international journal of energy science

Editorial Board

ISSN: 2218-6026 (Print) ISSN: 2304-3679 (Online) http://www.ijesci.org ﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡

Editor-in-Chief Prof. José Antonio Orosa Garcia, the University of A Coruña, Spain Editorial Board Dr. Yinjie Tang, Washington University, USA Dr. Gang Quan, Florida International University, USA Dr. Hansong Tang, The City University of New York, USA Dr. Linxia Gu, University of Nebraska, USA Dr. Shun-Chung Lee, National Cheng Kung University, Taiwan Dr. Qais H. Alsafasfeh, Tafila Technical University, Jordan Dr. Lukumon O. Oyedele, Queen's University Belfast, UK Dr. Berhan Ahmed, Melbourne University, Australia Dr. Ji-Hyoung Ryu, Chonbuk National University, Korea Dr. Alireza Bahadori, Curtin University, Australia

Prof. Jamal Mahmoud Nazzal, Jordan Cooperation Group, Jordan

Prof. Qifeng Zhang, University of Washington, USA

Prof. Yu Bo, China University of Petroleum, China

Dr. Ali Z. Hamadani, Isfahan University of Technology, Iran

Dr. Dragos Isvoranu, Polytechnic University of Bucharest, Romania

Dr. Chao Xu, Chinese Academy of Sciences, China

Prof. Pawan Tyagi, University of the District of Columbia, USA

Prof. Zvonimir Glasnovic, University of Zagreb, Croatia

Dr. Nakorn Tippayawong, Chiang Mai University, Thailand

Dr. Karmen Margeta, University of Zagreb, Croatia

Dr. Qingzhao Wang, University of Florida, USA

Dr. Wojciech M. Budzianowski,Wrocław University of Technology, Poland

Dr. Yongfu Huang, United Nations University World Institute, Finland

Dr. Zuhdi Hamdi Salhab, Palestine Polytechnic University, Palestine

Dr. Bindeshwar Singh, Kamla Nehru Institute of Technology, India

Dr. Messaouda Azzouzi, "Ziane Achour" University of Djelfa, Algeria

Dr. Vijay Kumar Thakur, Nanyang Technological University, Singapore

Dr. Sanjeev Kumar Aggarwal, M.M. Engineering College, India

Dr. Xiao-Sen Li, The Chinese Academy of Sciences, China

http://www.ijesci.org/

http://www.ijesci.org/file/21.pdf


http://www.ijesci.org/file/23.doc

http://www.ijesci.org/file/24.doc



﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡

TABLE OF CONTENTS

Volume 2, Issue 6 December 2012 Solar Radiation Forecast Using Artificial Neural Networks

Fernando Ramos Martins, En io Bueno Pereira, Ricardo André Guarn ieri………………………………………..217

Air Cells Using Negative Metal Electrodes Fabricated by Sintering Pastes with Base Metal Nanoparticles for Efficient Utilization of Solar Energy

Taku Saiki, Takehiro Okada, Kazuhiro Nakamura, Tatsuya Karita, Yusuke Nishikawa, Yukio Iida…228

Blends of Diesel – used Vegetable Oil in a Four-Stroke Diesel Engine

Charalampos Arapatsakos .……………………………………………………………………………………………………………….235

Catalytic Pyrolysis by Heat Transfer of Tube Furnace for Produce Bio-Oil

Kittiphop Promdee, Tharapong Vitidsant.............................………………………………………………………………..241

CMOS Bandgap Reference and Current Reference with Simplified Start-Up Circuit

Guo-Ming SUNG, Ying-Tzu LAI, Chien-Lin LU………..……………………………………………………….…………247

Transient Analysis of Three-Phase Self Excited Induction Generator Using New Approach

Vivek Pahwa, K. S. Sandhu………………………………………………..……………………………………………………………255

On the Sensitivity of Principal Components Analysis Applied in Wound Rotor Induction Machines Faults Detection and Localization

J. Ramahaleomiarantsoa, N. Heraud, E. J. R. Sambatra, J. M. Razafimahenina…………..……………………262

Evaluation of the Quality of Service Parameters for Routing Protocols in Ad-Hoc Networks

Zeyad Ghaleb Al-Mekhlafi, Rosilah Hassan, Zurina Mohd Hanapi………………………………………...……….272

An Investigation of Power Performance of Small Grid Connected Wind Turbines under Variable Electrical Loads

Md. Alimuzzaman, M.T.Iqbal, Gerald Giroux.………………………………………...……….…………………………….282

International Journal of Energy Science (IJES)

Journal Information IJES is published bimonthly by Science and Engineering Publishing Company. This journal is a multi-disciplinary focus on activities relating to the development, assessment and management of energy-related programs. As a professional journal, it aims at becoming a first-class international journal. It is hoped that this publication will prove to be an important factor in raising the standards of discussions, analyses, and evaluations relating to energy programs. The benefits to publish your papers in this journal include: ※RAPID PUBLICATION: Manuscripts are peer-reviewed and published within 50 days, and accepted papers are immediately published online. ※INDEXING SERVICES: All the published papers will be sent to be indexed by any third parties, including Google Scholar, Scirus, INSPEC, etc. ※REFERENCE SERVICES: All the published papers will be sent to the related researchers for potential reference by email.

C COPYRIGHT

Copyright ©2012 Science and Engineering Publishing Company (website: www.seipub.org). Address: 7800 STATE ROAD 46 E, PO Box 551, Riley, Indiana, 47871 All Rights Reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, scanning or otherwise, except as described below, without the permission in writing form of the Publisher. Copying of articles is not permitted except for personal and internal use, to the extent permitted by national copyright law, or under the terms of a license issued by the national Reproduction Rights Organization. Request for permission for other kinds of copying, such as copying for general distribution, for advertising or promotional purposes, for creating new collective works or for resale, and other enquiries should be addressed to the Publisher. Statements and opinions expressed in the articles and communications are those of the individual contributors and not the statements and opinion of Science and Engineering Publishing Company. We assumes no responsibility or liability for any damage or injury to persons or property arising out of the use of any materials, instructions, methods or ideas contained herein. We expressly disclaim any implied warranties of merchantability or fitness for a particular purpose. If expert assistance is required, the services of a competent professional person should be sought. Pricing: Personal subscription: $100

International Journal of Energy Science Vol. 2 Iss. 6, December 2012

217

Solar Radiation Forecast Using Artificial Neural Networks Fernando Ramos Martins, Enio Bueno Pereira, Ricardo André Guarnieri

Center for Earth System Science

Brazilian Institute for Space Research

São José dos Campos, Brazil. 12227-010

[email protected]

Abstract

The fast increase in importance of the solar energy resource as viable and promising source of renewable energy has boosted research in methods to evaluate the short-term forecasts of the solar energy resource. There is an increase on demand from the energy sector for accurate short-term forecasts of solar energy resources in order to support the planning and management of the electricity generation and distribution systems. The Eta model is the mesoscale model running at CPTEC/INPE for weather forecasts and climate studies. It provides outputs for solar radiation flux at the surface, but these solar radiation forecasts are greatly overestimated. In order to achieve more reliable information, Artificial Neural Networks (ANN) were used to refine short-term forecast for the downward solar radiation flux at the surface provided by Eta/CPTEC model. Ground measurements of downward solar radiation flux acquired in two SONDA sites located in Southern region of Brazil (Florianópolis and São Martinho da Serra) were used for ANN training and validation. The short-term forecasts produced by ANN have presented higher correlation coefficients and lower deviations. The ANN removed the bias observed in solar radiation forecasts provided by Eta/CPTEC model. The skill improvement in RMSE was higher than 30%when ANN was used to provide short-term forecasts of solar radiation at the surface in both measurement sites.

Keywords

Solar Energy Forecast; Short-Term Forecast; Artificial Neural Network; Energy Meteorology

Introduction

The scientific community points out that the fossil fuel expenditure is the major reason of the observed growth of the greenhouse gases concentrations in atmosphere along the last century [1]. Developed countries and advanced economies have been charged for the environmental damages due to consumption of conventional energy sources to meet their energy demand. However, emerging economies such as Brazil,

India, China, and Russia are increasingly sharing this responsibility as a result of their growing demand for energy to support their fast growing economic development

The commitment to reduce the emissions of carbon dioxide (and other greenhouse gases) established at the Kyoto Protocol and the perspectives of oil depletion in next decades are key factors to boost the research and development on alternatives and renewable energy sources such as solar and wind [2, 3].

Furthermore, the search for improvement on energy security has been driving the government policies and incentive programs to stimulate the employment of alternative renewable energy sources even in countries with large share of clean energy in their electricity generation matrix. For example, in Brazil, where hydroelectric energy is responsible for more than 70% of the electricity matrix, an energy shortage happened in 2001 due to very low precipitation during the wet season of the previous year [4]. After this event, Brazilian government created incentive programs for renewable energy sources like wind energy.

The solar energy is one of the promising alternatives in Brazil since most of its territory is located in the inter-tropical region where solar energy resources are accessible all year round [5]. The main obstacles to the commercial exploitation of solar energy resources are the highest cost compared to the conventional electricity generation technologies, lack of information on resource assessment and variability, and the deep dependency on the weather and climate conditions [4]. The investment costs are expected to fall during the next decades due to technological advances and market demands [6]. The growing market for solar energy leads to an increase on the demand for more reliable information concerning to solar resources, including its spatial and temporal variability in short and long terms.


218

In addition, the management of electricity generation and distribution systems is also asking for more accurate short-term solar energy forecasts.

Several methodologies were developed in order to provide solar radiation forecast in high temporal resolutions and short-term horizons [7, 8]. Some of them use numerical weather models (NWP). Such models have radiation parameterization codes to simulate the radiative atmospheric processes. Nevertheless, solar irradiation forecasts provided by NWP models for one or two days in advance have shown large deviations from solar irradiation data acquired at surface [9]. The major factors responsible for such deviations are related to the solar irradiation dependence on clouds and weather conditions which intrinsically involve non-linear physical processes [10].

Absorption and scattering interactions are the atmospheric radiative processes that attenuate the solar radiation flux. Therefore, the atmospheric optical properties should be known in order to correctly evaluate the solar irradiation at any specific site and time. Clouds are the main factor that modulates the solar radiation incidence at the surface [11, 12, 13, 14, 15]. Atmospheric aerosols also have an important role in atmospheric radiative processes, mainly in some regions where anthropogenic emissions from biomass or fossil fuel burning takes place.

The Eta/CPTEC mesoscale model runs operationally in the Center of Weather Forecast and Climate Studies at Brazilian Institute for Space Research (CPTEC/INPE) and provides short-term forecasts for many meteorological variables, including surface solar irradiation. However, the references [11] and [12] showed that Eta/CPTEC model systematically overestimates the surface solar irradiation, as well as the sensible and latent heat fluxes at surface. A common issue in numerical atmospheric radiation codes is the excess of the incoming shortwave radiation at the surface as a result of the deficient parameterization of extinction interactions with water vapor, atmospheric aerosols and clouds. Several methodologies were published in order to improve solar forecasts provided by numerical weather models [9, 16, 17, 18].

This work aims to present a methodology to reduce deviations of solar irradiation forecasts provided by Eta/CPTEC model by using a statistical post-processing applied to the model outputs. This paper presents the results obtained when Artificial Neural Networks

(ANN’s) were used as statistical tool to refine the solar radiation forecast provided by Eta/CPTEC model.

Artificial neural networks (ANN) are data-driven instead of model-driven techniques once the results provided by them depend on the available data used to feed the ANN. Relationships between predictors (input data) and predictions are developed after building a system which simulates the physical processes in atmosphere. Artificial neural networks have been applied in renewable energy research for modeling and design solar systems and to provide short-term forecasts for energy resources [19]. Reference [20] indicated that the ANN systems are able to predict the solar radiation time series more effectively than the conventional procedures based on the clearness index. The authors observed that the forecasting ability can be further enhanced with the use of additional meteorological parameters like temperature and wind direction. References [21] and [22] discussed different methodologies using ANN to provide short-term forecasts for solar radiation by extracting knowledge from a long ground data series. Reference [23] compared some statistical models and ANN systems using meteorological data as input data. The authors concluded that ANN systems were a promising alternative to the traditional approaches for estimating global solar radiation, especially in cases where solar radiation measurements are not readily available.

This paper presents an attempt to get better predictability for the solar energy resources using operational Eta/CPTEC model and it constitutes an important application of the meteorology science to the energy planning and decision-making processes in energy sector. The target is to provide more precise and reliable information on future availability of solar resources in order to optimize electricity generation and distribution systems.

Methodology

Forecasting solar irradiation depends on prospecting the future atmospheric conditions. Despite the intrinsic uncertainties, NWP models provide information about many meteorological variables, including solar radiation data and atmospheric optical properties for several future timeframes. However, earlier studies demonstrated that solar radiation data provided by such models presents a large bias making its use inappropriate to electricity system management where several solar power plants are connected [10, 16, 17, 18].


219

This work employed the weather forecast outputs provided by the Eta/CPTEC model together with environmental data to feed Artificial Neural Network (ANN). The main goal was to achieve a short-term forecast for solar irradiation with lower deviations than the ones provided by the Eta/CPTEC model. The solar radiation data acquired in two SONDA ground sites located in the Southern region of Brazil was used as reference for training and performance evaluation of the ANN.

Model Eta/CPTEC

The Eta/CPTEC model is used for operational weather forecasting, climate investigation, regional climate change studies and research on several issues like pollutant transport [24]. The Eta model, which has been running at CPTEC since 1996, was set up and optimized to the South America atmospheric conditions. The Eta/CPTEC model runs routinely for South America continent and neighboring oceans: latitudes from 50.2ºS to 12.2ºN, and longitudes from 83ºW to 25.8ºW. The horizontal resolution equals to 40km and 38 vertical layers were used for this study.

The Eta/CPTEC model employs the “finite difference” scheme to solve the equations system that describes the physical processes in atmosphere. The model uses the vertical coordinate “Eta”, η, defined as:

(1)

where pt is the pressure at the top of the model atmosphere, pref is the reference pressure to the vertical profile, and psfc and zsfc are the pressure and height of the lower boundary surface, respectively. The Eta coordinate was adopted to reduce the large errors observed in several numeric weather forecast models that use the sigma surfaces [12]. These deviations are related to the determination of the horizontal pressure gradient force, as well as the advection and the horizontal diffusion on a steeply sloped coordinate surface [25, 26].

The discretization of the space domain uses the Semi-Staggered Arakawa E-grid on the horizontal and the Lorenz grid on the vertical. The radiation modeling uses the schemes described in [27] for shortwave radiation, and in [28] for long wave radiation. More detailed descriptions about the physical parameterizations adopted in Eta/CPTEC model can be found in [26, 29, 30, 31].

The Eta/CPTEC model was executed using initial conditions at 00UT provided by NCEP analyses. The CPTEC Atmospheric Global Circulation Model (AGCM) provided the lateral boundary conditions.

The outputs provided by Eta/CPTEC model for 2001 till 2005 were used. The output file contains forecasts for 58 atmospheric variables at the synoptic timeframes (6, 12, 18 and 24UT) for 7 days in advance. The model provided the total amount in atmospheric column for forty-nine variables, and vertical profile values at 19 atmospheric pressure levels for the remaining nine variables. Only 33 out of the 58 atmospheric variables were used in this study. All vertical profile data were discarded together with 16 variables not representative of the atmospheric condition like topography, soil temperature and humidity for levels under surface.

Table I presents a complete list of model output data used for this work with a short description of them. Instantaneous values at each synoptic time were recorded for most of the data. However, average values regarding to the 6-hour period before each synoptic time were stored for some of the meteorological output variables, such as “ocis”.

SONDA network

SONDA (Brazilian System for Environmental Data applied to the Energy Sector) is a network of ground measurement sites, operated and managed by INPE. The goal is to acquire reliable surface solar irradiation and wind data at different climate areas in Brazil in order to develop, improve and validate numerical models used for renewable energy resources assessment and environmental research. The SONDA database will provide valuable information applied to the research on the energy meteorology in Brazil.

In this work, the SONDA ground data acquired at two SONDA sites was used for the ANN training and configuration as described later in this paper. Besides that, ground data were used to evaluate the deviations presented by short-term forecast provided by both methodologies: Eta/CPTEC model and ANN. Both measurement sites were located in the Brazilian Southern region:

São Martinho da Serra (SMS) – 29.44ºS/53.82ºW.

Florianópolis (FLN) – 27.60ºS/48.52ºW;

Fig. 1 shows the location of measurement sites of SONDA network featuring SMS and FLN sites. These both sites were chosen in order to evaluate the performance of ANN and Eta/CPTEC model in two


220

different climate conditions. SMS is located in the continental area at 500m above the sea level. FLN is located at the coastal area of Brazilian Southern region presenting the largest total precipitation along the year in Brazilian territory. The SMS has been collecting data since June 2004 and FLN has been acquiring data since 1995.The other SONDA sites are more recent and have smaller databases. The SONDA website (http://sonda.ccst.inpe.br) presents all information about measurement sites and describes the data quality assurance program.

For this work, data acquired from January/2001 to October/2005 in FLN and from July/2004 to October/2005 in SMS were used. The Kipp&Zonen CM-21 pyranometers [32] were used to acquire global solar irradiation data. One-minute average solar irradiation data was stored and its quality was checked. Both sites take part in Baseline Solar Radiation Network (BSRN) and meet all the quality criteria established by World Meteorological Organization (WMO).

FIGURE 1 LOCATION OF GROUND SITES OF SONDA NETWORK. FLORIANÓPOLIS AND SÃO MARTINHO DA SERRA WERE USED

FOREVALUATION OFSHORT-TERM FORECASTS.

After data-quality verification, 1150 days for FLN and 472 days for SMS were available for this work. The ground database was divided into 3groups as follows:

Training group: with 575 days for FLN and 236 days for SMS;

Validation group: with 288 days for FLN and 118 days for SMS;

Investigation group: with 287 days for FLN and 118 days for SMS.

The training group was used for the ANN training. The validation group was employed to evaluate and establish the end of the training step. The investigation group was used to evaluate the reliability of ANN outputs. More details on each these three steps are described latter in this paper.

Data Management

As explained earlier, the solar and meteorological database used to feed ANN comprises the output data provided by the model Eta/CPTEC (Table I). In addition, other three variables were calculated in order to supply ancillary information for the ANN: solar radiation flux at TOA (STOA), mean air mass (airm), and mean solar zenith angle (szam). Altogether, 36 variables were used as ANN predictors.

As described on Table I, the solar irradiation data provided by the Eta/CPTEC model, “ocis”, represents the 6-hour average solar irradiation. In order to achieve the same time-scale, the solar irradiation data acquired in FLN and SMS sites were averaged over the same 6-hour intervals. In summary, ground and model data of solar irradiation represents the total energy in the 6-hour period and they are expressed in MJ.m-2 (mega joules per squared meter).

The 6-hour average solar radiation flux at the top of the Earth’s atmosphere (STOA) was calculated taking into consideration local latitude, solar zenith angle, eccentricity and solar declination [13, 14]. As the ground solar irradiation data and “ocis”, the STOA solar radiation flux was also expressed in MJ.m-2.

Relative humidity, atmospheric pressure, air temperature, wind velocities and all other instantaneous data, provided by Eta/CPTEC model for synoptic time (Table I),were averaged by taking the two consecutive values. The averages were assigned to the second synoptic time in order to set up the database in a similar way used for ground data. This procedure aims to better represent the atmospheric and meteorological variability in the 6-hour interval.

In addition, the solar zenith angle (szam) and the air mass (airm) were obtained and stored for the same 6-hour intervals. Thus, the “ocis” data and all 36 variables used to feed ANN have the same temporal resolution and represent the equivalent timeframes.

The 36 predictors and the ground data are disposed into four timeframes each day: 6:00, 12:00, 18:00 and 24:00UT. Each timeframe represents the corresponding


221

time interval: 0-6UT (Rad06UT), 6-12UT (Rad12UT), 12-18UT (Rad18UT), and 18-24UT (Rad24UT). This paper only presents results for the Rad18UT timeframe. The

Rad18UT was chosen because the highest fraction (63% – 80%) of solar radiation flux occurs during the 12-18UT intervals throughout the year at both ground sites [35].

TABLE 1 THE METEOROLOGIC DATAUSED AS PREDICTORS IN ANN. ALL DATA WAS PROVIDED BY MODELETA/CPTEC

VARIABLE DESCRIPTION (UNITS) KEY FEATURES

rh2m Relative humidity at 2m-height (0 to 1 – adimensional) Instantaneous values

pslc Pressure at surface (hPa) Instantaneous values

tp2m Temperature at 2m-height above the surface (K) Instantaneous values

dp2m Dew Point Temperature at 2m above the surface (K) Instantaneous values

u10m Zonal wind at 10m-height above the surface (m s-1) Instantaneous values

v10m Meridional wind at 10m-height above the surface (m s-1) Instantaneous values

wnds Wind velocity at 10m-height above the surface (m s-1) Instantaneous values

prec Total rainfall (kg m-2 dia-1) Total in the 6h period

prcv Convective rainfall (kg m-2 dia-1) Total in the 6h period

prge Large scale rainfall (kg m-2 dia-1) Total in the 6h period

clsf Latent Heat Flux at the surface (MJ m-2) Average value in the 6h period

cssf Sensible Heat Flux at the surface (MJ m-2) Average value in the 6h period

ghfl Heat Flux in the soil (W m-2) Average value in the 6h period

tsfc Surface Temperature (K) Instantaneous values

qsfc Specific humidity at the surface (kg(H2O) kg(air) -1) Instantaneous values

lwnv Cloud cover Index for low clouds (0 a 1 - adimensional) Instantaneous values

mdnv Cloud cover Index for average clouds (0 a 1 - adimensional) Instantaneous values

hinv Cloud cover Index for high clouds (0 a 1 - adimensional) Instantaneous values

cbnt Mean Cloud cover Index (0 a 1 - adimensional) Instantaneous values

ocis Downward shortwave radiation flux at the surface (MJ m-2) Average value in the 6h period

olis Downward longwave radiation flux at the surface (MJ m-2) Average value in the 6h period

oces Upward shortwave radiation flux at the surface (MJ m-2) Average value in the 6h period

oles Upward longwave radiation flux at the surface (MJ m-2) Average value in the 6h period

roce Upward shortwave radiation flux at the TOA (MJ m-2) Average value in the 6h period

role Upward longwave radiation flux at the TOA (MJ m-2) Average value in the 6h period

albe Albedo (0 a 1 - adimensional) Instantaneous values

cape Available potential convective energy (m2 s-2) Instantaneous values

cine Energy to avoid convection (m2 s-2) Instantaneous values

agpl Instantaneous precipitable water amount (kg m-2) Instantaneous values

pcbs Pressure at the bottom of the clouds (hPa) Instantaneous values

pctp Pressure at the top of the clouds (hPa) Instantaneous values

tgsc Soil temperature at the surface layer (K) Instantaneous values

ussl Soil humidity at the surface (0 a 1 - adimensional) Instantaneous values


222

Artificial Neural Networks (ANNs)

Artificial Neural Networks (ANN) is computing systems, which attempt to simulate the structure and function of biological neurons. Generally, the ANN consists of a number of interconnected processing elements, called neurons. Fig. 2 presents an artificial neuron. The ANN usually consists of an input layer, some hidden layers and an output layer. Signals flow from the input layer through to the output layer via unidirectional connections (synapses). Synapses connect neurons of neighboring layers. The input data (xi) is weighted by values associated with each synapse (wij), called synaptic weights. Knowledge is usually stored as a set of connection weights (presumably corresponding to synapse efficacy in biological neural systems). The activity level of a neuron (υj) is determined by summing up all its weighted values together with its bias (bj). The neuron output is a result from an activation function (φ(υj)). Generally, the activation function is a linear or hyperbolic-tangent function. The non-linear activation functions allow ANNs to simulate non-linearity behaviors and complex patterns [19].

The ANN architecture depends on the physical process, the training method and the kind of data that the neural network will simulate. The multi-layer perceptron (or feed forward ANN) is the most widely ANN architecture used in meteorological topics [23]. A schematic diagram of typical multilayer neural network architecture is shown in Fig. 3. The input layer consists on one neuron for each input data (called predictor), and the output layer consists of one neuron for each simulated data (called predictant). The number of hidden layers and their total amount of neurons are not a priori established. There is no standard procedure to identify the best combination of neurons and layers.

The most widespread training algorithm used for multilayer perceptrons is the back propagation algorithm [33]. In this work, we use a modified version of back propagation, called Resilient Back propagation or Rprop [34]. The validation dataset was employed to verify the performance of the ANN with an independent data sample – data not used in training process. This procedure allowed to check the generalization capacity achieved by the ANN along the training and to find out the appropriate moment to stop the training step in order to avoid overlearning. After

training, the weights and bias are fixed and the ANN is ready to be used in simulations.

For this study, preliminary experiments revealed that better ANN performances were achieved using two hidden layers of neurons. These experiments were developed in two different situations. First, the 36 variables described earlier were used as input to the ANN; and, in the second situation, only a set of 8 out of the 36 input variables were used. Table II shows the best neurons distributions verified for each ANN-model. On both cases, only one neuron is the output layer to provide information on solar radiation flux at surface. The number of neurons in the input layer is equal to the number of predictors used to feed ANN.

The investigation dataset was used to evaluate the performance of ANN to provide reliable solar irradiation forecast. The next topic discusses the statistical parameters used to evaluate deviations of the ANN and Eta/CPTEC outputs and the skills of each model to provide reliable forecasts.

TABLE 2NUMBER OF ARTIFICIAL NEURONS IN EACH ANN LAYER

ANN-36p ANN-8p

Input layer 36 8

First hidden layer 36 16

Second hidden layer 18 8

Output layer 1 1

ANN-36p – ANN using 36 variables as predictors

ANN-8p – ANN using 8 variables as predictors

FIGURE 2 SYMBOLIC REPRESENTATION OF AN ARTIFICIAL NEURON AND ITS PARAMETERS.


223

FIGURE 3SCHEMATIC DIAGRAM OF A FEEDFORWARD ANN USED IN THIS STUDY.

Statistical analysis of ANN and Eta/CPTEC outputs

The outputs (forecasts – F) were compared with measured values (observations – O), and deviations between them (F - O) were calculated. The performance of the Eta/CPTEC and ANN models was checked with two statistical indices: mean error (ME) or bias, and root mean squared error (RMSE). ME values provide information about the systematic deviations of the forecasts indicating if the models overestimateor underestimate the actual solar irradiation at the two measurement sites. RMSE is a measure of how effectively the models predict ground observations. Since the deviations are squared, large deviations have greater contribution to RMSE. For this study, both ME and RMSE indices were normalized and expressed as percentage of the average solar irradiation in the two measurement sites, as shown in eq. (2) and (3).

(2)

(3)

where N is the number of data pairs (forecast and observation) used in the evaluation – 287for FLN and 118 for SMS.

In addition, the Pearson’s correlation coefficient (R) was computed as described in eq. (4):

(4)

In order to compare the performance of ANN and Eta/CPTEC model, the skill-score index was used as defined in eq. (5):

(5)

where Score can be the ME% or the RMSE% values obtained for a particular model (Eta/CPTEC or ANN) in evaluation, Scoreref is the score calculated for a reference method and Scoreper f is the score value expected for perfect-forecast.

Results and Discussion

Initially, the Eta/CPTEC forecast and ground data for solar radiation flux were compared. As demonstrated in previous studies [10, 11], a significant positive bias (overestimation) was observed in the solar radiation flux provided by Eta/CPTEC model. Table III shows the performance scores obtained for Eta/CPTEC estimates using only the investigation dataset (N = 287 for FLN; N = 118 for SMS). Similar scores were obtained when complete dataset was used for comparison between model estimates and ground data. Based on these results, it was assumed that the investigation dataset are representative of the complete dataset. Since ANN performance must be evaluated using the investigation dataset, only the Eta/CPTEC performance scores using this dataset were considered from this point on.

TABLE 3PERFORMANCE SCORESOBTAINED BY MODEL ETA/CPTEC

Scores Florianópolis São Martinho da Serra

N =1150 N =287* N =472 N =118*

R 0.747 0.720 0.790 0.775

R2 0.558 0.519 0.624 0.600

ME% 24.7% 24.6% 27.8% 28.0%

RMSE% 39.7% 40.0% 41.9% 43.2%

* - results obtained using only the investigation dataset.

As previously mentioned, various statistical analysis and simulations were performed using different subsets of the predictors listed in Table I in order to find a reduced dataset of predictors which produces a performance similar to that obtained when all 36 predictors are used. These analysis point out a set of 8

%)(

)(100%

1

1

∑

∑

=

=

−⋅= N

ii

N

iii

O

OFME

%)(1

)(1

100%

1

1

2

∑

∑

=

=

−⋅= N

ii

N

iii

ON

OFNRMSE

∑∑

∑

==

=

−⋅−

−−=

N

ii

N

ii

N

iii

OOFF

OOFFR

1

2

1

2

1

)()(

))((

refperf

ref

ScoreScoreScoreScore

refScoreSkill−

−=),(


224

predictors: solar radiation flux at TOA (STOA), relative humidity (rh2m), surface temperature (tsfc), precipitable water amount (agpl), zonal wind speed at 10 m height (u10m), and predictors for cloud fractions (cbnt, hinv and mdnv). Hereafter, the ANNs using 36 and 8 predictors will be called ANN-36p and ANN-8p, respectively.

Table IV presents the performance scores obtained for ANN-36p and ANN-8p using the investigation dataset for both ground sites. As noticed, there is a very similar performance in terms of correlation (R) and RMSE deviations. However, the ANN-8p provided solar irradiation forecasts for both sites with 50% less ME than the ANN-36p.

As noticed by comparing Tables III and IV, the ANN-36p and ANN-8p provided solar irradiation forecasts presenting larger correlation with ground observations in both sites. The ANN-8p outputs presented the lowest systematic deviation while Eta/CPTEC forecasts showed the largest deviations (ME and RMSE) for both ground sites.

Fig. 4 and 5 present four scatter-plots comparing forecast values and observations. Besides the scatter-plots for Eta model, ANN-36p and ANN-8p, it is also showed a plot for a forecast method called persistence. The persistence forecast is the simplest method to predict meteorological data and it consists in taking the value observed in a previous day as the forecast for the current day. Any forecast method is useful if it can lead to better results than the persistence forecast.

According to Fig. 4 and 5, the solar radiation flux outputs provided by Eta/CPTEC model are better than persistence forecasts, in general. However, it can be observed the positive bias mentioned before. The Eta/CPTEC model overestimated the observations, especially for cloudy days when solar radiation flux at the surface is lower. TABLE 4 PERFORMANCE SCORESOBTAINED BY ANN-36P AND

ANN-8P


ANN-36p ANN-8p ANN-36p ANN-8p

R 0.804 0.790 0.839 0.848

R2 0.646 0.625 0.704 0.720

ME% -2.1% -0.8% -1.7% -0.7%

RMSE% 26.2% 26.9% 28.8% 27.6%

All results obtained using investigation dataset.

Meanwhile, the scatter-plots for ANNs showed better agreement between forecasts and observations – most of the data points are located near the perfect-forecast line (diagonal line). Small difference was observed when ANN-8p is used instead ANN-36p, indicating that the 8 selected predictors was able to provide solar irradiation forecast as reliable as the forecast obtained by using the 36 predictors.

TABLE 5SKILL-SCORE CALCULATED WITH RMSE% VALUES FOR ANN TAKING MODEL ETA/CPTEC AND PERSISTENCE AS

REFERENCE METHODS


ANN-36p

ANN-8p ANN-36p ANN-8p

Skill(RMSE%, persistence)

0.429 0.414 0.464 0.487

Skill(RMSE%, Eta) 0.344 0.328 0.333 0.361

* - results obtained using investigation dataset.

FIGURE 4 SCATTER-PLOTS OF FORECASTS VERSUS GROUND

DATAFOR FLN: (A) PERSISTENCE METHOD, (B) MODEL ETA/CPTEC, (C) ANN-36P, AND (D) ANN-8P

Fig. 6 shows a short temporal series taken from the investigation dataset prepared for FLN and SMS sites. Outputs from Eta/CPTEC model and ANN were put together with observations acquired in Winter/2005 and Summer/2004-2005. Fig. 6 demonstrates the best agreement between the ANN forecasts and ground data. The deviations for each day are presented in Fig. 7. It is clear that an important improvement in short-term forecast for solar radiation flux is achieved when ANN is used to refine solar irradiation outputs provided by


225

model Eta/CPTEC. However, no significant differences were observed between ANN-36p and ANN-8p. Again, the analysis of Fig. 7 demonstrates that the eight selected predictors provide enough information to ANN simulate the atmospheric processes with good performance. To quantify the improvement acquired by the use of ANNs, the skill-score values were calculated using RMSE% score, and the results are presented in Table V. In general, the ANNs lead to skill-scores in RMSE% 30% higher if compared to model Eta/CPTEC.

FIGURE 5SCATTER-PLOTS OF FORECASTS VERSUS GROUND

DATAFORSMS: (A) PERSISTENCE METHOD, (B) MODEL ETA/CPTEC, (C) ANN-36P, AND (D) ANN-8P

Conclusions

Currently, the renewable sources of energy are getting more importance into electricity generation systems. Therefore, there is an increasing demand from the energy sector for accurate forecasts of solar energy resources in order to support and manage electricity generation and distribution systems. The forecasts provided by numerical weather models could supply this demand but, in general, these forecasts present large deviations reducing their confidence and reliability. In Brazil, the Eta/CPTEC model provided solar irradiation forecasts with bias around 25%. Lower deviations were observed when ANN was used to refine the forecasts provided by the Eta/CPTEC model. The comparison between solar irradiation forecasts and ground data showed a bias reduction from 25%for Eta/CPTEC forecasts till -1% for the ANN outputs. Both ANNs, ANN-36predictors and ANN-8predictors, have presented very similar performances. The skill-score

indices showed that both ANNs have improved the confidence and reliability on the solar radiation forecasts in more than 30% for both sites: Florianópolis in coastal area and São Martinho da Serra in continental region. The improvements in predictability were also observed as indicated by the correlation coefficients: from 0.72 to 0.80 in FLN, and from 0.78 to 0.85 in SMS.

FIGURE 6 SHORT TIME SERIES COMPARING FORECASTS AND GROUND DATA FOR SOLAR RADIATION FLUX AT SURFACE IN

FLN AND SMS

FIGURE 7 DEVIATIONS BETWEENFORECASTS AND GROUND DATA FOR SOLAR RADIATION FLUX AT SURFACE IN FLN AND SMS. THE MODEL ETA/CPTEC PROVIDED ESTIMATES

WITH LARGER DEVIATIONS.

ACKNOWLEDGMENTS

The authors would like to thank FINEP (project22.01.0569.00) and PETROBRAS/CENPES (project 0050.0019.104.06.2) for their finance support to


226

the SONDA project. Thanks are also due to the following colleagues: Silvia V. Pereira, Sheila A. B. Silva, Rafael Chagas and to the technologists of Laboratório de Instrumentação Meteorológica (LIM/CPTEC). In addition, acknowledgments are due to CPTEC/INPE and CNPq (grants 151700/2005-2, 141844/2006-0, 132148/2004-8, 555764/2010-9).

REFERENCES

[1] IPCC, 2007. In: Solomon, S., Quin, D., Manning, M.,

Chen, Z., Marquis, M., Averyt,K.B., Tignor, M., Miller,

H.L. (Eds.), IPCC Climate Change 2007: The

physicalscience basis. Contribution of Working Group I

to the Fourth AssessmentReport of the

Intergovernmental Panel on Climate Change.

CambridgeUniversity Press, Cambridge, UK and New

York, p. 996 Online: <htpp://www.ipcc.ch>.

[2] Bentley, R.W., 2002: Global Oil and Gas Depletion: An

Overview. Energy Policy., 30, 189-205.

[3] Geller, H. S., 2003: Revolução Energética: políticas para

um futuro sustentável. Relume Dumará, 299pp.

[4] Martins and Pereira, 2011.Enhancing information for

solar and wind energy technology deploymentin Brazil.

Energy Policy., 39, 4378-4390.

[5] Pereira, E. B., Martins, F. R., Abreu, S. L. and Ruther R.,

2006. Brazilian Atlas for Solar Energy. Brazilian Institute

for Space Research, 70pp.

[6] Viana et al, 2011. Assessing the potential of

concentrating solar photovoltaic generationin Brazil

with satellite-derived direct normal irradiation. Solar

Energy, 85, 486-495.

[7] Reikard, G., 2008: Predicting solar radiationat high

resolutions: a comparison of time series forecasts. Solar

Energy, 83, 342-349.

[8] Hammer, A.; Heinemann, D.; Hoyer, C. and Lorenz, E.,

2001: Satellite Based Short-Term Forecasting of Solar

Irradiance - Comparison of Methods and Error Analysis,

Proceedings of the EUMETSAT Meteorological Satellite

Data Users' Conference, Antalya, Turkey, 01.-05.10.2001.

[9] Heinemann, D.;Lorenz, E.; Girodo, M., 2004: Forecasting

of Solar Radiation. Proc. International Workshop on

Solar Resource from the Local Level to Global Scale in

Support to the Resource Management of Renewable

Electricity Generation. Institute for Environment and

Sustainability, Joint Research Center, Ispra, Italy.

[10] Hinkelman, L. M., Ackerman, T. P., and Marchand, R.

T., 1999: An evaluation of NCEP Eta model predictions

of surface energy budget and cloud properties by

comparison with measured ARM data. J. Geophys. Res.,

104(D16), 19535-19549.

[11] Chou, S. C., Tanajura C. A. S., Xue, Y., and Nobre, C. A.,

2002: Validation of the coupled Eta/SsiB model over

South América. J. Geophys. Res., 107(D20), LBA 56, 1-20.

[12] Mesinger, F., 2005: The Eta: brief history, summary,

results, discussion. Proceedings of ICTP Regional

Weather Predictabilityand Modeling Workshop. Centre

for Weather Forecast and Climate Studies – Brazilian

Institute for Space Research (CPTEC/INPE), São José

dos Campos, Brazil. Available online at

<http://etamodel.cptec.inpe.br/refer.shtml>.

[13] Kidder, S. Q., and Vonder Haar, T. H., 1995: Satellite

Meteorology: an introduction. Academic Press, 466pp.

[14] Iqbal, M., 1983: An Introduction to Solar Radiation.

Academic Press Canada, 390pp.

[15] Liou, K. N., 1980: An Introduction to Atmospheric

Radiation. Academic Press Inc., 392pp.

[16] Armstrong, M. A., 2000: Comparison of MM5 Forecast

Shortwave Radiation with Data Obtained from the

Atmospheric Radiation Measurement Program. Master

of Science Scholarly Paper, University of Maryland, USA.

[17] Zamora, R. J.; Solomon, S.; Dutton, E. G.; Bao, J. W.;

Trainer, M.; Portmann, R. W.; White , A. B.; Nelson, D.

W.; Mc Nider, R.T., 2003: Comparing MM5 Radiative

Fluxes with Observations Gathered During the 1995

and 1999 Nashville Southern Oxidants Studies. Journal

of Geophysical Research,108(D2), 4050-4060.

[18] Barbosa, H. M. J.; Tarasova, T. A.; Cavalcanti, I. F. A.,

2008: Impacts of a New Solar Radiation

Parameterization on the CPTEC AGCMClimatological

Features. Journal of Applied Meteorology and Climatology,

47(5), 1377-1392.

[19] Kalogirou, S. A., 2001: Artificial neural networks in

renewable energy systems applications: a review.

Renewable and Sustainable Energy Reviews, 5, 373-401.

[20] Sfetsos, A.; Coonick, A. H., 2000. Univariate and

multivariate forecasting of hourly solarradiation with

http://etamodel.cptec.inpe.br/refer.shtml


227

artificial intelligence techniques. Solar Energy, 68(2), 169-

178.

[21] Mihalakakou, G.; Santamouris, M.; Asimakopoulos, D.

N., 2000. The total solar radiation time series simulation

in Athens,using neural networks. Theoretical and Applied

Climatology, 66, 185-197.

[22] Hocaoglu, F. O.; Gerek, O. N.; Kurban, M., 2008: Hourly

solar radiation forecasting using optimal coefficient 2-

Dlinear filters and feed-forward neural networks. Solar

Energy, 82, 714-726.

[23] Tymvios, F.S.; Jacovides, C.P.; Michaelides,S.C.; Scouteli,

C., 2005: Comparative study of Angström’s and

artificial neuralnetworks methodologies in estimating

global solar radiation. Solar Energy, 78, 752-762.

[24] Chou, S. C.; Bustamante, J. F.; Gomes, J. L., 2005:

Evaluation of seasonal precipitation forecasts over

South America using Eta model. Nonlin. Proc. Geophys.,

12, 537-555.

[25] Staudenmaier Jr., M., 1996: A description of the Meso

Eta model. Western Region Technical Attachment no. 96-06,

5pp. Available online at

<http://www.wrh.noaa.gov/wrh/96TAs/TA9606/ta96-

06.html>.

[26] Ničković, S., Mihailović, D., and Rajković, B., 1988:

Scientifc documentation of the Eta model. Word

Meteorological Organization.

[27] Lacis, A. A., and Hansen, J. E., 1974: A parameterization

of the absorption of solar radiation in earth's

atmosphere. J. Atmosph. Sci., 31, 118-133.

[28] Fels, S. B., and Schwarzkopf, M. D., 1975: The simplified

exchange approximation: a new method for radiative

transfer calculations. J. Atmosph. Sci., 32(7), 1475-1488.

[29] Black, T. L., 1994: The new NMC mesoscale Eta model:

Description and forecast examples. Weather and

Forecasting , 9(2), 265-278.

[30] Janjić, Z. L., 1994: The step-mountain Eta coordinate

model: Further developments of the convection, viscous

sublayer, and turbulence closure schemes. Monthly

Weather Review, 122(5), 927-945.

[31] Mesinger, F.; Janjić, Z. L.; Ničković, S., and Deaven, D.

G., 1988: The step-mountain coordinate: Model

description and performance for cases of Alpine lee

ciclogenesis and for a case of Appalachian

redevelopment. Monthly Weather Review,116, 1493-1518.

[32] Kipp & Zonen, 2006: Kipp & Zonen Co. Webpage:

manufacturer of meteorological and industrial sensors.

Available online at <http://www.kippzonen.com/>.

[33] Rumelhart, D. E., Hinton G. E., and Williams, R. J., 1986:

Learning representations by back-propagation errors.

Nature, 323, 533-536.

[34] Riedmiller, M., and Braun H., 1993: A direct adaptive

method for faster backpropagation learning: The

RPROP algorithm. Proceedings, IEEE International

Conference on Neural Networks.

[35] Guarnieri, R. A., 2006: Emprego de redes Neurais

Artificiais e Regressão Linear Múltipla no refinamento

das previsões de radiação solar do modelo Eta.Insituto

Nacional de Pesquisas Espaciais, 171pp.

http://www.wrh.noaa.gov/wrh/96TAs/TA9606/ta96-06.html

http://www.wrh.noaa.gov/wrh/96TAs/TA9606/ta96-06.html


228

Air Cells Using Negative Metal Electrodes Fabricated by Sintering Pastes with Base Metal Nanoparticles for Efficient Utilization of Solar Energy Taku Saiki, Takehiro Okada, Kazuhiro Nakamura, Tatsuya Karita, Yusuke Nishikawa, Yukio Iida

Department of Electrical and Electronic Engineering,

Faculty of Engineering Science, Kansai University

3-3-35 Yamate, Suita, Osaka. 564-8680, Japan

[email protected] Abstract

Research on the produce of renewable energy as a source of solar power has continuously advanced. We have proposed an energy cycle that uses solar-pumped pulse lasers and base metal nanoparticles. Here, Fe and Al nanoparticles were prepared by laser ablation in liquid for the energy cycle . Solar power was confined in base metal nanoparticles. Metal plates were fabricated by sintering metal paste with base metal nanoparticles. Electricity was generated by air cells using the sintered metal plate . A highly repetitive laser pulse, which was an alternative to lasers driven by solar power, was used for laser ablation in liquid, and metal oxides (Fe3O4 or Al2O3) were reduced and metal nanoparticles were fabricated. Metal plates with a low electrical resistance were fabricated by sintering them at a low temperature of 520 K.

The electrical properties of the air cells fabricated using sintered paste with nanoparticles as negative electrical cathodes were the same as those of the air cells fabricated in a blast furnace. It was found that the sintered metal nanopaste could be used for air cells.

Keywords

Solar Power; Laser; Renewable Energy; Air Cells; Nanoparticles

Introduction

Recently, concerning the desire to develop methods for the reduction in the amounts of gases that causes the global earth warming, low-cost and energy-saving method are encouraged. Also, research on the production of renewable energies has continuously advanced [1-3].

High temperatures are generated by focusing solar light on metal oxides, thereby reducing the metal oxides by the generated high temperatures. As a result

of this process, chemical energy is stored as the difference of chemical potential between metal and metal oxide, and hydrogen is generated [1]. In previous research, CW lasers have been generated using solar energy is used for the reduction of magnesium oxides. Here, an energy cycle in which reduced magnesium is used as a renewable energy is proposed [3, 4]. Our group is also developing solar-pumped lasers [3, 5-7] that employ solar light for pumping laser materials. We propose an energy cycle using base metallic nanoparticles and solar-pumped pulsed lasers.

A Nd/Cr:YAG ceramic laser has been used as a solar-pumped laser [6, 7]. Its lasing wavelength is 1064 nm, which is the same as that of the Nd:YAG laser. The ceramic laser has a special lasing mechanism because its photon energy includes thermal energy due to the phonon-assisted cross-relaxation effect [8]. The maximum theoretical optical-optical (from solar light to laser) conversion efficiency reaches close to 80%. Very high opt.-opt. conversion efficiency close to 60% has been achieved in experiments. The generation of efficient high peak-power and highly repetitive laser pulses from solar-pumped lasers, whose conversion efficiency is close to that of CW laser [6], has been realized.

The reduction and production of metal nanoparticles using laser ablation in liquid and solar-pumped pulsed lasers has been performed here. The method is different from that adopted by Yabe et. al [4].

Laser ablation in liquid with a high fluence and a high intensity of laser pulses can produce nanoparticles and


229

reduce metal oxides [9-13]. The physics of thermal ablation and coulomb explosion [14-16] have been proposed for the ablation of metal oxides in liquid, which is remarkably different from that using a CW laser.

It has been considered that reduced metal oxides can be used as a negative electrode of air cells. Electricity can be obtained by reacting metals with oxygen in air. Research on air cells has been carried out for 40 years since Ferro developed Zn air cells [17, 18]. Air cells are expected as a future electrical power source because their energy storage density per unit weight is higher than that of the Li ion battery. Metal oxides, such as Fe3O4 and Al2O3, exist in large quantities in the ground.

Our aim in this paper is to develop primary air cells [17, 18]. The electrical property of air cells using sintered metal paste was investigated and compared with that of air cells using conventional metals. No such researches have been performed previously. The possibility of using sintered paste with reduced metal oxide particles fabricated by laser ablation in liquid as a negative electrode was investigated.

Experimental

Reduction of metal oxide powder and production of nanoparticles

By using pulse laser ablation in liquid, metal oxides were reduced, and metal nanoparticles such as Fe and Al were fabricated. The chemical formulas for the reduction are shown. We first show the chemical formula for Fe3O4 and Fe:

13

Fe3O4 → Fe +23

O2 , ΔH＝373 kcal/mol. (1)

The chemical formula for Al2O3 and Al is shown next:

12

Al2O3 → Al +34

O2, ΔH＝838 kcal/mol. (2)

Fe oxides are resolved and vaporized at a temperature of above 1600 K. Furthermore, Al oxides are resolved and vaporized at a temperature of above 3300 K.

FIG. 1 LASER SYSTEM FOR LASER ABLATION IN LIQUID USING MICROCHIP LASER

The laser ablation method in liquid is described in the paragraph below. When laser pulses are irradiated onto metal oxides in liquid, the metal oxides melt and resolve and the melted oxides are set outside the metal nanoparticles. The surrounding liquid cools the metal nanoparticles rapidly. The merits of the use of this method are as follows: 1) We do not need to use specific materials for reduction. 2) A high reduction rate of metal oxides is obtained because the recombination between oxygen and metals is prevented. 3) The re-collection of nanoparticles is easier than in other methods. 4) It has a very low cost. We obtained reduced Fe nanoparticles with 20 nm diameters by this method [13]. The experimental setup for laser pulse ablation is shown in Fig. 1. A microchip Nd:YAG laser was used in this experiment. The maximum output averaged laser power was 250 mW, the laser wavelength was 1064 nm, the repetitive rate of the laser pulses was 18 kHz, and the pulse duration was 8 ns. A beam with a diameter of 6 mm (1/e2) was focused using a lens with a focal length of 50 mm. Thus, the diameter of the focused beam was 20 µm at the front of each glass bottle. Glass bottles with a size of 20 mmΦ x 5 mm were used in the experiment.

Fe3O4 andα-Al2O3 powders (Koujyundo Chemical Laboratory) were mixed with water in each glass bottle for the experiment. The mean size of each Fe3O4 andα-Al2O3 was 1 µm. Each glass bottle was set after the focused laser beam. The weight of the Fe3O4 andα-Al2O3 powders was measured using an electronic force balance. Their measured weight was 200 mg. 4mL of pure water was placed in each glass bottle. Laser pulses were irradiated to the water with the metal oxide in glass bottle for 20 minutes. Here, the fluence of the irradiated laser pulse was estimated to be 4.5 J/cm2. Ketones, such as acetone, are used as liquids for laser ablation in liquid. Here, we neglect the oxidation at the surface of metal nanoparticles. A magnetic stirrer was used to mix the liquid. After the irradiation of the laser pulses for 20 minutes in the case of Fe3O4, the surface of the nanoparticles was black. Thus, it has been presumed that the surfaces of metal nanoparticles were surrounded by Fe3O4. In the case of α-Al2O3 powder, its color changed to gray, which is close to the color of Al powder. The powder after irradiating laser pulses in the water was dried to not change chemically.


230

Fabrication of metal plates by sintering paste

TABLE I MEASURED ELECTRICAL RESISTANCE

Original powder[Ω] Paste[Ω]

Paste after sintering[Ω]

Fe Large (unmeasurable)

Large (unmeasurable)

0.0

Al Large (unmeasurable)

Large (unmeasurable)

0.0

The dried Fe and Al nanopowders were mixed with 3mg of Ag nanopastes (NAG-10 Daiken Chemical); the viscosity of the paste was high. Fe paste was sintered using an electrical hot plate for 5 minutes (1 minute at 520 K, 4 minutes at 570 K). In the case of the Al paste, the paste was also sintered for total 5 minutes (1 minute at 510K, and 4 minutes at 550K) to generate less gas per unit time. The sintered metal plates are shown in Figs. 2(a) and 2(b). The sintered Fe plate is shown in Fig. 2(a), and the sintered Al plate is shown in Fig. 2(b). The opposite surface of either plate was not metalized by oxidation.

A tester measures the electrical resistance of the metal oxide, and the metal paste, and sintered metal paste. The Fe3O4 andα-Al2O3 powders, and Fe and Al pastes were set on a glass sheet with 1 mm thickness and sintered. The results are shown in Table II. The distance between the needles of the tester was 8 mm. The electrical resistances of the Fe3O4 andα-Al2O3 powders were very high; they could not be measured

because both powders are insulators. The electrical resistances of the metal pastes mixed with reduced Fe and Al nanoparticles by irradiating laser pulses were also measured. However, they were also very high and thus could also not be measured. Finally, after sintering the metal pastes, the measured electrical resistance is 0.0Ω. It was prospected that the Fe and Al pastes were both metalized. For the sintered Al paste prepared by this method, a weak ferromagneticity was observed, and thus it was presumed that the metal structure of Al is markedly different from the structure of common metals. Crystal structure analysis using XRD was performed to check the quantities of Fe3O4 andα-Al2O3 in the sintered metal paste samples. The results are shown in Figs. 3(a) and 3(b). An XRD instrument (MAXima_X XRD-7000 Shimazu Japan) was used for the experiment. K-α X ray radiation of Cu was irradiated onto the samples. The results for analyzing Al and Fe plates are shown in Fig. 3(a) and Fig. 3(b), respectively. In the case of Al, a large spectrum of Al and a small of spectrum of α-Al2O3 were slightly observed. The Al spectrum contained components of a little Ag nanoparticles and the Al stage. Thus, the quantity was not determined.

0

100

200

300

400

500

600

20 30 40 50 60 70 80

Inte

nsity

(arb

. uni

t)

2θ (degree)

AlAl2O3

(a)

0

50

100

150

200

250

300

350

20 30 40 50 60 70 80

Al

FeFe3O4

Inte

nsity

(arb

. uni

t)

2θ (degree)

(b)

FIG. 3 RESULTS OF XRD ANALYSIS: (A) AL AND (B) FE

(a)

(b)

FIG. 2 SINTERED METAL PASTE. (A) FE AND (B) AL


231

Moreover, determining the quantity of Fe was difficult because, normally, the peak spectrum of the Fe3O4 is larger than that of Fe at the same quantity [19]. However, we can see the two components at angles of 45 and 65 degrees in Fig. 3(b), but the Fe component was clearly recognized. The spectrum peak of Fe at 44.7 degrees evaluated by extracting the Al background component is nearly three times higher than that of Fe3O4 at 43 degrees, and it was found that the sintered Fe paste contains negligible quantity of Fe3O4 [19]. Additionally, the sintered structure was observed by a microscopy, and it was found that the porous structure consisted of small metal particles in the metal plates.

Air cell

An experimental setup for testing air cells is shown in Fig. 4. Fe (JIS G3141) and Al (JIS 1050) plates were used as the metal plates in the negative electrodes. The dimensions of the metal plates were set to be 20 mm x 15 mm x 0.5 mm. However, the size of the metal paste was set to be 8 x15 mm2, which is almost half of the metal plate, because of the difference in duration of the output voltage between the use of the metal pastes and that of the conventional metal plates. The electrical property of the sintered metal pastes on the metal plates was compared with that of the conventional metal plates. Here, a Pt-doped carbon electrode with a layer for diffusing oxygen was used as the positive electrode. A separator made of 8 pieces of papers piled up was set between the positive electrode and the negative metal electrode. The thickness of each piece of paper was 0.1 mm. Saturated salt water was injected into each piece of papers at intervals of 5 minutes. The chemical formulas of Fe air cells are shown below:

12

O2 + H2O + 2e− → 2OH− ,

Fe + 2OH− → Fe(OH)2 + 2e− . (3 )

The chemical formulas of Al air cells are also shown:

34

O2 +32

H2O + 3e− → 3OH− ,

Al + 3OH− → Al(OH)3 + 3e− . (4)

The theoretical electromotive forces of Fe and Al air cells are 1.2 and 2.7 V, respectively. However, using NaCl dissolved in water, the output voltages of the cells were 0.4 and 0.7V, respectively. Because the valence of the Al ion is 3 and 3 electrons are emitted

from the negative electrodes per chemical reaction, the electrical power density per unit weight is high. Metal nanopaste could help to connect different metal tightly. It is expected that the output voltage of the air cells will be improved by using different metal junctions. The air cells using different species of metal junctions utilize the phenomenon of galvanic corrosion. For example, new local cells are constructed between Fe nanopaste and an Al plate, and a new electromotive force is generated. Here, Fe corrosion occurred before Al corrosion occurred resulting in the generation of a high output voltage depending on the use of Al as a total air cell.

Electrical Property of Air Cells

Previously, as shown in section 2, metal pastes with Fe and Al were sintered on metal plates, and air cells were fabricated. The electrical properties of the air cells were then investigated. The open output voltage of air cells using a negative Mg electrode was 1.8 V, and that of air cells using negative Al electrode was 1.2 V. That of air cells using a negative Fe electrode was 0.6V. Fe paste on a Fe plate was sintered. Al paste on an Al plate was also sintered. The measured shortcut currents of the air cells using Fe and Al plates were both 70 mA. In the case of Fe and Al pastes, the extracted shortcut current was 40 mA because the area was half of that using Fe and Al plates.

Small connecting load

The temporal property of the output voltage when the electrical load connected to the air cells is small is shown in Fig. 5(a). A 1.0 kΩ resistor was connected to the constructed air cell, and their output voltages were measured. The obtained output voltages of the Fe, Fe paste–Fe and Fe paste–Al, air cells were all 0.4V, one of the Al and Al paste–Al air cells were all 0.7V, and

FIG. 4 EXPERIMENTAL SETUP FOR METAL AIR CELLS


232

that of the Mg and Al paste–Mg air cell were all 1.4V. Current can be evaluated by dividing the output voltage by the 1.0 kΩ resistance. The evaluated output current of the Mg air cell was 1.4mA, that of one of the Al air cells was 0.7 mA, and that of one of the Fe air cells was 0.4 mA.

Large connecting load

A large load, which was applied using an electrical motor (Solar motor, Mabuchi RF-500TB Tamiya, Japan) with a propeller, was connected to the air cells to induce the flow of large currents. The temporal property of the output voltage is shown in Fig. 5(b). The motor rotated normally. However, when connected to the Fe air cells, the motor did not rotate owing to low output voltage. The temporal property of the output voltages was measured when the motor was connected to the air cells. In the case of the Fe paste –Al air cell, the initial output voltage was 0.7 V. In the case of the Al paste –Al air cell, the initial output voltage was also 0.7 V. The output current was 26 mA. The duration of the output voltage was 60 minutes for the Al air cell, and that for the Al paste –Al air cell was

30 minutes, half of that for the Al air cell. Moreover, that for the Fe paste –Al air cell was 30 minutes, which was also half that for the Al air cell. The initial output voltage in the case of the Mg air cell was 1.4 V, and that in the case of the Al paste –Mg air cell was also 1.4 V. The output currents were 28 and 27 mA, respectively. The duration of the output voltage was 80 minutes for the Mg air cell, and that for the Al paste –Mg air cell was 40 minutes, which was also half of that for the Mg air cell.

DISCUSSION

We conducted experiments to produce nanoparticles and reduce metal oxides by laser ablation in liquid. It has been proved in the experiments that metal plates prepared by sintering metal pastes can be obtained, and that the electrical resistances of the metal plates are close to those of conventional metal plates. The Clarke numbers of Al and Fe are 3 and 4, respectively. They exist abundantly in the ground of the earth. The proposed method of reducing and metalizing metal oxides is an alternative to the conventional electrolysis method in aluminium refining. Reduction is performed in two steps; laser ablation and sintering. Ag nanopaste has less ability to reduce metal oxides. Ag paste with only Fe3O4 powder could not be sintered, and no perfect metal plate was fabricated in fact. The inner surface of the paste was not sintered into the metal.

Some heat sources or the solar light can sinter the metal paste. When sintering metal paste, the degradation of surface energy induces heat generation during the chemical reaction. When the paste reaches a given temperature, sintering starts automatically with the generated heat. No heat is required to melt a common metal plate. Required energy to start sintering is adequately lower than the stored energy in the sintered metals owing to the vanishing surface energy.

Here, we consider the input-output energy balance. The injected total laser energy determined by calculation was 290 J when the averaged laser power was 250 mW and the irradiation time was 20 minutes, considering surface loss of each glass bottle. The minimum chemical energy per mol to reduce Fe3O4 to Fe is 373 kJ, that to reduce Al2O3 to Al is 838 kJ. In the case of Fe, when its weight is 200mg, the required minimum energy to reduce it is evaluated to be 960 J. This energy is 3.3 times as large as the injected total laser energy in liquid. In the case of Al, when its

0

0.5

1

1.5

2

-10 0 10 20 30 40 50 60 70

MgAlFeFe paste -AlAl paste -MgFe paste -FeAl paste -Al

Time (Min.)

V out（V（

(a)

0

0.5

1

1.5

2

0 20 40 60 80 100 120

Mg

Al

Fe paste -Al

Al paste -Mg

Al paste -Al

27 mA

28 mA

Time (Min.)

V out（V（

23 mA

(b) FIG. 5 MEASURED OUTPUT VOLTAGE OF METAL AIR CELLS.

THE LOAD WAS (A) 1KΩ AND (B) A MOTOR


233

weight is 200 mg, the required minimum energy to reduce it is evaluated to be 3300J. This energy is 11 times as large as the injected total laser energy. Moreover, more energy for reduction is needed because the generated metal and oxide atoms must be removed far away from metal atoms. These results strongly show that the physics of the reduction process does not depend on conventional thermal ablation. Because the estimated irradiated laser fluence was 4.5J/cm2 in this case, the ablation of the metal oxides occurred by the coulomb explosion. Also, it has been thought that reduction of 200 mg of Fe3O4 or Al2O3 powder can be almost perfectly performed when the averaged laser power is 250 mW, and the irradiation time is 10 minutes. In coulomb explosion, avalanche ionization occurs in metal oxides. Some of the electrons of Fe3O4 are ejected into water. Fe and O atoms are ionized and exploded by the coulomb repulsion between ions. Finally, local plasma is generated. Air cells using sintered metal nanopaste were constructed and their electrical property was investigated. It was clarified that the fabricated air cells using sintered metal nanopaste can be used as primary air cells.

If the weight of oxygen is neglected, the potential electrical energy density per unit weight of Fe used in air cells is estimated to be 1160 Wh/kg, and that of Al is estimated to be 8100 Wh/kg. The potential energy of Al is 7 times higher than that of Fe. The atomic weight of Fe is 55.6, and that of Al is 27. Thus, the potential current of Al is higher than that of Fe, indicating that Al has an advantage for generating electricity.

The electrical resistance of metal plates must be adequately low because currents are extracted from such plates. A large plate made from sintered paste may have a high electrical resistance, resulting in the degradation of the output voltage. Because we did not fabricate large-scale metal plates, the sintered metal plates had lower electrical resistances, and their resistance could not be estimated. However, the resistance is as low as that of common metal plates. A dissimilar metal joint is important for improving the electromotive force of air cells.

By sintering metal paste on common metal plate, a more rigid connection between them is obtained, and contact resistance is reduced. From the experimental result, it has been thought that a dissimilar metal joint between the Al and Mg plates for the negative metal electrode has a low contact resistance. When using a Mg plate, the output voltage of air cells decreased 80

minutes after connecting electric codes. However, when using Al paste and a Mg plate as an air cell, the output voltage was sustained 40 minutes after connecting the codes. Because the output voltages of the Al paste -Al air cells are the same as that of the Al air cells when the load is low, the output voltage will be maintained near 1.4 V for 80 minutes if the area of the metal paste is twofold. After using Fe and Al in air cells, metal oxides are generated. These metal oxides return to Fe and Al by laser ablation in liquid. The negative metal electrode used is exchanged to new ones. In this experiment, to improve the electromotive force of air cells, we had used a Mg plate as a base metal plate. Using Li plates will improve the electromotive force when sintered Al paste is used, and a higher electrical stored energy per unit weight of the air cells will be obtained.

Solar energy or other natural energies are considered as the sources of laser power for laser ablation in liquid. However, the most suitable lasers for energy cycles are considered to be solar-pumped lasers because common lasers have a low electro-optical conversion efficiency and their generated energy gains are vanished.

Repetitive usage of metal oxide by laser ablation and researching the maximum stored energy gain of metals are the future objects.

Conclusions

As an example of renewable energies for utilizing solar energy efficiently, a renewable energy obtained using metallic nanoparticles fabricated by laser ablation in liquid was proposed in this study. Laser pulses are generated using solar power. It has been clarified that metal nanoparticles can be used in metal air cells to generate electricity.

By using high repetitive laser pulses, an alternative to pulsed solar-pumped lasers, Fe3O4 and Al2O3 were reduced, and Fe and Al nanoparticles were fabricated. Metal plates were obtained by sintering paste with Fe and Al nanoparticles at around 520 K. It was found that Fe and Al metal plates have low electrical resistances.

The electrical properties of the air cells using Fe and Al plates fabricated by sintering Fe and Al nanopaste had been compared with those using Fe and Al plates fabricated in a blast furnace. Very close output voltages and currents were obtained. Different species of junction metal plates were used as negative electrodes


234

for the air cells. A higher output voltage of the Al paste–Mg air cells than that of the Al air cells has been obtained. Al was connected to Mg tightly by sintering Al nanopaste. The measured out put voltage of Al paste-Mg air cell was 1.4 V when an electrical motor was connected, which is the same as that of the Mg air cell. Also, a high output voltage of the Fe paste-Al air cell than that of the Fe air cell was obtained.

REFERENCES

[1] P. Charvin, S. Abanades, F. Lemort, and G. Flamant, “Hydrogen Production by Three-Step Solar

Thermochemical Cycles Using Hydroxides and Metal Oxide Systems”, Energy & Fuels, vol. 21, pp.2919-2928, July 2007.

[2] D. G. Rowe, “Solar-powered lasers”, Nature Photonics, vol. 4 pp.64-65, Feb. 2010.

[3] T. Yabe, T. Okubo, S. Uchida, K. Yoshida, M. Nakatuska, T. Funatsu, A. Mabuti, A. Oyama, K. Nakagawa, T. Oishi, K. Daito, “High-efficiency and

economical solar-energy-pumped laser with Fresnel lens and chromium codoped laser medium”, Appl. Phys. Lett., vol. 90 pp.261120-261122, June 2007.

[4] M. S. Mohamed, T. Yabe, C. Baasandash, Y. Sato, Y. Mori, L. Shi-Hua, H. Sato, and S. Uchida, ”Laser-induced magnesium production from magnesium oxide using reducing agents”, J. Appl. Phys., vol. 104

pp.113110-113116, Dec. 2008. [5] M. Weksler and J. Shwartz”, Solar-pumped solid-state

lasers”, IEEE J. Quantum. Electron, vol. 24 pp.1222-1228, July 1988.

[6] T. Saiki, S. Motokoshi, K. Imasaki, K. Fujioka, H. Yoshida, H. Fujita, M. Nakatsuka, and C. Yamanaka, ”Highly-repeated laser pulses amplified by Nd/Cr:YAG ceramic amplifier under CW arc-lamp

light pumping”, Opt. Comm., vol. 282, pp.2556-2559, May 2009.

[7] T. Saiki, S. Motokoshi, K. Imasaki, K. Fujioka, H. Yoshida, H. Fujita, M. Nakatsuka, and C. Yamanaka, “Nd3+- and Cr3+-doped yttrium aluminum garnet ceramic pulse laser using Cr4+-doped yttrium aluminum garnet crystal passive Q-Switch”, Jpn. J. Appl. Phys., vol. 48 pp.122501-1-7, Jan. 2009.

[8] T. Saiki, M. Nakatsuka, K. Imasaki, “Highly efficient lasing action of Nd3+- and Cr3+-doped yttrium aluminum garnet ceramics based on phonon-assisted cross-relaxation using solar light source”, Jpn. J. App. Phys., vol. 49 pp.082702-1-8, Aug. 2010.

[9] A. Henglein, “Physicochemical properties of small metal particles in solution: "microelectrode" reactions, chemisorption, composite metal particles, and the atom-to-metal transition”, J. Phys. Chem., vol. 97 pp.5457-5471, May 1993.

[10] M. S. Sibbald, G. humanov, and T. M. Cotton, “Reduction of cytochrome c by halide-modified, laser-ablated silver colloids”, J. Phys. Chem., vol. 100

pp.4672-4678, Mar. 1996. [11] M. Kawasaki and N. Nishimura,”Laser-induced

fragmentative decomposition of ketone-suspended Ag2O micropowders to novel self-stabilized Ag nanoparticles”, J. Phys. Chem. C, vol. 112 pp.15647-15655, Sep. 2008.

[12] H. Q. Wang,, A. Pyatenko, K. Kawaguchi, X. Y. Li , Z.

Swiatkowska-Wackocka, and N. Koshizaki, ”Selective pulsed heating for the synthesis of semiconductor and metal submicrometer spheres”, Angew. Chem. Int. Ed., vol. 49 pp.6361-6364, Aug. 2010.

[13] S. Taniguchi, T. Saiki, T. Okada, T. Furu, “Synthesis of reactive metal nanoparticles by laser ablation in liquids and its applications”, Report on The 421th Topical Meeting of The Laser Society of Japan, Laser

Technology for 21st Century, No. RTM-11-56, (2011) pp.25-30 [in Japanese].

[14] J. Purnell, E. M. Snyder, S. Wei, and A. W. Castleman, Jr, “Nuclear fusion induced by coulomb exprosion of heteronuclear clusters”, Chem. Phys. Lett., vol. 229 pp.333-339, Nov. 1994.

[15] K. Yamada, K. Miyajima, and F. Mafune, “Ionization of

gold nanoparticles in solution by pulse laser excitation as studied by mass spectrometric detection of gold cluster ions”, J. Phys. Chem. C, vol. 111, pp.033401-1-4, July 2007.

[16] M. Shoji, K. Miyajima, and F. Mafune,”Ionization of gold nanoparticles in solution by pulse laser excitation as studied by mass spectrometric detection of gold cluster ions”, J. Phys. Chem. C, vol. 112 pp.1929-1932,

Jan. 2008. [17] K. Ikawa, T. Horiba, Air Cells, Japan Patent 258782, Oct.

8, 1993. [18] Y. Kudo, M. Suzuki, Al slid-stage Air Cells, Japan

Patent 147442, July 20, 2006. [19] Y. Kataoka, and S. Itushiki, For Analysis by X radiation,

Production Research vol. 12 (1960) p.311 [in Japanese].

International Journal of Energy Science Vol. 2 Iss. 6, November 2012

235

Blends of Diesel – used Vegetable Oil in a Four-Stroke Diesel Engine Charalampos Arapatsakos1

Department of Production and Management Engineering, Democritus University of Thrace

V. Sofias Street, 67100, Xanthi, Greece 1*[email protected]

Abstract

In the days before the proliferation of large cities and industry, nature’s own systems kept the air fairly clean. Wind mixed and dispersed the gases, rain washed the dust and other easily dissolved substances to the ground and plants absorbed carbon dioxide and replaced it with oxygen. With increasing urbanization and industrialization humans started to release more wastes into the atmosphere than nature could cope with. Since then, more pollution has been added to the air by industrial, commercial and domestic sources. There are several many types of air pollutant. These include smog, acid rain, the greenhouse effect and holes in the ozone layer. The atmospheric conditions such as the wind, rain, stability affect the transportation of the air pollutant. This paper examines the use of diesel-used vegetable oil mixtures in a four-stroke diesel engine. The mixtures that have been used are the following: diesel-5% used vegetable oil, diesel-10% used vegetable oil, diesel-20% used vegetable oil, diesel-30% used vegetable oil, diesel-40% used vegetable oil, diesel-50% used vegetable oil. For those mixtures the gas emissions of carbon monoxide (CO), hydrocarbons (HC), nitrogen monoxide (NO), smoke are being measured. Also the gas emissions temperatures are being measured and the consumption for any fuel mixture is examined. The fuel temperatures were 30oC and 40oC.

Keywords

Gas Emissions; Vegetable Oil; Biofuels; Fuel Temperature

Introduction

Air pollution is one of the most serious environmental problems confronting our civilization today. Air pollution is the presence of toxic chemicals or compounds in the air. These compounds may be found into the air in two major forms, in a gaseous and

in a solid form. The most common causes of air pollution are various human activities, including industry, construction, transport agriculture etc. However, there are some natural processes such as volcanic eruptions and wildfires too [1, 2, 3]. The effects of air pollution vary from simply coughing or skin problems to serious diseases, such as cancer, chronic respiratory disease, heart disease etc. People of all ages can be affected from air pollution and particularly from sources such as vehicle exhausts and residential heating, but mainly those with existing heart and respiratory problems are in an extra risk. Air pollutants are also responsible for the acidification of forests and water ecosystems and eutrophication of soils and waters and corrode buildings and materials [4, 5, 6]. One of the main causes of air pollution is transportation and particularly the increased emissions from the road traffic. In order to improve air quality scientists are focusing in the use of alternative fuels that can give energy without harming the environment. Biomass offers a physical way to produce energy without damaging the environment. Biofuels are alcohols, ethers, esters, and other chemicals made from cellulosic biomass such as herbaceous and woody plants, agricultural and forestry residues, and a large portion of municipal solid and industrial waste. The term biofuels can refer to fuels for electricity and fuels for transportation. Unlike petroleum, which is a non-renewable natural resource, biofuels are renewable and inexhaustible source of fuel. Biofuel is used to produce power, heat and steam and fuel through a number of different processes. Consequently, it can be used to power vehicles, heat homes and for cooking. Vegetable oil is an alternative renewable fuel for diesel engines [7, 8, 9]. There are two main types of vegetable oil fuels, the straight vegetable oil and the waste vegetable oil. Straight vegetable oil is the relatively unprocessed or unadulterated oil pressed from a variety of vegetables


236

and plants. These oils can be used for cooking and power vehicles too. Some examples of vegetable oil are palm oil, cottonseed oil and corn oil. Waste vegetable oil is the oil that has already been used for cooking and can no longer be used for that purpose. Both types of oil can be used just as they are or they can be mixed with diesel fuel in engines modified to use them. The use of vegetable oils has many benefits. First of all it is better for your engine as it provides additional lubrication and reduces engine deposits. It is less likely to cause a fire or explosion in the case of an accident. It also results in lower emissions, as the carbon dioxide produced by burning vegetable oil is less than the amount absorbed by the plants from which the oil is obtained, vehicles running on vegetable oil produce no net increase in atmospheric carbon dioxide. Finally, vegetable oil fuel is indefinitely renewable. However, in order to use vegetable oil either straight or waste, it requires engine modification, which is inconvenient and expensive [10].

The major issue is how a four-stroke diesel engine behaves on the side of pollutants and operation, when it uses directly mixed fuel of diesel – used vegetable oil [11].

Instrumentation and Experimental Results

In the experiment stage has been used directly used

vegetable oil (used sunflower oil that emanated from cooking) in the mixture of diesel in to a four – stroke diesel engine. Specifically it has been used diesel, mixture diesel-5% used vegetable oil (tig5), diesel-10 used vegetable oil (tig10), diesel-20% used vegetable oil (tig20), diesel-30% used vegetable oil (tig30), diesel-40% used vegetable oil (tig40), diesel-50% used vegetable oil (tig50) in a four-stroke diesel air-cooled engine named Ruggerini type RD-80, volume 377cc, and power 8.2hp/3000rpm, who was connected with a pump of water centrifugal. Measurements were made when the engine was functioned on 1000, 1500, 2000 and 2500rpm. The fuel temperatures were firstly 30oC and secondly 40oC. During the experiments, it has been counted: The percent of CO, the ppm of HC, the ppm of NO, the percent of smoke, the gas emissions temperature and the fuel consumption. The measurement of rounds/min of the engine was made by a portable tachometer (Digital photo/contact tachometer) named LTLutron DT-2236. Smoke was measured by a specifically measurement device named SMOKE MODULE EXHAUST GAS ANALYSER MOD 9010/M, which it has been connected to a PC unit. The CO and HC emissions have been measured by HORIBA Analyzer MEXA-324 GE. The NO emissions have been measured by a Single GAS Analyser SGA92-NO. The experimental results are shown at the following figures:

PIC. 1 EXPERIMENTAL LAYOUT


237

30oC

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

1000 1500 2000 2500

rpm

CO%

dieseltig5tig10tig20tig30tig40tig50

FIG. 1 THE CO VARIATION ON DIFFERENT ENGINE RPM REGARDING TO THE MIXTURE, WHEN THE FUEL TEMPERATURE IS 30OC

40oC

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

1000 1500 2000 2500

rpm

CO%


FIG. 2 THE CO VARIATION ON DIFFERENT ENGINE RPM REGARDING TO THE MIXTURE, WHEN THE FUEL TEMPERATURE IS 40OC

30oC

0

10

20

30

40

50

60

1000 1500 2000 2500

rpm

HC(

ppm

)


FIG. 3 THE HC VARIATION ON DIFFERENT ENGINE RPM REGARDING TO THE MIXTURE, WHEN THE FUEL TEMPERATURE IS 30OC


238

40oC

0

10

20

30

40

50

60

1000 1500 2000 2500

rpm

HC(

ppm

) dieseltig5tig10tig20tig30tig40tig50

FIG. 4 THE HC VARIATION ON DIFFERENT ENGINE RPM REGARDING TO THE MIXTURE, WHEN THE FUEL TEMPERATURE IS 40OC

30oC

0

200

400

600

800

1000

1200

1400

1600

1800

1000 1500 2000 2500

rpm

NO(p

pm)


FIG. 5 THE NO VARIATION ON DIFFERENT ENGINE RPM REGARDING TO THE MIXTURE, WHEN THE FUEL TEMPERATURE IS 30OC

40oC

0

200

400

600

800

1000

1200

1400

1600

1800

1000 1500 2000 2500

rpm

NO(p

pm)


FIG. 6 THE NO VARIATION ON DIFFERENT ENGINE RPM REGARDING TO THE MIXTURE, WHEN THE FUEL TEMPERATURE IS 40OC

International Journal of Energy Science Vol. 2 Iss. 6, November 2012

239

30οC

0

10

20

30

40

1000 1500 2000 2500

rpm

smok

e%


FIG. 7 THE SMOKE VARIATION ON DIFFERENT ENGINE RPM REGARDING TO THE MIXTURE, WHEN THE FUEL TEMPERATURE IS

30OC

40oC

0

10

20

30

40

1000 1500 2000 2500

rpm

smok

e%


FIG. 8 THE SMOKE VARIATION ON DIFFERENT ENGINE RPM REGARDING TO THE MIXTURE, WHEN THE FUEL TEMPERATURE IS

40OC

0

50

100

150

200

250

300

1000 1500 2000 2500

rpm

gas t

empe

ratu

re (o C)


FIG. 9 THE GAS TEMPERATURE VARIATION ON DIFFERENT ENGINE RPM REGARDING TO THE MIXTURE

In the case of 30oC as fuel temperature:

From figure 1 it can be noticed that the most constant behaviour appeared in the mixture of tig40, while the best behaviour appeared in the case of diesel at

1500rpm.

From figure 3 it can be noticed that the biggest reduction of HC emissions regarded to diesel presented in the mixture of tig 40. Figure 5 show that


240

the biggest reduction of NO emissions regarding to diesel appeared in the mixture of tig40. Finally, from figure 7 it can be said that the biggest reduction of smoke emissions regarding to diesel appeared in the mixtures of tig30 and tig40.

In the case of 40oC as fuel temperature:

From figure 2 it is clear that mixtures tig5, tig10, tig20, tig30, tig40 and tig50 presented lower CO emissions regarding to diesel. From figure 4, it can be seen a reduction of HC emissions when using different mixtures than diesel. In figure 6 it is also presented a reduction of NO emissions regarding to diesel with the exception of the engine functioned on 2000 rpm, in where the diesel presented lower NO emissions than the mixtures. Finally, from figure 8, it can be seen that mixtures tig10, tig20, tig30, and tig50 presented lower smoke emissions than diesel. However, when the engine functioned on 1000, 1500 and 2000 rpm, the mixture tig40 presented higher smoke emissions than diesel. On the other hand, the mixture tig5 presented lower smoke emissions than diesel with the exception of the engine functioned on 2500 rpm, in where the smoke emissions were higher than diesel.

From the above figures it can be concluded that the use of different mixtures can constitute changes to CO, HC, and NO and smoke too. It is also important to mention that there were no changes in the rounds of the engine, as well as in the supply of water during the use of mixtures. As far as the gas emissions temperature (fig. 9) and the fuel consumption is concerned, did not observed any changes with the use of different mixtures on the different fuel temperatures.

Conclusion

The use of mixtures diesel-used vegetable oil has as result the gas emissions variation. Better behaviour presented in the mixtures of tig30 and tig40. The density and viscosity of those mixtures did not create any problems in the spraying of fuel. As it has already been mentioned above the different fuel temperatures (30oC, 40oC) differentiate the gas emissions.

It is also important to mention, that during the combustion of the mixtures there was not presented any reduction in the power of the engine.

Finally, it has not been presented engine malfunction from the directly use of fuel mixtures diesel - used vegetable oil.

References

[1] C. Arapatsakos, “Air and water influence of two

stroke outboard engine using gasoline - ethanol

mixtures”. Transaction of SAE, Book SP-1565, 2000.

[2] C. Arapatsakos, “Testing the tractor engine using

diesel – ethanol mixtures under full load conditions”.

International Journal of Heat & Technology, Vol. 19,

n.1, 2001.

[3] C. Arapatsakos, A. Karkanis, P. Sparis, “Gas

emissions and engine behavior when gasoline-

alcohols mixtures are used” Journal of

Environmental Technology, Vol. 24, pp. 1069-1077.

[4] C. Arapatsakos, A. Karkanis, P. Sparis,

“Environmental Contribution of Gasoline- Ethanol

Mixtures” WSEAS Transactions on Environment and

Development, Issue 7, Volume 2, July 2006.

[5] S. Siddharth. “Green Energy-Anaerobic Digestion.

Converting Waste to Electricity” WSEAS

Transactions on Environment and Development,

Issue 7, Volume 2, July 2006.

[6] William Ernest Schenewerk “Automatic DRAC

LMFBR to Speed Licensing and Mitigate CO2”

WSEAS Transactions on Environment and

Development, Issue 7, Volume 2, July 2006.

[7] Timothy T. Maxwell and Jesse C. Jones “Alternative

fuels: Emissions, Economics and Performance”

Published by SAE, 1995.

[8] C. Arapatsakos, A. Karkanis, P. Sparis,

Environmental pollution from the use of alternative

fuels in a four-stroke engine, International journal of

environment and pollution 21 (2004) 593-602.

[9] C. Arapatsakos, A. Karkanis, P. Sparis, Tests on a

small four engine using gasoline-ethanol mixtures as

fuel, Advances in air pollution 13 (2003) 551-560.

[10] C. Arapatsakos, A. Karkanis, P. Sparis, Gas

emissions and engine behaviour when gasoline-

alcohol mixtures are used, Environmental

technology 24 (2003) 1069-1077.

[11] C. Arapatsakos, D. Christoforidis, A. Karkanis, The

use of vegetable oils us fuel on diesel engine

International journal of heat and technology. Vol 29,

No 1, pp. 25-31, 2011.


241

Catalytic Pyrolysis by Heat Transfer of Tube Furnace for Produce Bio-Oil Kittiphop Promdee1,2*, Tharapong Vitidsant3 1 Inter-Department of Environmental Science, Chulalongkorn University, Bangkok, Thailand

2 Department of Environmental Science, Chulachomklao Royal Military Academy, Nakorn Nayork, Thailand

3 Department of Chemical Technology, Chulalongkorn University, Bangkok, Thailand [email protected]; 2 [email protected] ; 3 [email protected]

Abstract

Catalytic pyrolysis by heat transfer model can be solved the control temperature in tube furnace to produce bio-oil by continuous pyrolysis process and this study concern the products yield of bio-oil from mixed biomass consist of Cogongrass, Manilagrass and the leaf of trees, which conducted temperatures in the range of ~ 400-550°C, considering the feed rate of 150, 350, and 550 rpm (r·min−1)]. Preliminary result of proximate analysis was founded that the high volatile matter, low ash and moisture . The products yield calculation showed that the liquid yield of bio-oil was highest of 55.60 %, and 45.45%., at 350 rpm and 550 rpm., respectively, the solid yield of bio-oil was highest of 27.35 %, at 350 rpm, and the gas yield of bio-oil was highest of 43.60 %, at 150 rpm. Indicated that biomass from mixed biomass had good received yields because of low solid yield and gas yield and high liquid yield. The compounds detected in bio-oil from mixed biomass showed that the functional groups, especially; phenols. For the purpose that; in this research not only concern the feed rate and the heat transfer for contact biomass but also concern the control gas flow and temperature balanced.

Keywords

Catalytic Pyrolysis; Heat Transfer; Continuous Pyrolysis Reactor; Received Oil Yield

Introduction

This research was conducted by using mixed biomass transformed to bio-oil by continuous pyrolysis reactor on standard criteria and analysis the properties of material and products. In present, the fuel is being concerned in every country [1-3]. Now we are looking at the fuel which synthesized from natural matter, especially; residual plant[5-6], by using the pyrolysis method combined with the theory of heat transfer for control the temperature balance in the continuous pyrolysis reactor (tube furnace)(Fig 1). The fuels from natural matter have a good solve and can reduce a

waste in widespread areas of central part of Thailand. Continuous pyrolysis reactor is a one excellent of technology for synthesized bio-oil [7-9]. In this case want to produce bio-oil in high potential performance of yield and properties by applied the heat transfer model for control some criteria of reactor to an generate the alternative energy source.

FIG. 1 CONDUCTION IN A CONTINUOUS PYROLYSIS CYLINDER WITH UNIFORM HEAT GENERATION

To determine the temperature distribution in the cylinder reactor, the appropriate from of the heat equation. For constant thermal conductivity k, Equation (1) reduces to

0.1=+

=

•

kq

drdTr

drd

r (1)

Separating variables and assuming uniform generation, this expression maybe integrated to obtain

=⋅drdTr 1

2

2Cr

kq

+•

(2)

Repeating the procedure, the general solution for the temperature distribution becomes

212 ln

4)( CrCr

kqrT ++−=•

(3)

To obtain the constants of integration C1 and C2, we apply the boundary conditions

mailto:[email protected]



242

00=

=rdrdT

and so TrT =)( (4)

From the foregoing symmetry condition at r = 0 and Equation (2), it is evident that C1 = 0. Using the surface boundary condition at r = ro with Equation (3), we then obtain

22 4 os r

kqTC•

+= (5)

The temperature distribution is therefore

so

o Trr

krqrT +

−

=

•

2

22

14

)( (6)

Evaluation Equation (6) at the centerline and dividing the result into Equation (6), we obtain the temperature distribution in nondimensional form

2

1)(

−=

−−

oso

s

rr

TTTrT

(7)

Where To is the centerline temperature. The heat rate at any radius in the pyrolysis cylinder may, of course, be evaluated by using Equation (6) with Fourier’s law [10,11]. To relate the surface temperature, Ts, to the temperature, T∞, of the cold fluid, either a surface energy balance or an overall energy balance may be used. Choosing the second approach, we obtain

))(2(( 2∞−= TTLrhLrq soo

o ππ

Or

+=

•

∞ hrqTTs 2

(8)

The foregoing approach may also be used to obtain the temperature distribution in cylindrical and in solid spheres for a variety of boundary conditions [11].

Experimental

Feedstock and Experimental Set-up

Preparation of mixed biomass, crust and bring to oven at 100 oC for ~ 2 hr until it is completed dry or less than 5 percent moisture. The samples were separated through a sieve to the approximate 450-1,000 microns.

The samples were fed to continuous reactor (Fig. 2), for pyrolysis process at ~ 400-550 oC and control the N2 flow rate around 0, 50, 100, 150, 200, 400 ml/hr and feed samples averaging of 150, 350, and 550 rpm

(r·min−1). The bio-oil product were analyzed by Ultimate analyzer, Proximate analyzer, calculate the received oil yields and analyze the chemical compound by Gas Chromatography with Mass Spectrometer.

FIG. 2 SCHEMATIC DIAGRAM OF EXPERIMENT SETUP: 1.

PYROLYSIS REACTOR (TUBE FURNACE) 2. NITROGEN TANK 3. ROTAMETER 4. HOPPER 1,2 5. SEPARATOR 6. CONDENSER 7. FLASK IN ICE BUCKET 8. ELECTRICAL COIL HEATER WITH

TEMPERATURE CONTROLLER 9. ENCLOSED DEIONIZER WATER TANK 10. VACUUM PUMP

Proximate Analysis

Proximate analysis is the most used analysis for characterizing biomass in connection with their utilization, this experiment was analysis by ASTM D 3173-3175. The process are determined the distribution of products obtained when the sample is heated under specified conditions. Proximate analysis separates the products into 4 groups: (1) moisture, (2) volatile matter, (3) fixed carbon, the nonvolatile fraction of char, and (4) ash.

Ultimate Analysis

In the experiment was analysis form of element components of bio-oil concerned with determination of only Carbon (C), Hydrogen (H) and Nitrogen (N) in a sample, these analyzed by Ultimate analyser [12-13].

Received Oil Yield

% Liquid yield

×=

ini

Liq

WW

100

% Solid yield

×=

ini

R

WW100

% Gas yield −=100 % Liquid yield - % Solid yield

Wini = Initial weight

WR = Residual solid weight

WLiq = Liquid product weight


243

Chemical analysis

Gas Chromatography with Mass Spectrometer, GC-MS was used to analyse the light components in bio-oil and investigating the molecular compositions qualitatively [14-15]. The analyses detective and identify organic compounds both aliphatic hydrocarbon and aromatic hydrocarbon.

Results and Discussion

Preliminary, proximate analysis of mixed biomass used in the three species (Cogongrass, Manilagrass and the leaf of trees) was founded that the fixed carbon of mixed biomass was 17.28 wt.%, which will have a major effect on the quality of bio-oil as well. The other three proximate analysis as following; The moisture, ash and volatiles of mixed biomass were 2.51, 17.00 and 63.21 wt.%, respectively (Table. 1). The results showed that the stability for the range of material compound in mixed biomass can be synthesized bio-oil in high efficiency, because consist of the high volatile matter and low ash and moisture.

TABLE 1 PROXIMATE ANALYSIS AND ULTIMATE ANALYSIS OF MIXED BIOMASS

Proximate analysis (wt.%)

mixed biomass Ultimate

analysis (wt.%)

mixed biomass

Moisture 2.51 C 38.23

Ash 17.00 H 5.27

Volatiles 63.21 N 1.00

Fixed carbon 17.28 O 55.16

The ultimate analysis of mixed biomass was found that the element contents as following; carbon, hydrogen, nitrogen and oxygen were 38.23, 5.27, 1.00 and 55.16 %, respectively (Table. 1)., according to the result of safflower seed [12,13] showed the carbon, hydrogen, nitrogen, and oxygen of 49.5, 6.9, 3.0, and 40.6, respectively.

The result of products yield (3-phase; gas, liquid and solid) of bio-oil by during pyrolysis, which takes place at temperatures in the range of ~ 400-550°C, to compare the received oil yield from mixed biomass at a feed rate of feed rate of 150, 350 and 550 rpm; revolutions per minute (r·min−1)]. Preliminary calculate of the product oil yield of mixed biomass, the result showed that the gas yield of bio-oil obtained mixed biomass were 43.6, 19.05 and 29.25 %, at 150, 350 and 550 rpm., respectively., (Fig. 3).

FIG. 3 GAS YIELD OF BIO-OIL OBTAINED FROM MIXED BIOMASS

Liquid yield of bio-oil obtained from mixed biomass was highest of 55.6 %, at 350 rpm. And the another of liquid yield obtained from mixed biomass were 29.55 and 45.45 %, at 150 and 550 rpm., respectively (Fig. 4). Indicated that the liquid yield of bio-oil obtained from mixed biomass was high volume (> 50 %) by the heat control in continuous pyrolysis reactor and can be improving to high efficiency of bio-oil production on next step.

FIG. 4 LIQUID YIELD OF BIO-OIL OBTAINED FROM MIXED

BIOMASS

FIG. 5 SOLID YIELD OF BIO-OIL OBTAINED FROM MIXED

BIOMASS


244

Solid yield of bio-oil obtained from mixed biomass was highest of 27.35 %, , at 350 rpm. And the another of solid yield obtained from mixed biomass were 25.85and 24.3 %, at 150 and 550 rpm., respectively (Fig. 5).

TABLE 2 COMPOUNDS DETECTED IN BIO-OIL FROM MIXED BIOMASS

Compound

% formula MW Detection

1,2-Benzenediol 0.5

×

1,2-Cyclopentanedione, 3-methyl-

0.82 C6H8O2 112.12 ×

2-Cyclopenten-1-one, 2,3-dimethyl- 1.08 ×

Phenol 19.78

C6H6O 94.11 ×

Phenol, 2,3-dimethyl- 2.23

C8H10O 122.16 ×


×


C8H10O 122.16 ×

Phenol, 2,6-dimethoxy-

12.47

×

Phenol, 2,6-dimethyl- *

C8H10O 122.16

Phenol, 2-ethyl- 0.85

C8H10O 122.16 ×

Phenol, 2-methoxy- 3.81

×

Phenol,2-methoxy-4-(1-propenyl)-, (E)- * C10H12O2 164.19

Phenol, 2-methoxy-4-methyl- 0.72

C10H12O2 164.19 ×

Phenol, 2-methoxy-4-propyl-

C10H12O2 164.19 ×

Phenol, 2-methyl- 2.87

C7H8O 108.13 ×


×

Phenol, 3-methyl- 3.86

C7H8O 108.13 ×

Phenol, 4-ethyl- 1.54

C8H10O 122.16 ×

Phenol, 4-ethyl-2-methoxy-

2.23 C9H12O2 152.18 ×

*can not determined

The compounds detected in bio-oil from mixed biomass showed that the hydrocarbon compounds compose of hydroxyl and carboxyl groups, especially; phenols (Phenol, 2,3-dimethyl-, Phenol, 2,6-dimethoxy-4-(2-propenyl)-, Phenol, 2-ethyl-4-methyl-, Phenol, 2-methoxy-, Phenol, 3-methyl-, Phenol, 4-ethyl-2-methoxy-), alcohols, and ketones (Table. 2) as same the result of pyrolysis two energy crops [15] and the other result of pyrolysis biomass [3-5-16,17,18,19,20].

Conclusions

The continuous pyrolysis reactor for produce bio-oil from mixed biomass showed the proximate analysis of mixed biomass presented a high volatiles content and showed a moderate level of fixed carbon. The amount of the elemental composition of mixed biomass, can found the concentration of carbon was relatively high volume. The products oil yield showed that liquid yield of bio-oil obtained from mixed biomass was a good result was 55.6 %, at 350 rpm. Also the result of solid yield bio-oil obtained from mixed biomass as same a high volume at 350 rpm. The compounds detected in bio-oil from mixed biomass can found the phenols, alcohols, and ketones, especially; phenols group. Thus, in this research, the process of continuous pyrolysis depended on the mechanism of heat transfer with cylinder shape. If control the N2 flow, control temperature system balance, according to as good as the heat transfer model, the yields and qualities of bio-oil should to be high efficiency and the another concern that to the overall performance system of the continuous pyrolysis reactor.

ACKNOWLEDGMENT

This work was supported by the Higher Education Research Promotion and National Research University Project f Thailand, Office of the Higher Education Commission (Project Code : EN272A), and thank you for Department of Chemical Technology, Faculty of Science, Chulalongkorn University, were advised and supported the laboratory for experiment and analysis in this research.

REFERENCES

[1] M. Patel, M. Neelis, D. Gielen, J. Olivier, T. Simmons,

and J., Theunis, “Carbon dioxide emissions from non-

energy use of fossil fuels: Summary of key issues and

conclusions from the country analyses,” Resources,

http://www.sciencedirect.com/science/article/pii/S0921344905000704?_alid=1857914214&_rdoc=12&_fmt=high&_origin=search&_docanchor=&_ct=11052&_zone=rslt_list_item&md5=0bf9249dd3b80567faae6d71ec8932b2




245

Conservation and Recycling , Vol 45, Issue 3, pp.195-209,

Jul. 2005.

[2] M. Garcia-Perez, J. Shen, X. S., Wang, and C-Z Li,

“Production and fuel properties of fast pyrolysis

oil/bio-diesel blends,” Fuel Processing Technology, Vol

91, Issue 3, pp.296-305, Mar. 2010.

[3] G. Duman, C. Okutucu, S. Ucar, R. Stahl, and J., Yanik,

“The slow and fast pyrolysis of cherry seed,”

Bioresource Technology, Vol 102, Issue 2, pp.1869-1878,

Jan. 2011.

[4] D. Wu, S. Zhang, J. Xu and T., Zhu, “The CO2

Reduction Effects and Climate Benefit of Beijing 2008

Summer Olympics Green Practice ,” Energy Procedia,

Volume 5, pp. 280-296, Apr.2011.

[5] Q. Lu, W-Z. Li, and X-F., Zhu, “Overview of fuel

properties of biomass fast pyrolysis oils,” Energy

Conversion and Management, 50, pp.1376–1383, May

2009.

[6] H. Li, Q. Xu, H. Xue, and Y., Yan. 2009, “Catalytic

reforming of the aqueous phase derived from fast-

pyrolysis of biomass,” Renewable Energy, pp.1-6, Apr.

2009.

[7] H.S. Heo, H.J. Park, J-H. Yim, J.M. Sohn, J.H. Park, S-S.

Kim, C.K. Ryu, J-K. Jeon, and Y-K., Park. “Influence of

operation variables on fast pyrolysis of Miscanthus

sinensis var. purpurascens,” Bioresource Technology, Vol

101, Issue 10, 2010, pp.3672-3677.

[8] K. Promdee, T. Vitidsant, and S. Vanpetch.

“Comparative study of some physical and chemical

properties of bio-oil from Manila grass and Water

hyacinth transformed by pyrolysis process”,

International Journal of Chemical Engineering and

Applications, Vol.3, No.1, pp.72-75. Feb. 2012.

[9] K. Promdee, and T. Vitidsant, and S. Vanpetch: J. Chem

Eng and App. 3, 1, Feb 2012.

[10] F.P. Incropera, and D. P. De Witt. Fund of Heat and Mass

Transfer. 2 1985.

[11] T.G. Hicks. Mechanical engineering formulas. 1, 2003.

[12] S. Sevgi, and D., Angin, “Pyrolysis of safflower

(Charthamus tintorius L.) Seed press cake: Part 1. The

effect of pyrolysis parameters on the product yields,”

Bioresource Technology, 99, pp.5492-5497. Sep. 2008.

[13] S. Sevgi, and D., Angin, “Pyrolysis of safflower

(Charthamus tintorius L.) Seed press cake in a fixed-bed

reactor: Part 2. Structural characterization of pyrolysis

bio-oils,” Bioresource Technology, 99, pp.5498-5504, Sep.

2008.

[14] P. Duan, and P. E., Savage, “Upgrading of crude algal

bio-oil in supercritical water,” Bioresource Technology,

Vol 102, pp.1899-1906, Jan. 2011.

[15] C.A. Mullen, A. Charles, and A., Akwasi, “Chemical

Composition of Bio-oils Produced by Fast Pyrolysis of

Two Energy Crops,” Energy & Fuels, pp.2104-2109, Apr.

2008.

[16] P.M. Mortensen, J-D. Grunwaldt, P.A. Jensen, K.G.

Knudsen, and A.D., Jensen,”A review of catalytic

upgrading of bio-oil to engine fuels,” Applied Catalysis,

Vol 407, Issues 1-2, pp.1-19, Aug. 2011.

[17] R. Razuan, Q. Chen, N.K.Finney, V.N. Russell, N.V.

Sharifi, and J., Swithenbank. Combustion of oil palm

stone in a pilot-scale fluidized bed reactor. Fuel

Processing Technology, 92 (2011), p. 2219-2225.

[18] L. Mei-kuei, T. Wem-tien; S. Yi-lin and L. Sheau-horng.

Pyrolysis of Napier Grass in an Induction-heating

Reactor. Analytical and Applied Pyrolysis, (2010),Vol 88,

Issues 2, pp. 110-116, Jul. 2010.

[19] H. Kazemi Esfeh, B. Ghanavati, and T. GhaleGolabi.

“Properties of modified bitumen obtained from natural

bitumen by adding pyrolysis fuel oil”, International

Journal of Chemical Engineering and Applications, Vol.2,

No.3, pp.168-172. Jun. 2011.

[20] U. Jena, and K. C. Das. “Comparative evaluation of

thermochemical liquefaction and pyrolysis for bio-oil production from microalgae. Energy฿fuels”, 25 (2011),

p.5472-5482, Sep. 2011.

Kittiphop Promdee was born in Ubonrachathani Province, Thailand, in 1975. He received the M.S. degree in environmental science from the school of environmental science, Kasetsart University, Bangkok, Thailand, in 2004. He is currently a lecturer with Department of Environmental Science, Chulachomklao Royal Military Academy, Nakorn Nayork, Thailand. His research interests include agricultural residuals, renewable energy, fuel, environmental energy, green and biomass technology and catalytic pyrolysis processes. He is currently pursuing the Ph.D. degree with Inter-Department of Environmental Science, Chulalongkorn University, and Bangkok, Thailand.

http://www.sciencedirect.com/science/journal/03783820

http://www.sciencedirect.com/science?_ob=PublicationURL&_hubEid=1-s2.0-S0378382010X00029&_cid=271404&_pubType=JL&view=c&_auth=y&_acct=C000228598&_version=1&_urlVersion=0&_userid=10&md5=77b8627e13bcf2224e5834faef8c2c17



http://www.sciencedirect.com/science/article/pii/S0960852410012411?_rdoc=1&_fmt=high&_origin=article&_docanchor=&_ct=101&_zone=related_art_hover&md5=b3d1e89cd16561b51fa210de3108e2ce

http://www.sciencedirect.com/science/article/pii/S1876610211009866?_alid=1857914214&_rdoc=5&_fmt=high&_origin=search&_docanchor=&_ct=11052&_zone=rslt_list_item&md5=d21d8342bc9e70317d4f6a02a7eb2a41



http://www.sciencedirect.com/science/article/pii/S096085240901743X?_rdoc=1&_fmt=high&_origin=article&_docanchor=&_ct=11616&_zone=related_art_hover&md5=824f19db144d3c40153be23463afa6e6




http://www.sciencedirect.com/science/article/pii/S0960852410013659?_alid=1857899632&_rdoc=3&_fmt=high&_origin=search&_docanchor=&_ct=11521&_zone=rslt_list_item&md5=3f9fe84494717220678c5f14691331c4



http://www.sciencedirect.com/science/article/pii/S0926860X11005138?_alid=1857899632&_rdoc=9&_fmt=high&_origin=search&_docanchor=&_ct=11521&_zone=rslt_list_item&md5=92b8c08763257863708b206788f78faf




246

Tharapong Vitidsant received the B.S. degree and the M.S. degree in c hemica l engineering from Chula longkorn University, Bangkok, Thailand. He received the Ph.D. degree from. Institut National Polytechnique de Toulouse (INPT), France, in 1999. He is currently an associate professor with

Department of Chemical Technology, Chulalongkorn University. His research interests include renewable energy, fuel energy, catalytic pyrolysis processes and reaction engineering.


247

CMOS Bandgap Reference and Current Reference with Simplified Start-Up Circuit Guo-Ming SUNG, Ying-Tzu LAI, Chien-Lin LU

Department of Electrical Engineering, National Taipei University of Technology

1, Sec. 3, Chung- Hsiao E. Rd. Taipei 10608 Taiwan [email protected]; [email protected]; [email protected]

Abstract

This paper presents a CMOS bandgap reference and current reference based on the resistor compensation. The proposed architecture consists of various high positive temperature coefficient (TC) resistors, a two-stage operational transconductance amplifier (OTA) and a simplified start-up circuit in 0.35-µm CMOS process. In the proposed bandgap reference and current reference, numerous compensated resistors, which have a high positive temperature coefficient (TC), are added to the parasitic n-p-n and p-n-p bipolar junction transistor devices, to generate a temperature-independent voltage reference and current reference. With the simplified start-up circuit, the proposed resistor-compensation bandgap reference and current reference can be started at 43 ns at a minimum supply voltage of 1.35 V. The measurements verify the current reference of 735.6 nA, the voltage reference of 888.1 mV, and the power consumption of 91.28 µW at a supply voltage of 3.3 V. The voltage TC is 49 ppm/ in the temperature range from 0oC to 100oC and 12.8 ppm/ from 30oC to 100oC. The current TC is 119.2 ppm/ at temperatures of 0oC to 100oC. Measurement results also demonstrate a stable voltage reference at high temperature (> 30oC), and a constant current reference at low temperature (< 70oC).

Keywords

Bandgap Reference; Current Reference; Resistor Compensation; Temperature Coefficient; Start-up Circuit

Introduction

A bandgap reference is extensively adopted in several integrated circuits, including analog, mixed-mode and memory circuits. In CMOS bandgap reference design, the sub-1-V curvature-compensated CMOS bandgap reference is favored for use with resistive subdivision methods [1-3], parasitic n-p-n and p-n-p bipolar junction transistor devices [4], and the compensation approaches associated with layout, operational amplifiers, and VEB linearization [5-7]. However, the low-voltage bandgap reference frequently functions with a higher temperature coefficient than the conventional bandgap reference [4]. To reduce the

temperature coefficient further, many curvature compensations have been introduced. Guan et al. developed a current mode curvature-compensated BGR (bandgap reference) using the trimming approach [8]. Leung et al. introduced second-order curvature compensation based on resistors with opposing temperature coefficients [9]. Audy introduced a third- order curvature compensation with a low-TCR resistor in parallel with a high-TCR resistor and in series with low-TCR tail resistors [10]. Malcovati had designed a high-order curvature temperature compensation method with Malcovati topology [7]. Chen also presented a programmable and high-precision temperature independent current reference by adding a positive TC current to a negative TC circuit [11]. However, the simulation result had not been verified yet.

Start-up circuits are required to prevent the bandgap reference from operating at the zero point. The components of a start-up circuit must be limited to three or less. Xing et al. developed a start-up circuit that comprised three NMOSs, MS1-MS3 [5]. Xiaokang Guan et al. also introduced a start-up circuit with a resistor R6 and two PMOSs, M7-M8 [8]. Ker further developed two start-up circuits in the proposed bandgap reference. One consists of one PMOS and two NMOSs, MSN1-MSN3. The other comprises a NMOS and two PMOSs, MSP1-MSP3 [4]. Unfortunately, as is well known, the rise times of the proposed start-up circuits are long. More effort must be made to design a high-speed start-up circuit. Additionally, the current trend in industry is toward new bandgap references with simultaneous voltage reference and current reference [3], [12]. If a temperature-independent current reference is required, then the bandgap reference must generally be divided by a resistance. However, the resistance depends on temperature. Accordingly, a current reference also depends on a curvature-compensation method. This work presents a resistor-


248

compensation CMOS bandgap and current reference with a current reference of 735.6 nA and a voltage reference of 888.1 mV at a supply voltage of 3.3 V in a standard 0.35-µm CMOS process. The proposed bandgap and current reference, which comprises a two-stage operational transconductance amplifier (OTA) with differential NMOS input stage, is validated. Notably, the voltage TC is 49 ppm/ in the temperature range from 0oC to 100oC, and 12.8 ppm/ from 30oC to 100oC; the power consumption is 91.28 µW. Hence, section II describes the basic principles and circuit designs associated with the proposed bandgap and current reference. Section III presents and discusses experimental results. Conclusions are finally drawn in Section IV, along with recommendations for future research.

Basic Principles and Circuit Design

Basic Principles of CMOS Bandgap Reference

The temperature-independent bandgap reference is the conventionally adopted voltage reference because temperature commonly skews the operating point and affects noise in the semiconductor. A bandgap reference working with zero TC, which has both positive TC and negative TC, is therefore required. Figure 1 depicts the basic framework of a bandgap reference [13-15]. The output reference voltage, Vref, is

tEBref VKVV ⋅+= (1)

where K is a constant which is normally equal to 17.2 [13], VEB is the voltage difference between the emitter and the base in bipolar junction transistors (BJT) and Vt is the thermal voltage. VEB typically exhibits a negative-TC characteristic, while thermal voltage Vt usually has a positive TC. Vt is realized from the difference between two emitter-base voltages, ∆VEB, which is proportional to the absolute temperature (PTAT). However, the output voltage of the bandgap reference suffers from variation with temperature, even when curvature-compensation is considered. To solve this problem, an improved cascading CMOS bandgap reference (BGR) with second-order curvature-compensated circuit has been presented [16]. However, it does not plot a temperature-independent current reference Iref. To have both voltage reference and current reference in a temperature-independent bandgap reference, the curvature-compensated method must be further investigated.

FIG. 1 BASIC FRAMEWORK FOR BANDGAP REFERENCE

Proposed Resistor Compensation Circuit

Figure 2 presents the schematic of the proposed temperature-independent bandgap reference and current reference where a two-stage operational transconductance amplifier (OTA) is used to replace a traditional cascade current mirror. Notably, Resistors, R3 and Rb3, and BJT, Q3, compensate for the output voltage reference, Vref, in what is called second-order curvature compensation [9]. Resistors Ra1 and Ra2 have two purposes. The first is to increase the input voltages, Vin+ and Vin-, of the N-type stage OTA to ensure that the OTA works properly. The second is to compensate for the temperature-dependent variation of Vin+ and Vin-. The OTA is employed to equality Vin+ and Vin-. Transistors Q2 and Q1 are vertical PNPs with a base-emitter area ratio of 5:2, passing the same current, such that the current though R2 is PTAT. Transistors Q1 and Q3 are identical. Rb1, Rb2 and Rb3 are added to produce the second-order curvature compensation circuit [16]. Notably, the base-emitter area ratio of Q2 and Q1 is smaller than the traditional ratio, because current compensation is performed using several resistors. Therefore, the positive TCs of Rb1, Rb2 and Rb3 compensate for the negative temperature coefficients of Q1, Q2 and Q3, respectively. MOSFETs M1-M4, are also identical to each other. They equalize the currents I1, I2, I3 and Iref, and provide a temperature-independent current reference Iref. Restated, the proposed bandgap and current reference simultaneously provides both a temperature-independent voltage reference Vref and temperature-independent current reference Iref.

Next, the resistor-compensation circuit is described in deta il . The emitt er-base voltage VE B of BJT has a negat iv e TC w her eas t he emitt er -b ase v oltage difference ∆VBE and all resistances have a positive TC.


249

FIG. 2 SCHEMATIC OF PROPOSED RESISTOR-COMPENSATION

BANDGAP AND CURRENT REFERENCE WITH VARIOUS COMPENSATED RESISTORS AND A TWO-STAGE

OPERATIONAL TRANSCONDUCTANCE AMPLIFIER (OTA)

As the temperature (T) increases, the voltages, VC, VD and VE, of nodes C, D and E, will drop because of the negative TC of VEB; meanwhile, the values of compensated resistors, Rb1, Rb2 and Rb3, are increased. Three compensated currents flow into the three BJTs, Q1, Q2, and Q3, respectively, compensating for the emitter currents, which increase according to a positive TC such that IR2 = (VEB1-VEB2)/R2 = ∆VEB/R2. This increase yields three temperature-independent currents, I1, I2 and I3. Simultaneously, voltages Vin-, Vin+ and Vref, will be compensated for to maintain constant with resistances, Ra1, Ra2 and R3, because their TCs are positive. Hence, both voltage reference and current reference will be independent of temperature. We here need to emphasize that the compensated resistances, Rb1, Rb2 and Rb3, must be very large and be fabricated with n-wells. However, the resistance R2, which is fabricated with n+-diffusion, is low and a TC of around one-third of that of n-well. Figure 3 presents the five-corner simulations of current reference Iref as a function of temperature for the proposed bandgap and current reference.

The proposed resistor-compensation circuit is analyzed mathematically. For simplicity, consider components Ra2, Rb2, R2 and Q2 in Fig. 2. Suppose that OTA is ideal and that the counterpart resistors are equal, such that Vin+ = Vin-, Ra1 = Ra2 and Rb1 = Rb2; now, I1 = I2. Therefore,

22211 RRaEBRaEB VVVVV ++=+ (2)

221121 aRaaRaRaRa RIRIVV ×=×== (3 )

21222 RRbRRbRa IIIII +=+= (4)

FIG. 3 OUTPUT CURRENT REFERENCE VERSUS TEMPERATURE

OBTAINED BY FIVE-CORNER SIMULATIONS OF THE PROPOSED BANDGAP AND CURRENT REFERENCE

where VEB1 and VEB2 are the emitter to base voltages of the bipolar transistors, Q1 and Q2; VRa1, VRa2 and VR2 are the voltage declines across Ra1, Ra2 and R2, and IRa1, IRa2 and IR2 are the currents through Ra1, Ra2 and R2, respectively. Differentiating Eqn. (4) with respect to temperature (T) yields

TI

TI

TI RRbRa

∂∂

+∂∂

=∂∂ 222 (5)

If a temperature-independent current reference IRa2 is required, ∂IRa2/∂T = 0 is set; thus,

022 =∂∂

+∂∂

TI

TI RRb (6)

where IRb2 is a negative-TC current because IRb2 = IRb1 = VEB1/Rb1, whereas IR2 is a positive-TC current because IR2 = (VEB1-VEB2)/R2 = ∆VEB/R2. Combining the first differential item, ∂IRb2/∂T with the second differential item, ∂IR2/∂T, yields a temperature-independent current IRa2. Passing through the current mirror, which has MOSFETs M1-M4, a temperature-independent current reference Iref is generated in the proposed bandgap and current reference.

The voltage reference Vref associated with the bandgap and current reference can be expressed as,

333 EBRref VRIV +×= (7)

where VEB3 is the emitter to base voltage of the BJT Q3 and IR3 is the current through resistor R3. Notably, IR3 equals I3, which is a temperature-independent current. Differentiating the above equation with respect to temperature (T) yields

TV

TRI

TIR

TV EB

RRref

∂∂

+∂∂

×+∂∂

×=∂

∂ 333

33 (8)


250

A temperature-independent bandgap reference is obtained by setting ∂Vref/∂T ≈ 0 and ∂IR3/∂T ≈ 0. Therefore,

0333 =

∂∂

+∂∂

×T

VTR

I EBR (9)

where R3 is a positive-TC resistor, whereas VEB3 is a negative-TC voltage. Properly setting the value of R3 yields a temperature-independent voltage reference, ∂Vref/∂T ≈ 0. Notably, R3 not only compensates for the temperature variation of VEB3, but also adjusts the output voltage as required. Moreover, resistors R3 and Rb3 are fabricated using an n-well. The resistance R3 exceeds Ra2, while Rb3 is less than Rb2.

Figure 4 schematically depicts the complete circuit of the proposed bandgap and current reference, which simultaneously provides temperature-independent voltage reference Vref and current reference Iref. The left-hand side of Fig 4 presents an OTA circuit with N-type input [14], which is used to ensure that the positive input Vin+ equals the negative input Vin- of OTA. Notably, the MOSFET, Mr, must be operated in the triode region as a resistor. The simulations of the two-stage telescopic OTA indicate those dc gain, bandwidth and phase margins are 61.35 dB, 9.52 MHz and 64o, respectively. If an input offset voltage of OTA, owing to asymmetries, is considered, it will introduce error in the voltage reference Vref . Thus,

( )33 2 3

2

lnref EB T OS RbRV V V m V I RR

= + × − +

(10)

where VOS is the input offset voltage of OTA, VT is the thermal voltage, and m is the base-emitter area ratio of Q2 to Q1, producing m ≈ 5/2. In this work, two methods are employed to lower the effect of VOS. One is that the OTA is a telescopic topology to minimize the offset because of symmetry and the other is that the layout of OTA incorporates common centroid and dummy in a large device.

Proposed Start-up Circuit

As shown in Fig. 2, the start-up circuit comprises three MOSFETs, Mn, Mp and MS, where the gate-source voltage VGS, of Mn is shorted to turn off Mn with a huge resistance. Restated, the gate voltage of MS, VMS,G, is half of the supply voltage VDD in the initial stage because both Mp and Mn are cut-off. As the supply voltage VDD increases slowly, VOTA,out, which is connected to the gate terminal of Mp, traces VDD because the OTA is off and the voltage difference

between source (VDD) and gate of Mp is about zero due to the Cgs of Mp, M1 and M2. When MS is turned on under the condition VDD-VMS,G ≥ |Vthp| with a threshold voltage of PMOS, |Vthp|, a small conduction current IMS flows. Then the conduction resistance between drain and source, rDS, of Mp is reduced. By comparing with the huge resistance of Mn, the gate voltage of Ms increases. Mp flows current until VDS is reduced to nearly 0 V. The voltage of VMS,G keeps VDD due to parasitic capacitance.

Finally, the operation mode of Mp is changed from saturation to the linear region and Ms will be turned off. Note that the Mn is used not only to speed-up the rise time, but also to save power without driving current. Figure 5 plots the simulated results concerning the start-up circuit over time, where the symbols , , and represent the power-supply voltage (VDD), the gate voltage of MS (VMS,G), the output voltage of OTA (VOTA,out) and the source current of Ms (IMS), respectively. Importantly, the minimum power supply, VDD,min, is approximately 1.35 V, and the difference between the power-supply voltage (VDD) and the gate voltage of MS (VMS,G), VDD – VMS,G, exceeds 0.7 V; the start time is around 43 ns, and the source current of Ms, IMS, falls to zero.

Experimental Results

The proposed bandgap and current reference was fabricated with 0.35-µm 2P4M CMOS process. The layout is carefully considered to minimize the mismatches of the resistor and that of the transistor. Additionally, resistors Ra1 and Ra2 are implemented using n-well because of its large temperature coefficient, while n+-diffusion occurs in resistor R2 with a small temperature coefficient. The temperature-dependent performance was measured over operating temperatures from 0oC to 100oC. Figures 6 and 7 plot the measured voltage reference Vref and current reference Iref, respectively, of the proposed bandgap and current reference against temperature. Figure 6 indicates that the measured voltage reference is proportional to the temperature in the range 0oC to 30oC, and is roughly constant from 30oC to 100oC. The temperature coefficient of the voltage reference is approximately 49 ppm/oC and the maximum variation of Vref is approximately 4.35 mV at a supply voltage of 3.3 V and temperatures from 0oC to 100oC. The corresponding values are 12.8 ppm/oC and 0.8 mV from 30oC to 100oC. Figure 7 reveals that the measured current reference is almost constant from 0oC to 70oC, but increases rapidly in


251

FIG. 4 COMPLETE CIRCUIT OF THE PROPOSED BANDGAP AND CURRENT REFERENCE WHICH SIMULTANEOUSLY PROVIDES

TEMPERATURE-INDEPENDENT VOLTAGE REFERENCE AND CURRENT REFERENCE

temperature range 70oC to 100oC. The temperature coefficient of the current reference is about 119.3 ppm/oC and the maximum variation of Iref is about 8.77 nA at temperatures from 0oC to 100oC. Additionally, the variations in voltage reference are measured, and plotted in Fig. 8 versus the power supply voltage from 0 V to 3.6 V. The measurements also demonstrate that output voltages from 0.871 V to 0.888 V are roughly proportional to the power supplied from 1.4 V to 3.4 V. Notably, the proposed bandgap and current reference can be started up at a supply voltage of 1.35 V, and so is suitable for operating with battery cell.

Table I presents the measurements of the proposed bandgap and current reference. Table II compares the proposed resistror-compensation bandgap and current r efer ence pr esent ed her ein w ith ot her pr ior- art curvature-compensation bandgap references. In table II, the best voltage TC of 12.8 ppm/ is superior to that of other bandgap references, except [5], and the best current TC of 119 .2 ppm/ is acceptab le by compar ing w it h r ef er ence [ 19] . Based on t he compar ison, the pr oposed bandgap and curr ent reference is suitable for use at temperatures of over 30oC. However, the averaged current reference is lower. Note that large compensated resistors, Rb1, Rb2 and Rb3, were selected herein to reduce power consumption of

(a)

(b)

FIG. 5 SIMULATED RESULTS CONCERNING START-UP CIRCUIT OVER TIME IN NS. (A) VARIATIONS OF VDD (), VMS,G () AND VOPA,OUT() IN VOLT (V). (B) SOURCE CURRENT OF MS ()

IN MICRO AMPERES (µA)


252

the chip and high positive-TC resistors, Ra1 and Ra2, were adopted to compensate for the temperature-dependent variation of Vin+ and Vin- in OTA. Figure 9 presents a die microphotograph of the proposed bandgap and current reference fabricated in a 0.35-µm CMOS process. In this chip, two capacitors, C1 and C2, are connected to power supply and voltage reference, respectively, to alleviate the unstableness.

FIG. 6 MEASURED VOLTAGE REFERENCE (V) AS A FUNCTION OF

TEMPERATURE (OC)

FIG. 7 MEASURED CURRENT REFERENCE (NA) AS A FUNCTION OF

TEMPERATURE (OC)

FIG. 8 VARIATION OF OUTPUT REFERENCE VOLTAGE (V)

AGAINST SUPPLY VOLTAGE (V) FOR THE PROPOSED BANDGAP AND CURRENT REFERENCE

FIG. 9 DIE MICROPHOTOGRAPH OF THE PROPOSED BANDGAP

REFERENCE AND CURRENT REFERENCE FABRICATED IN A 0.35-µM CMOS PROCESS

TABLE I MEASUREMENTS OF PROPOSED BANDGAP REFERENCE AND CURRENT REFERENCE

Parameters Measurements

Typical power supply (V) 3.3

Minimum power supply (V) 1.35

Averaged voltage reference (mV) (0~100) 888.1

Maximum variation of voltage reference (mV) (0~100) 4.35

Voltage temperature coefficient (ppm/)

(0~100) 49

Averaged voltage reference (mV) (30~100) 888.7

Maximum variation of voltage reference (mV)(30~100) 0.8

Voltage temperature coefficient (ppm/)

(30~100) 12.8

Voltage reference settling time (V/µs) 38.1

Averaged current reference (nA) (0~100) 735.6

Maximum variation of current reference (nA) (0~100)

8.77

Current temperature coefficient (ppm/)

(0~100) 119.2

Power dissipation (µW) 91.28

Chip area (µmµm) 237256

Conclusions

A resistor-compensation CMOS bandgap reference and current reference with a current reference of 735.6 nA and a voltage reference of 888.1 mV at a supply voltage of 3.3 V was presented. It consumes a maximum power of 91.28 µW. The voltage TC was 49


253

TABLE 2 COMPARISON AMONG THE CURVATURE-COMPENSATED BANDGAP REFERENCES

This work [17] [18] [19] [5] [3]

Technology 0.35 µm 0.35 µm 0.25 µm 0.18 µm 0.18 µm 0.13 µm

Typical VDD (V) 3.3 3.3 NA 1.0 1.1 1.2

Minimum VDD (V) 1.35 NA 0.85 NA 0.90 NA

Averaged voltage reference (0~100) 888.1 mV 1173.2 mV 238.2 mV 598.5 mV 657 mV 630 mV

Voltage TC (0~100) 49 ppm/ 47 ppm/

(trimming) 58 ppm/ 125 ppm/ 10 ~ 40

ppm/ 29 ppm/

Voltage TC (30~100) 12.8 ppm/

Averaged current reference(nA) (0~100) 735.6 nA NA NA 144 µA NA 50.2 µA

Current TC (0~100) 119 ppm/ NA NA 185 ppm/ NA 18 ppm/

Temperature Range 0~100 -75~75 -10~120 0~100 0~150 -10~100

ppm/ at temperatures from 0oC to 100oC, and 12.8 ppm/ from 30oC to 100oC. The current TC was 119.2 ppm/ from 0oC to 100oC. The measurements also reveal that a good temperature-independent voltage reference Vref is realized at high temperature and a fine temperature-independent current reference Iref is performed at low temperature. With a simplified start-up circuit, the proposed bandgap and current reference was verified to be effective in a standard 0.35-µm CMOS process. Restated, the proposed resistor-compensation CMOS bandgap and current reference, which is compensated with various high positive TC resistors, simultaneously provides both a temperature-independent voltage reference Vref and temperature-independent current reference Iref.

Furthermore, this work verifies that both n-well and n+-diffusion are suitable for developing a new resistor-compensation technique in bandgap reference or current reference. To further improve the performance of bandgap reference or current reference, the resistor-compensation technique can be utilized except high-order curvature compensation [5].

ACKNOWLEDGMENT

The authors would like to thank the National Science Council of the Republic of China, Taiwan, for financially supporting this research under Contract No. NSC 95-2221-E-027-138-MY3. The CIC is appreciated for fabricating the test chip and Ted Knoy is appreciated for his editorial assistance.

REFERENCES

[1] H. Banba, H. Shiga, A. Umezawa, T. Miyaba, T.

Tanzawa, S. Atsumi and K. Sakui, “A CMOS bandgap

voltage reference circuit with sub-1-V operation,” IEEE J.

Solid-State Circuits, vol. 34, no. 5, pp. 670–674, May 1999.

[2] G. Giustolisi, “A low-voltage low-power voltage

reference based on subthreshold MOSFETs,” IEEE J.

Solid-State Circuits, vol. 38, no. 1, pp. 151–154, Jan. 2003.

[3] D. O. Han, J. H. Kim and N. H. Kim, “Design of

bandgap reference and current reference generator with

low supply voltage,” in Proc. ICSICT’08, Oct. 2008, pp.

1733–1736.

[4] M. D. Ker and J. S. Chen, “New c urvature-

compensation technique for CMOS bandgap

reference with sub-1-V operation,” IEEE Trans. Circuits

Syst. II, Express Briefs, vol. 53, no. 8, pp. 667–671, August

2006. [5] X. Xing, Z. Wang and D. Li, “A low voltage high

precision CMOS bandgap reference,” Norchip, pp. l-4,

Nov. 2007.

[6] K. N. Leung and P. K. T. Mok, “A sub-1-V 15-ppm/C

CMOS bandgap voltage reference without requiring low threshold voltage device,” IEEE J. Solid-State

Circuits, vol. 37, pp. 526-530, April 2002.

[7] P. Malcovati and F. Maloberti, “Curvature-compensated

BiCMOS bandgap with 1 V supply voltage,” IEEE J.

Solid-State Circuits, vo1. 36, no. 7, pp. l076-1081, July 2001.

[8] X. Guan, A. Wang, A. Ishikawa, S. Tamura, Z. Wang


254

and C. Zhang, “A 3V 110uW 3.1ppm/ curvature-compensated CMOS bandgap reference,” IEEE Int.

Symp. Circuits Sys. pp. 2861-2864, 2006.

[9] K. N. Leung, P. K. T. Mok and C. Y. Leung, “A 2-V 23-

uA 5.3-ppm/oC curvature-compensated CMOS bandgap

voltage reference,” IEEE J. Solid-State Circuits, vol. 38, no.

3, pp. 561-564, March 2003.

[10] J. M. Audy, “Bandgap voltage reference circuit and

method with low TCR resistor in parallel with high TCR

and in series with low TCR portions of tail resistor,” U.S.

Patent 5291122, Mar. 1, 1994.

[11] J. Chen and B. Shi, “1 V CMOS current reference with 50

ppm/oC temperature coefficient,” Electronics Letters, vol.

39, issue 2, pp. 209-210, Jan. 2003.

[12] T. V. Cao, D. T. Wisland, T. S. Lande, F. Moradi and Y.

H. Kim, “Novel start-up circuit with enhanced power-

up characteristic for bandgap references,” in Proc. IEEE

Int. SOC Conference, Sept. 2008, pp. 123-126.

[13] B. Razavi, Design of Analog CMOS Integrated Circuit ,

McGraw Hill, 2001.

[14] P. E. Allen and D. R. Holberg, CMOS Analog Circuit

Design, 2nd Edition, Oxford University Press, 2002.

[15] D. A. Johns and K. Martin, Analog Integrated Circuit

Design, New York, John Wiley, 1997. [16] W. Wu, W. Zhiping and Z. Yongxue, “An improved

CMOS bandgap reference with self-biased cascoded

current mirrors,” in Proc. IEEE Conf. on Electron Devices

and Solid-State Circuits, Dec. 2007, pp. 945-948.

[17] J. P. M. Brito, H. Klimach and S. Bampi, “A design

methodology for matching improvement in bandgap

references,” in Proc. 8th Int. Symp. on Quality Electronic

Design (ISQED’07), March 2007, pp. 586-594.

[18] M. D. Ker, J. S. Chen and C. Y. Chu, “A CMOS bandgap

reference circuit for sub-1-V operation without using

extra low-threshold-voltage device,” IEICE Trans.

Electronic, vol. E88-C, no. 11, pp. 2150-2155, Nov. 2005.

[19] A. Bendali and Y. Audet, “A 1-V CMOS current

reference with temperature and process

compensation,” IEEE Trans. on Circuits and Systems I,

vol. 54, pp. 1424-1429, 2007.

Guo-Ming Sung received the B.S. and M.S. degrees in biomedical Engineering from the Chung-Yuan University in

1987 and 1989, respectively, and the PH.D. degree in electrical engineering from the National Taiwan University, Taipei, in 2001. In 1992, he joined the Division of Engineering and Applied Sciences, National Science Council, Taiwan, where he became an Associate Researcher in 1996. Since 2001, he has been with the Electrical Engineering

Department, National Taipei University of Technology, where he is an Associate Professor. His research interests include magnetic sensors, integrated circuits and systems for analog and digital circuits, motor control ICs, and mixed-mode ICs for XDSL.

Ying-Tzu Lai received the M.S. degree in electrical engineering from Lunghwa University of Science and Technology, Taoyuan, Taiwan, R.O.C., in 2005, and now studying Ph.D. degrees in electrical engineering from National Taipei University of Technology since 2006. Her research interests include mixed-mode integrated circuit design, analog-to-

digital converters, and switched-current delta-sigma modulator.

Chien-L in Lu received the B.S. degree from the Department of Communications Engineering, Feng Chia University in 2004, and now studying M.S. degrees in Electronic Engineering from National Taipei University of Technology since 2008. He has been a member with the National Chip Implementation Center (CIC), Taiwan,

R.O.C. His research interests include analog circuit design, RF circuit design, and analog to digital converter (ADC).


255

Transient Analysis of Three-Phase Self Excited Induction Generator Using New ApproachVivek Pahwa1, K. S. Sandhu2 1Department of Electrical and Electronics Engineering, Haryana College of Technology and Management, Kaithal, India 2Department of Electrical Engineering, National Institute of technology, Kurukshetra, Kurukshetra, India

[email protected], [email protected]

Abstract

In this paper, Matlab/Simulink based new saturation model is proposed to investigate the transient performance of a three-phase induction machine. The model as proposed is used to predict the transient performance of three-phase induction generator under different operating conditions. Closeness of simulated results with experimental results on a test machine proves the effectiveness of the proposed model.

Keywords

Modeling; Self Excited Induction Generator; Simulation; Transient Analysis

Introduction

Global environmental concerns and growing demand of isolated power plants are some of the major issues due to which self excited induction generators are getting more and more popular. Further operational and constructional advantages are some other factors which are responsible for rapid establishment of suitable self excited induction generators in contrast to conventional synchronous generators. Further in the absence of grid a wind turbine generator in self excited mode is found to be very useful for isolated and remote locations.

The self excited induction generator is essentially a three phase induction machine in which the magnetizing current is furnished by the static capacitors connected across the stator terminals [1, 2]. Whenever driven by a suitable prime-mover under favourable conditions, voltage build up occurs and power is transferred to connected load. The types of loads experienced on such isolated power plants consisting of self excited induction generators may be static/dynamic in nature. Sudden switching of such loads cause transients in the system, which are of

immense interest. Therefore transient analysis of a machine is must for design consideration and some researchers tried to investigate the dynamic performance of such generators [3-8].

[9, 10] used the well tested d-q axis based conventional model to investigate the transient behaviour of three phase induction machine. [11-14] describes the basic concept of transient modeling of the machine.

Matlab / Simulink is found to be very useful tool for modeling electrical machine and it may be used to predict the dynamic behavior of the machines. In this paper Matlab / Simulink based new saturation model is proposed to study the dynamic behaviour of three phases self excited induction generator. Effects of ‘capacitor switching’, ‘load variation’, ‘input variation’ and ‘variation in moment of inertia’ on the transient performance of self excited induction generator have been taken in the present work.

Mathematical Modeling

The voltage equation of the induction machine model in rotor reference frame is given by [11-14];

[ ] [ ] [ ]iZvG = (1)

where [ ]Gv and [ ]i represents voltage and current

matrices and are given as Tdrqrdsqs vvvv

and

Tdrqrdsqs iiii

respectively.

Impedance matrices as defined in equation (1) may be given as,


256

[ ]

+−+

−+−+

=

pLRpLpLRpL

pLpLpLRLLpLLpLR

Z

rrm

rrm

msrsssr

mrmsrss

0000

ωωωω

FIGURE 1 TWO-AXIS MODEL OF THREE-PHASE SELF EXCITED

INDUCTION GENERATOR

And, the electromagnetic torque is

)(22

3qrdsdrqsme iiiiLPT −= (2)

Equation of motion

Equation of motion used to relate the electromagnetic torque developed and load torque may be defined as;

rLe pPJTT ω)/2(=− (3)

Capacitor side equations are

[ ] ( ) [ ]cG icvp /1= (4)

A n d [ ] [ ] [ ]Lc iii += (5)

[ ] TLdLqL iii ][=

[ ] [ ]Tcdcqc iii =

Load side equation

[ ] [ ] [ ]LLLLG iRipLv += (6)

Saturation model

Saturation curve of the machine may be represented [14] in two ways as discussed below.

1) Conventional Saturation model

This curve is a graphical representation between rms values of air gap voltage and magnetization current. Therefore mathematically air gap voltage is a function of magnetizing current and is represented as:

E = f ( Im ) (7)

where

E = Per phase air gap voltage (rms).

Im = Per phase magnetizing current (rms).

This curve for test machine [Appendix- A] may be used to develop the polynomial relationships between Xm and Im [Table-1] and is used to account the saturation in the simulated model. This is the conventional way to account the effect of saturation in electrical machines and is generally adopted by most of the researchers [2, 15-17] for the transient and steady state analysis of induction machine.

2) Proposed Saturation model

This curve is a graphical representation between instantaneous values of air-gap flux linkage and magnetizing current. Therefore air-gap flux linkage is a function of magnetizing current and is represented as;

ψ = f (im) (8 )

where

ψ = Instantaneous value of flux linkage.

im = Instantaneous value of magnetizing current.

Such relationships have been used by [13, 14, and 17] in case of transformer and reactors. However none of the research persons used such representations for the analysis of induction generators. Conventional magnetization curve as shown in figure A.2 may be modified as in figures B.1 [Appendix-B], which show the variation of instantaneous values of flux linkages and magnetizing current. This is used to develop the relationship between xm and im [Table-1]. For the first time, such relationship is proposed to account the saturation in three phase induction generator.

TABLE1 RELATION BETWEEN MAGNETIZING REACTANCE AND MAGNETIZING CURRENT

Results And Discussions

Figure 2 shows the MATLAB/SIMULINK based conventional and proposed models of a three-phase self excited induction generator used for simulations.

Relationship between magnetizing reactance and magnetizing current

Due to conventional saturation curve

Xm = 0.3372 I3 m- 1.8650I2m+

9.1425 Im + 108.2482Ω

Due to proposed saturation curve

xm = 0.1205 i3 m- 0.7154i2m+

8.6888 im + 100.4563Ω


257

FIGURE 2 MATLAB/SIMULINK MODEL OF THREE-PHASE SELF

EXCITED INDUCTION GENERATOR WITH PROPOSED AND CONVENTIONAL METHODOLOGY

Figure 3 shows the comparison of simulated and experimental results on test machine [Appendix-A]. Proposed saturation model yields better simulation results in contrast to conventional saturation model. This proves the effectiveness of proposed model in contrast to conventional model. Therefore it is recommended to use this model to predict the transient behavior of a self excited induction generator.

FIGURE 3 GENERATED VOLTAGES AND LOAD CURRENTS

Figure 4 to figure 7 show the comparison of simulated results with conventional and proposed modeling on test machine in self excited generating mode to analyze the effects of following:

• Capacitor switching

• Load switching

• Change in input power

• Change in moment of inertia

Capacitor switching

Figure 4 shows the effect of capacitor switching on the voltage build of induction generator under no load operation. The machine under consideration runs as a self excited induction generator with change of capacitance at 2, 3 and 4 seconds. At these instants capacitance is varied from 40 to 30, 30 to 20 and 20 to 10 microfarads respectively. From this figure following observations may be drawn:

Proposed model results into high value of voltage in contrast to conventional model. In addition initial build up from zero to final value is dependent upon the type of modeling adopted for simulation.

With decrease in capacitance the voltage decreased and ultimately it leads to voltage collapse, irrespective of the model used for simulation purpose.

FIGURE 4 EFFECT OF CAPACITANCE SWITCHING UNDER NO LOAD OPERATION, SPEED=1500 RPM

Load switching

Figure 5 shows the effect of load switching on the transient behaviour of stator current of the test machine when load resistance is changed at 2 and 4 second. The load resistance is changed from 100 ohms to 200 ohms and 200 ohms to 400 ohms at the respective instants. Both models give the same transient response during the load switching. However initial response from zero value to final value of current found to be dependent upon the type of model used for simulation purpose.

Ia_proposed

Vb_proposed

Va_proposed

Ib_proposed

Vc_proposed

Ic_proposed

Va_conventional

Ia_conventional

Vb_conventional

Ib_conventional

Vc_conventional

Ic_conventional

In1

Out1

Out2

Out3

Out4

Out5

Out6

proposed

In1

Out1

Out2

Out3

Out4

Out5

Out6

conventional

y3r

To Workspace2

y3i

To Workspace1

Tmech

Scope3

Scope1

Mux

Mux2

Mux

Mux1

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50

200

400

Va(

V)

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50

2

4

Ia(A

)

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50

200

400

Vb(

V) simulated results with proposed model

simulated results with conv. modelexperimental results

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50

2

4

Ib(A

)

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50

200

400

Vc(

V)

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50

2

4

Ic(A

)

time(sec)

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50

100

200

300

Va(

V)

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50

100

200

300

Vb(

V)

simulated results with proposed modelsimulated results with conv. model

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50

100

200

300

Vc(

V)

time(sec)


258

FIGURE 5 EFFECT OF LOAD SWITCHING, C = 40 MICROFARADS,

SPEED = 1500 RPM

Change in input power

Figure 6 shows the effect of change of input mechanical power applied by the prime mover on the voltage build of induction generator. The machine under consideration runs as a self excited induction generator with change of input mechanical power at 1 and 3 seconds. At these instants input mechanical power is varied from 1 pu to 0.5 pu and 0.5 pu to 0.25 pu respectively. From this figure following observations may be drawn:

With decrease in input power the voltage as well as the current is decreasing simultaneously.

Initial build up from zero to final value is dependent upon the type of modeling adopted for simulation. Proposed model results into high value of voltage in contrast to conventional model.

FIGURE 6EFFECT OF CHANGE IN INPUT POWER, C = 40

MICROFARADS, SPEED = 1500 RPM

Change in moment of inertia

It is well known that moment of inertia greatly affects the transient performance of three-phase induction machine in motoring mode. Figure 7.a to figure 7.c show the simulated results to look the effects of moment of inertia on the transient performance in self excited generating mode. It is observed that:

Any change in the moment of inertia of the machine affects the voltage build up of generator, irrespective of the type of model adopted for analysis. However this effect is more pronounced in case of proposed model in contrast to conventional model.

FIGURE 7 a EFFECT OF CHANGE IN MOMENT OF INERTIA, C =

40 MICROFARADS, SPEED = 1500 RPM WITH J = 0.913 KGM2

FIGURE 7 b EFFECT OF CHANGE IN MOMENT OF INERTIA, C =


0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50

1

2

3

Ia(A

)

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50

1

2

3

Ib(A

)


0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50

1

2

3

Ic(A

)

time(sec)

0 0.5 1 1.5 2 2.5 3 3.5 40

200

400

Va(

V)

0 0.5 1 1.5 2 2.5 3 3.5 40

2

4

Ia(A

)

0 0.5 1 1.5 2 2.5 3 3.5 40

200

400

Vb(

V) simulated results with proposed model

simulated results with conv. model

0 0.5 1 1.5 2 2.5 3 3.5 40

2

4

Ib(A

)

0 0.5 1 1.5 2 2.5 3 3.5 40

200

400

Vc(

V)

0 0.5 1 1.5 2 2.5 3 3.5 40

2

4

Ic(A

)

time(sec)

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

200

400

Va(

V)

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

2

4Ia

(A)

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

200

400

Vb(

V)


0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

2

4

Ib(A

)

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

200

400

Vc(

V)

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

2

4

Ic(A

)

time(sec)

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

200

400

Va(

V)

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

2

4

Ia(A

)

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

200

400

Vb(

V)


0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

2

4

Ib(A

)

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

200

400

Vc(

V)

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

2

4

Ic(A

)

time(sec)


259

FIGURE-7 c EFFECT OF CHANGE IN MOMENT OF INERTIA, C =


Conclusion

Due to global acceptability of self excited induction generators in wind power conversion, in this paper an attempt is made to analyze the transient behaviour of such machines under capacitor and load switching. In addition simulated results were also taken to include the effects of ‘input mechanical power’ and ‘moment of inertia’ on the performance of the machine. Matlab/Simulink based new model is proposed to investigate the transient performance of a self excited induction generator. Simulated results as obtained with new proposed model are found to be closer to experimental results. This proves the effectiveness and superiority of proposed model in contrast to conventional model. Simulated results as shown in figures 4 to figure 7 may be used to draw the following observations.

Proposed model results into a delayed voltage build up in case of self excited mode for any given value of excitation capacitance, load, mechanical input and moment of inertia.

Delay in voltage build-up further increases with an increase in moment of inertia i.e. especially for large rated machines.

Nature of effects of ‘capacitor switching’, ‘load variation’, ‘input variation’ and ‘variation in moment of inertia’ is found to be same, irrespective of model used for simulation purpose. However simulated results (using proposed model) for such effects are found to be slightly different than those with conventional model.

From above observations it may be concluded that the proposed model, which is found to be superior to conventional model, results into different but reliable simulations. Therefore, it is strongly recommended to use this model for investigating the transient performance of self excited induction generator.

Appendix-A

3-hp, 3-phase, 50 Hz, 220 volts Induction Motor;

Stator Resistance, Rs= 3.35 ohms

Rotor Resistance, Rr = 1.7 ohms

Stator & Rotor Inductance,Ls =Lr = 15.44 mH

Moment of Inertia of test machine set up

With coupling,J = 0.913 kgm2

Two machines as shown in figure A.1 must run in the same direction in case fed individually. After that test machine is driven at synchronous speed with prime mover. Input current, power is recorded for different values of input voltage. Data as obtained is used to draw the magnetization curve of test machine as shown in figure A.2.

FIGURE A 1 SET UP FOR SYNCHRONOUS RUN TEST

FIGURE A 2 CONVENTIONAL MAGNETIZATION CURVE

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

200

400

Va(

V)

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

2

4

Ia(A

)

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

200

400

Vb(

V)


0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

2

4

Ib(A

)

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

200

400

Vc(

V)

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

2

4

Ic(A

)

time(sec)

0 0.5 1 1.5 2 2.5 30

50

100

150

200

250

Im(A)

E(V

)

Im1 Im2 Imn-1 Imn ImN-1 ImN

EN

EN-1

En

En-1

E2

E1


260

Appendix-B

Conventional magnetization curve as shown in figure A.2 may be converted to the proposed saturation curve with the procedure laid down by [13, 17] and it is shown in figure B.1:

FIGURE B 1 PROPOSED SATURATION CURVE.

Conversion guidelines:

1. For a sinusoidal input voltage of frequency, ω, the corresponding flux linkage is given by

ω/2 kk E=Ψ

k = 0, 1,-----n-1, n,-------, N

So, Ψ 0, Ψ 1, -----------, ΨN can be obtained.

2. For calculation of proposed value of current mi ,

∑=

−Ψ−Ψ=k

jjjjmk Gi

11)(

k = 1,-----n-1, n,-------, N

Gj is the slope of line joining points (j-1) and j, as seen from vertical axis [13, 17].

Nomenclature

sR = Stator Phase Resistance/phase

sL = Stator Self inductance/phase

mL = Mutual inductance/phase

rR = Rotor Phase Resistance/phase

rL = Rotor self inductance/phase

sω =Angular speed (radian/sec.) in synchronously rotating reference frame

J = Inertia of Motor

eT = Electrical Torque

LT = Load Torque p = Operator for differentiation

Subscripts:

q = Quadrature axis

d = Direct axis

s = Stator quantities

r = Rotor quantities

REFERENCES

[1] M. G. Say, “The Performance and Design of

Alternating Current Machines”, CBS Publishers and

Distributors, Third edition, 2002.

[2] A. Kishore, R. C. Prasad and B. M. Karan,

“MATLAB/SIMULINK based D-Q Modeling and

Dynamic Characteristics of Three Phase Self Excited

Induction Generator”, Progress in Electromagnetics

Research Symposium, Cambridge, USA, pp. 312-316,

March 26-29, 2006.

[3] B. Singh, L. Shridhar and C.S. Jha, “Transient Analysis

of Self-Excited Induction Generator supplying

Dynamic Load”, Electric Machines and Power Systems,

vol. 27, pp. 941-954, 1999.

[4] L. Wang and R.Y. Deng, “Transient Performance of an

Isolated induction Generator under Unbalanced

Excitation Capacitors”, IEEE Transactions on Energy

Conversion, vol. 14, no. 4, pp. 887-893, December 1999.

[5] S.K. Jain, J.D. Sharma and S.P. Singh, “Transient

Performance of Three-phase Self-excited Induction

Generator during Balanced and Unbalanced Faults”,

Proc. Inst. Elect. Eng., Gen., Transm., Distrib., vol. 149,

pp. 50-57, January 2002.

[6] Y. S. Wang and L. Wang, “Unbalanced Switched

Effects on Dynamic Performance of an Isolated Three-

phase Self-excited Generator”, Electric Machines Power

Systems, vol. 29, no. 4, pp. 375-387, April 2001.

0 1 2 3 4 5 60

50

100

150

200

250

300

350

im(A)

psi

im1 im2 imn-1 imn imN-1 imN

psiN psiN-1

psin

psin-1

psi2

psi1


261

[7] D. Seyoum, C. Grantham, and M.F. Rahman, “The

Dynamic Characteristics of an Isolated Induction

Generator Driven by a Wind Turbine”, IEEE

Transactions on Industry Applications, vol. 39, no. 4,

pp. 936-944, July/August 2003.

[8] F. Khater, R.D. Lorenz and D.W. Novotny, “Selection

of Flux Level in Field-Oriented Induction Machine

Controllers with Consideration of Magnetic Saturation

Effects”, IEEE Transactions on Industry Applications,

vol. 23, no. 2, pp. 276-282, March/April 1984.

[9] Julio C. Moreira and Thomas A. Lipo, “Modeling of

Saturated AC Machines including Air Gap Flux

Harmonic Components”, IEEE Transactions on

Industry Applications, vol. 28, no. 2, pp. 343-349,

March/April 1992.

[10] Paul C. Krause, O. Wasynczuk and S. D. Sudhoff,

“Analysis of Electric Machinery and Drive Systems”,

IEEE Press Series on Power Engineering, John Wiley &

Sons Inc. Publication, 2004.

[11] B. K. Bose, “Power Electronics and AC Drives”,

Pearson Prentice Hall, 2007.

[12] R. Krishna n, “Elec tr ic Motor Drives. Mode ling,

Analysis and Control”, Pearson Prentice Hall, 2007.

[13] C.M. Ong, “Dynamic Simulation of Electric

Machinery”, Prantice Hall PTR, 1998.

[14] S. N. Talukdar, J. K. Dickson, R. C. Dugan, M. J.

Sprinzen and C. J. Lenda , “On Modeling Transformer

and Reactor Saturation Characteristics for Digital and

Analog studies”, IEEE Transactions on Power

Apparatus and Systems , vol. PAS-94, no. 2, pp. 612-622,

1975.

[15] Nuh Erdogan, Humberto Henao and Richard Grisel,

“The Analysis of Saturation Effects on Transient

Behavior of Induction Machine Direct Starting”, IEEE,

pp. 975-979, 2004.

[16] O. I. Okoro, “MATLAB Simulation of Induction

Machine with Saturable Leakage and Magnetizing

Inductances”, The Pacific Journal of Science and

Technology, vo1. 5, no. 1, pp. 5-15, April 2003.

[17] S. Prusty and M.V.S. Rao, “A Direct Piecewise

Linearized Approach to Convert rms Saturation

Characteristics to Instantaneous Saturation Curve”,

IEEE Transactions on Magnetics, vol. 16, no. 1, pp.156-

160, January 1980.


262

On the Sensitivity of Principal Components Analysis Applied in Wound Rotor Induction Machines Faults Detection and Localization J. Ramahaleomiarantsoa1, N. Heraud2, E. J. R. Sambatra3, J. M. Razafimahenina4 1, 2Université de Corse, U.M.R. CNRS 6134 SPE, BP 52, 20250 Corte, France

1, 3Institut Supérieur de Technologie, BP 509, 201 Antsiranana, Madagascar

1, 4Ecole Supérieure Polytechnique, BP O, 201 Antsiranana, Madagascar

[email protected]; [email protected] ; [email protected]; [email protected]

Abstract

This paper deals with faults detection and localization of wound rotor induction machines based on principal components analysis method. Both, the localization and the detection approaches consist in analyzing detection index which is established on the latest principal components. Once the faults are detected, the affected state variables are localized by the variables reconstruction approach. The exponentially weighted moving average filter is applied to improve the faults detection quality by reducing the rate of false alarms. An accurate analytical modeling of the electrical machines is proposed and implemented on the Matlab software to obtain the state variables data of both healthy and faulted machines. Several simulation results are presented a nd analyzed.

Keywords

Principal Components Analysis; Wound Rotor Induction Machines; Faults Detection and Localization; Detection Index; Reconstruction Approach; EWMA filter

Introduction

The necessity for having reliable electric machines is more important than ever and the trend continues to increase. Now, advances in engineering and materials science allow building lighter machines while having a considerable lifetime.

Although researches and improvements have been carried out, these machines still remain the most potential failures of the stator and the rotor. The faults can be resulted by normal wear, poor design, poor assembly (misalignment), improper use or combination of these different causes. Indeed, for many years, faults detection in electrical machines has been the subject of reflection and research projects in various industrial and academic laboratories.

Several diagnosis and control methods exist and already used for the electrical machines monitoring. In this paper, Wound Rotor Induction Machines (WRIM) faults detection and localization based on Principal Components Analysis (PCA) is proposed. PCA is a statistical method used for data or state variables measurement of systems in operation to monitor their behavior.

The PCA principle consists in reducing the size of the representation space of the system [1]. In fault detection approach based on PCA, two methods are proposed [2, 3], Hotelling’s T2 statistical method and Squared Prediction Error (SPE) indicator. The T2 statistical is calculated with the “l” first principal components while the SPE indicator achieves detection with the residual space. However the two methods have limitations in faults detection [3, 4]. In case of sensors detection, the T2 indicator is not very efficient because the variations due to the failure may be masked by normal variations of the variables in the first principal components space. And when the considered systems are no longer linear, residues having high variance contain the modeling errors generated by the PCA. Thus the residue having a low variance will have less influence on the SPE quantity with respect to the residues having a higher variance, so that they correspond to the linear redundancy relations or quasi-linear. This sensitivity of the SPE indicator to the modeling errors can create many false alarms. F. Harkat, G. Mourot, and J. Ragot [2] proposed a new method for faults detection and localization based on the sums of squares of the last principal components.

mailto:1ramahaleojacques@


mailto:[email protected])


263

For our case, the method proposed by [2, 4] will be used for the WRIM faults detection and localization.

To improve the faults detection and to reduce the false alarms, the Exponentially Weighted Moving Average (EWMA) filter is used.

The first part of this article deals with the reconstruction principle of the PCA model followed by the WRIM modeling. The second part is focused on the faults detection method by the detection index (Di). The third part is focused on variables reconstruction combined with the fault indicator Di for fault location aim. The last part is reserved to the applications of the PCA approach on the WRIM. Several simulations result built with Matlab software are presented and analyzed to show the PCA method sensitivity.

Pca Method Implementation

PCA methods

The PCA method is based on a transformation of space representation of simulation data. The new space is smaller than that of the original space. This method is classified as without model methods [5] and can be seen as a full-fledged system identification method [6, 7]. Each variable to be monitored for the state of the WRIM are expressed by different units and scales. For that, it is preferable to apply a PCA on a centered and reduced measures matrix X (columns of zero means and units standard deviations) [8]. The orthogonal space defined by PCA is generated by the eigenvalues and eigenvectors of the matrix correlation R of X. These values are sorted in descending order in a diagonal matrix. The eigenvalues analysis of the correlation matrix R provides information on the number of principal components to be retained “l” for the PCA model reconstruction [1].

The orthonormal projection matrix P formed by the m eigenvectors associated with eigenvalues of the correlation matrix R is expressed as:

1 2[ , ,..., ]mP p p p= (1)

The diagonal matrix Λ of the correlation matrix R generated by the eigenvectors associated with eigenvalues λ sorted in descending order is done by:

1 2( , ,..., )mdiag λ λ λΛ = (2)

With 1 2 ... mλ λ λ≥ ≥ ≥

The orthogonal matrix which represents the projection of X in the PCA new space is T. Mathematically, the

PCA decomposes X as follows:

T XP= (3)

*N mT ∈ℜ and N is the number of carried out measures of variables to be monitored.

Determination of the structure of the model

To obtain the model structure, the components number “l” to be retained must be determined. This step is very important for PCA construction. Component number can be determined by using:

thcm

kk

l

ii

≥

∑

∑

=

= 100*

1

1

λ

λ (4)

With l<m

Where thc is an user defined threshold expressed as percentage. Now, user should retain only the components number “l” which was associated in the first term of (4). By reordering the eigenvalues, the minimum numbers of components are retained while still reaching the minimum variance threshold, [9, 10]. The vector of principal components is noted by:

1 2[ , ,..., ]mT t t t= (5)

Since the aim of PCA is to reduce the space dimension, the “l” first principal components (l << m) are the most significant and sufficient to explain the variability of a process. Therefore, the expression of centered and reduced matrix X can be written as follows:

pX X E= + (6)

The matrix Xp is the estimated principal part and the matrix E the residual part of X which represents information looses due to the X matrix dimension reduction. They are expressed as follow:

'

1

l

p i li

X PT=

= ∑ (7)

'

1

m

i li l

E PT= +

= ∑ (8)

'T is the transpose of the orthogonal matrix.

Wrim Analytical Modeling

Fig.1 shows the equivalent electrical circuit of the WRIM. Each coil, for both stator and rotor, is modelised with a resistance and an inductance connected in series configuration (Fig. 2).


264

Vj, Ij and Φj (j : A, B, C for the stator phases and a, b, c, for the rotor phases) are respectively the voltages, the electrical currents and the magnetic flux of the stator and the rotor phases, θ is the angular position of the rotor relative to the stator.

FIG. 1 EQUIVALENT ELECTRICAL CIRCUIT OF THE WRIM

FIG. 2 EQUIVALENT ELECTRICAL CIRCUIT OF THE WRIM

COILS

Rj and Lj are the resistances and the own inductances of the stator and the rotor phases. We note the voltages vector ([VS], [VR]), the currents vector ([IS], [IR]) and the flux vector ([ΦS], [ΦR]) of respectively the stator and the rotor:

[ ]A

S B

C

VV V

V

=

; [ ]A

S B

C

II I

I

=

; [ ]A

S B

C

φφ φ

φ

=

[ ]a

R b

c

VV V

V

=

; [ ]a

R b

c

II I

I

=

; [ ]a

R b

c

φφ φ

φ

=

Taking into account the above assumptions, both stator and rotor three phase voltages and currents are connected to the total magnetic flux by differential equations systems [11]. The stator and rotor voltages vectors expressions are given by:

[ ] [ ] [ ] [ ] S

S S S

dV R I

dtφ

= + (9)

[ ] [ ] [ ] [ ] R

R R R

dV R I

dtφ

= + (10)

[ ] [ ] [ ] [ ] [ ] S S S SR RL I M Iφ = + (11)

[ ] [ ] [ ] [ ] [ ] R R R RS SL I M Iφ = + (12)

[RS] and [RR] are the resistances matrix, [LS] and [LR] the own inductances matrix, and [MSR] and [MRS] the mutual inductances matrix between the stator and the rotor coils.

Equations (9) and (10) become:

[ ] [ ] [ ] [ ][ ] [ ][ ] S S SR R

S S S

d L I d M IV R I

dt dt= + + (13)

[ ] [ ] [ ] [ ][ ] [ ][ ] R R RS S

R R R

d L I d M IV R I

dt dt= + + (14)

By applying the fundamental principle of dynamics to the rotor, the mechanical motion equation is [12]:

t v em rdJ f C CdtΩ+ Ω = − (15)

ddtθ

Ω = (16)

[ ] [ ]( ) [ ] 1 * *

2t

em

d LC I I

dθ= (17)

Jt is the total inertia brought to the rotor shaft, Ω the shaft rotational speed, [I]=[IA IB IC Ia Ib Ic]’ the current vector, fv the viscous friction torque, Cem the electromagnetic torque, Cr the load torque applied to the machine, θ the angular position of the rotor with respect to the stator, and [L] the inductance matrix of the machine.

Introducing the cyclic inductances of the stator and the

rotor 32SC SL L= and 3

2RC RL L= (LS is the own inductance

of the each phase of the stator and LR is the own inductance of the each phase of the rotor), the mutual inductances between the stator and the rotor coils MSR and pole pair number p, the inductance matrix of the WRIM car be written as follow:

=

RCSRSRSR

RCSRSRSR

RCSRSRSR

SRSRSRSC

SRSRSRSC

SRSRSRSC

LfMfMfMLfMfMfM

LfMfMfMfMfMfMLfMfMfMLfMfMfML

L

000000

000000

][

123

312

231

132

213

321

(18)

With

)cos(1 θpf = (19)

)3

2cos(2πθ += pf (20)


265

)3

2cos(3πθ −= pf (21)

Differential equations system modelling

In choosing the stator and rotor currents, the shaft rotational speed and the angular position of the rotor relative to the stator as state variables, the differential equations a system modeling the WRIM is given by:

1[ ] [ ] ([ ] [ ][ ])X A U B X−= − (22)

With

'[ ] [ ]A B C a b cX I I I I I I θ= Ω ; [ ] 0 0

[ ] 0 00 0 1

t

LA J

=

;

[ ][ ]

0r

VU C

= −

; '[ ] [ ]A B C a b cV V V V V V V= ;

[ ][ ] 0 0

1 [ ][ ] [ ] 02

0 1 0

tv

d LRd

d LB I fd

θ

θ

+Ω = − −

This model of the WRIM will be used to simulate both healthy and faulted operation case of the stator and the rotor. The considered faults are resistances values increases of the stator or rotor windings due to a rise of their temperatures. The following table presents the different parameters of the WRIM:

TABLE I WRIM PARAMETERS

Symbol Parameter Value Units

Lsp Stator principal inductance 0.397 H

Lrp Rotor principal inductance 0.397 H

Lsl S tator leakage inductance 9.594 mH

Lrl Rotor leakage inductance 9.594 mH

Msr Stator-rotor mutual

inductance 0,3953 H

p Number of pole pairs 1 -

Jt Moment of inertia 0.024 Kg.m2

Rs Stator resistance 2.86 Ω

Rr Rotor resistance 2.756 Ω

fv Viscous friction coefficient 1,444 mNm/rad/s

The model has been implemented on Matlab in source

codes and allows us to obtain several matrix data for the PCA applications on WRIM faults detection and localization. The WRIM is considered faulted from t=2s and coupled to a mechanical load at time 2s. Nine state variables (m=9) have been chosen to be monitored and 10000 measures (N=10000) during 4s are considered.

Fig.3 represents the temporal variations of some state variables (Stator current, rotor current, shaft rotational speed, angular position and electromagnetic torque) showing the steady and transient states of faulted WRIM. Fig.4 shows the zoom of the same state variables variations but only the part during which the machine is in loaded.

FIG. 3 STATE VARIABLES VARIATIONS VERSUS TIME OF THE FAULTED WRIM LOADED (STEADY AND TRANSIENT

STATES)

FIG. 4 STATE VARIABLES VARIATIONS VERSUS TIME OF

THE FAULTED WRIM (STEADY STATE)

Considered faults

The considered faults are on the resistance values which increase due to a rise of their temperature. In normal operation, a resistance value variation


266

compared to its nominal value (in ambient temperature, 25°C) is faulted machine due to machine overload or coils fault [10, 13]. The resistance versus the temperature is expressed as:

)1(0 TRR ∆+= α (23)

R0 is the resistance value at T0 = 25°C, α the temperature coefficient of the resistance and ΔT the temperature variation.

Faults Detection Approach

Residues generation

For any measures vector x(k) the equation (6) becomes:

( ) ( ) ( )px k x k e k= + (24)

( )px k and ( )e k vectors represent respectively the

estimations vector and the estimation errors vector.

The principal components vector t(k) corresponds to x(k) is expressed as:

'( ) ( )t k P x k= (25)

t(k) = [tp(k) te(k)] (26)

*N lpt ∈ℜ and *( )N m l

et−∈ℜ are respectively the “ l ” first

principal components vector and the “ m-l ” last principal components vector.

With this expression (26), there is an similarity on the residue vector e(k) and the final components vector te(k).

Detection index “Di” calculation

The fault detection index is based on successive sums of squares of the last principal components [2, 4] and is defined as follows:

)()(1

2 ktkDm

imjeji ∑

+−=

= (27)

i = 1, 2,…, (m-l)

At time k, systems are malfunctioning sensing if Di is greater than a threshold index noted 2

,ατ i . α is the false

detection probability according to the Khi-2 law with “m” degree of freedom [14]. One can note a strong similarity between the detection index SPE and the detection index Di. Indeed, Di corresponds to the SPE indicator calculated by PCA model with (m-l)

principal components. Thus, this threshold detection can be calculated with an argument similar to that exposed in [4, 14].

The process is considered in default at time k if:

2,)( ατ ii kD > (28)

EWMA filter

To reduce false alarm and to improve the faults detection quality, the EWMA filter is applied at time k, and then the “jnth” filtered vector of the last principal components can be written as follow [3, 6 and 15]:

( ) (1 ) ( 1) ( )efj ej ejt k t k t kγ γ= − − + (29)

γ is the forgetting factor (0< γ <1) in taking as initial condition (0) 0ejt = and can be calculated by [6, 16]:

1 exp( 1/ )tγ = − − ∆ (30)

t∆ is the time step.

Finally, the detection index is expressed as:

)()(1

2 ktkDm

imjefjfi ∑

+−=

= (31)

And the filtered detection threshold of faults is given by [3, 10]:

2,

2, 2 αα τ

γγτ iif −

= (32)

It should be noted that many research works uses the threshold detection for sensor faults, but our case concerns faults detection of systems.

Faults Localisation Approach

When a fault is detected, it is necessary to localize or identify the involved variables. There are several methods for faults localization:

residues structuring approach,

partial PCA approach,

Calculation of variables contributions to the

detection indicator approaches.

But [45] showed the disadvantage of the methods mentioned above. Then, in this paper, faults location of WRIM state variables is based on the variables


267

reconstruction combined with the filtered detection index.

The localization of the WRIM affected state variables by the combined approach is based on two methods combination:

variables reconstruction by PCA,

detection index

The method consists in eliminating the fault influence on Di when the affected variable is reconstructed.

To localize faults on the indicator, faults directions projected in the residual space should not be collinear [8]. After the j variable number reconstruction, fault indicator in noted j

iD .

One can also use the EWMA filter to reduce the localization false alarms and to improve the localization indicator quality. If j

ifD is the filtered

detection index of the j variable number, the localization indicator can be obtained by:

2,ατ if

jifj

if

DL = (33)

The variable for which the localization indicator jifL is

less than one is the offending variable. This method can be used for the multiple faults localization in reconstructing the supposed faulted variables simultaneously.

Simulation Results and Discussion

To validate the proposed models and the efficiency of the chosen approaches, the established models have been implemented in Matlab. Nine WRIM state variables (stator three phase currents, rotor three phase currents, shaft rotational speed, angular position and electromagnetic torque) have been considered. The matrixes data of the healthy and faulted WRIM obtained by the analytical model of the machine are introduced in PCA model to show faults detection and localization performances.

For the electrical machines diagnosis, many methods are used to detect the presence or absence of faults, occurred at t=2s, and to locate the time when it began to appear on the machine windings. Two types of faults levels are considered in the system (10%, 30%). These values correspond to the rise of the stator or rotor coils resistance. We can mention the temporal

representation (Fig. 5, Fig. 6 and Fig. 7) and the signal frequency analysis. Although they have demonstrated their efficiency, the state variables representations between them also show their advantages. They can be performed without mathematical transformation (Fig. 7).

Also, the electromagnetic torque variations versus the shaft rotational speed clearly show the WRIM operation zone in the presence of faults (Fig. 7 ). After several simulations, we suggest some of these methods to highlight the place of PCA among them. In the Fig.5 and Fig.6, the figures clearly show that it is difficult to visualize changes in signals and the fault appearance time.

However, by analyzing the residues of the stator current by PCA (Fig.8, Fig.9), the fault appearance time is located on the two signals. The case of a healthy machine that has a zero residue is almost coincident with the x-axis. These observations are found in the case of the rotor current (Fig. 8). In Fig.5 and Fig.6, the presences of faults with the conventional temporal representation are no more evident than that using PCA method (Fig. 8 and Fig. 9). This one shows the residue analysis interest on PCA method. Fig.6 and Fig.9 expose the real and residue variations of the electromagnetic torque versus time. With Fig.6, the fault appearance time is not easy to locate. However, with PCA method, the variation of residues in the electromagnetic torque with and without faults can be easily detected. The real (Fig. 6) and residue (Fig. 9) variations of the electromagnetic torque versus shaft rotational speed of the machine show again that it is much more interesting to treat the state variables of the machine with PCA method to detect the presence or absence of faults on the windings. The difference between healthy and faulted operation (Fig. 9) are clearer. It is almost not found in the real variation representations (Fig. 6). Fig.8 to Fig.11 highlight the major potential benefits of state variables treatment by PCA method which easily shows faults detection and locate time of fault appearance.

With PCA method application, all representation (Fig. 10 and Fig. 11) well shows the differences between healthy and faulted WRIM. In the healthy case, residues are zero. When faults appear, the residue representations have an effective value with an absolute value greater than zero. In the Fig.10 and Fig.11, the healthy case is represented by a right line placed on the x-axis. Also, in taking into account “1” last principal component, Fig.10 shows peak variations


268

at t=2s (measure number 5000), time at which the faults are introduced. The peaks are attenuated immediately after but the signals are shifted.

With “2” last principal components, the detection is improved because after the first peaks, other peaks (in the presence of faults) are greater than those in the case of the healthy machine.

In the case of the faults detection, Fig.12 and Fig.13 show the variations of the filtered and the no filtered detection index of both “1” and “2” last principal components versus the measure number of the faulted WRIM. The threshold detection is represented on both figures. In the part where the machine is coupled to a mechanical load, the shapes exceed repeatedly the threshold index. This overrun corresponds to the presence of faults. The last principal components numbers do not have large influences on the curves shape for both filtered and no filtered detection index.

1.995 1.996 1.997 1.998 1.999 2 2.001 2.002 2.003 2.004 2.005-10

-8

-6

-4

-2

0

2

4

6

8

10

Sta

tor c

urre

nt IA

[A]

Time[s]

WRIM Healthy & Faulted: Stator current [A]

Healthy30 %10 %

FIG.5 REAL VARIATIONS VERSUS TIME OF THE STATOR

CURRENT OF THE HEALTHY AND FAULTED WRIM

-50 0 50 100 150 200 250 300 350-10

-5

0

5

10

15

20

25

30

35

40

Ele

ctro

mag

netic

torq

ue [N

m]

Shaft rotational speed [rad/s]

Healthy & Faulted WRIM

Healthy10 %30 %

FIG.6 REAL VARIATIONS OF ELECTROMAGNETIC TORQUE VERSUS THE SHAFT ROTATIONAL SPEED OF THE WRIM

However, in the case of the no filtered shape, excessive values appear. These values show a bad detection of the no filtered data compared to those of the filtered data. This behaviour can be corresponding to alarm false for some cases. Data filtering is therefore

important in faults detection process to avoid alarm false.

287.5 288 288.5 289 289.5 29010.3

10.35

10.4

10.45

10.5

Elec

trom

agne

tic to

rque

[Nm

]

Shaft rotationnal speed [rad/s]

10%30%

Healthy

FIG. 7 REAL VARIATIONS OF ELECTROMAGNETIC TORQUE

VERSUS THE SHAFT ROTATIONAL SPEED OF THE WRIM

1.99 1.992 1.994 1.996 1.998 2 2.002 2.004 2.006 2.008-1.5

-1

-0.5

0

0.5

1

Phas

e "a

" R

otor

cur

rent

Time [s]

Residues:Healthy & Faulted WRIM

Healthy 10% 30%

Early faulted

FIG.8 EARLY FAULTED IN VARIATIONS OF THE ROTOR CURRENT RESIDUES VERSUS OF THE HEALTHY AND

FAULTED WRIM

-0.03 -0.02 -0.01 0 0.01 0.02 0.03 0.04-0.2

-0.15

-0.1

-0.05

0

0.05

0.1

0.15

0.2

0.25

Elec

trom

agne

tic to

rque

Shaft rotational speed

Residues:Healthy & Faulted WRIM

30%

10%

Healthy

FIG. 9 VARIATIONS OF ELECTROMAGNETIC TORQUE

RESIDUES VERSUS THE SHAFT ROTATIONAL SPEED RESIDUES OF THE WRIM

Fig.14 and Fig.15 represent respectively the no filtered and the filtered localization index versus the WRIM state variables. The threshold of the localization index is represented on both figures. All variables having a localization index founding below the threshold variation are the affected variables.


269

2000 3000 4000 5000 6000 7000 8000

0

0.5

1

1.5

2

2.5x 10

-3 Detection index of 1 last principal component

Measure number

Healthy

30%10%

FIG. 10 DETECTION INDEX OF “1” LAST PRINCIPAL

COMPONENT VARIATIONS VERSUS THE MEASURE NUMBER OF THE WRIM

4000 4200 4400 4600 4800 5000 5200 5400 5600 5800 6000-1

0

1

2

3

4

5

6

7

8

9

x 10-3 Detection index of 2 last principal components

Measure number

30%

10%

Healthy

FIG. 11 DETECTION INDEX OF “2” LAST PRINCIPAL

COMPONENTS VARIATIONS VERSUS THE MEASURE NUMBER OF THE WRIM

0 1000 2000 3000 4000 5000 6000 7000 8000 9000 100000

0.2

0.4

0.6

0.8

1

1.2

1.4x 10

-12 Detection index of 1 last principal component

Measure number

No filtered DiFiltered DiThershold index

Thershold index

False alarm

FIG. 12 FILTERED AND NO FILTERED DETECTION INDEX OF

“1” LAST PRINCIPAL COMPONENT VARIATIONS VERSUS MEASURE NUMBER OF FAULTED WRIM

In the case of the no filtered localization index, only variable “4”and “6”correponding to the phase “a” and phase “c” rotor currents are not affected. In the filtered localization index case, stator phase “a” and phase “b” are not affected by faults. This last better reflects the WRIM behaviour in the case of the considered fault type. As in the case of the fault detection approach, the use of filter is necessary for faults localization.

0 1000 2000 3000 4000 5000 6000 7000 8000 9000 100000

0.5

1

1.5x 10

-12 Dectetion index of 2 last principal components

Measure number

No filtered Di Filtered DiThershold index

Thershold index

False alarm

FIG. 13 FILTERED AND NO FILTERED DETECTION INDEX OF

“2” LAST PRINCIPAL COMPONENT VARIATIONS VERSUS MEASURE NUMBER OF FAULTED WRIM

1 2 3 4 5 6 7 8 90

0.2

0.4

0.6

0.8

1

1.2

1.4

WRIM state variable

Non f

ilter

ed lo

caliz

ation

inde

x

State variableThreshold localization index

FIG. 14 NO FILTERED LOCALIZATION INDEX VARIATION

VERSUS THE STATE VARIABLES

1 2 3 4 5 6 7 8 90

0.2

0.4

0.6

0.8

1

1.2

WRIM state variable

Filte

red

loca

lizat

ion

inde

x

State variableThreshold localization index

FIG. 15 FILTERED LOCALIZATION INDEX VARIATION VERSUS

THE STATE VARIABLES

Conclusion

PCA method based on residues analysis has been established and applied on WRIM diagnosis. In the case of temporal variation and without PCA, the electromagnetic torque and the shaft rotational speed are the more affected by the considered fault type. An accurate analytical model of the machine has been proposed and simulated to perform the healthy and


270

faulted data for PCA approach need.

WRIM faults detection and localization approaches based on PCA method are proposed. For that, an accurate analytical modeling of the WRIM has been carried out. The established models are implemented in Matlab. Nine state variables of the machine have been considered. Simulation results show the efficiency of the detection and localization based on respectively the detection index and localization index. The use of EWMA filter on both detection and localization has helped to avoid some false alarm. Also, filtered localization index better show the affected variables.

ACKNOWLEDGMENT

This research was supported by MADES/SCAC Madagascar project. Authors are grateful to french cooperation for technical and financial support.

REFERENCES

[1] Y. Tharault, G. Mourot, J. Ragot, and D. Maquin, “Fault

detection and isolation with robust principal

component analysis,” International Journal of Applied

Mathematics and Computer Science, ED 11, vol.4, pp. 429-

442, 2008.

[2] F. Harkat, G. Mourot, and J. Ragot, “Différentes

méthodes de localisation de défauts basées sur les

dernières composantes principales,” Conférence

Internationale Francophone d’Automatique, CIFA’02,

Nantes, France, 2002.

[3] J. Mina, C. Verde, “Fault detection for large scale

systems using dynamic principal components analysis

with adaptation,” International Journal of Computers,

Communication & Control, vol. II, N°2, pp. 185-194, 2007.

[4] J. F. Ramahaleomiarantsoa, N. Héraud, E.J.R. Sambatra

and J.M. Razafimahenina, “Fault detection of a wound

rotor induction motor by principal components

analysis,” International Conference on Production, ICPR,

Stuttgart Germany, 2011.

[5] L. Lui, “Robust fault detection and diagnosis for

permanent magnet synchronous motors,” PhD

dissertation, College of Engineering, The Florida State

University, USA, 2006.

[6] S. Borquet and O. Léonard, “Coupling principal

component analysis and Kalman filtering algorithms

for on-line aircraft engine diagnostics,” Control

Engineering Practice, vol.17, pp. 494-502, 2009.

[7] B. Huang, “Process identification based on last

principal component analysis,” Journal of Process Control,

vol.11, pp.19-33, 2001.

[8] J. Karhunen, “Robust PCA methods for complete and

missing data,” Aalto University School of Science1,

Dept. of Information and Computer Science, Espoo,

Finland 2011.

[9] G.R. Halligan, “Fault detection and prediction with

application to rotating machinery,” PhD, Missouri

University of Science and Technology, 2009.

[10] J.F. Ramahaleomiarantsoa, E.J.R. Sambatra, N. Héraud,

and J.M. Razafimahenina, Performances of the PCA

method in electrical machines diagnosis using Matlab,

INTECH, MATLAB / Book 1, ISBN 979-953-307-774-0.

[11] M. Wieczorek, E. Rosołowski, “Modelling of induction

motor for simulation of internal faults,” Modern Electric

Power Systems, MEPS'10, Wroclaw, Poland, p. 29, 2010.

[12] A. Stefani, “Induction Motor Diagnosis in Variable

Speed Drives,” PhD in Electrical Engineering Final

Dissertation, University of Bologna, 2010.

[13] J.F. Ramahaleomiarantsoa, N. Héraud, E.J.R. Sambatra

and J.M. Razafimahenina, “Principal Components

Analysis Method Application in Electrical Machines

Diagnosis,” International Conference on Informatics in

Control, Automation and Robotics, ICINCO, Pays Bas,

2011.

[14] A. Benaicha, M. Guerfel, N. Bouguila, and K.

Benothman, “New PCA-based methodology for sensor

fault detection and localization,” International

Conference of Modeling Simulation, MOSIM’10,

Hammamel, Tunisia, 2010.

[15] G. Spitzlsperger “Fault detection for a via Etch process

using adaptive multivariate methods” IEEE,

Transaction on semiconductor manufacturing, vol. 18,

N° 4, pp. 528-533, 2005.

[16] J. Wang, S. J. Qin, “EWMA Kalmar Filter and Recursive

Least Squares - relationships and modifications,”

TWMCC, Spring Meeting, Austin, 2003.


271

Ramahaleomiarantsoa Jacques was born in November 1966 in Antananarivo Madagascar. He has obtained the Dipl. of engineering in electrical engineering power transmission option in 1995 at the Polytechnic School of Antsiranana (ESPA) Madagascar. PhD student at the

University of Corsica, France and the ESPA. Research professor at the ESPA and at the High Technology Institute of Antsiranana (IST D). His research focuses on fault diagnosis system and rural e lectrification based on renewable energy.

Heraud Nicolas was born in France on september 15, 1962. He received his Ph.D. degree in Automatic and Electrical Engineering from Institut National Polytechnique de Lorraine in 1991. Since 1992, he teaches at the University of Corse as professor and he is at the CNRS (UMR 6134). His fie ld of

interest includes data reconciliation and process diagnosis on renewable energy systems.

Sambatra Eric Jean Roy was born in Antsirabe, Madagascar, on December 11, 1975. He received his Ph.D. degree in Electrical Engineering from the Electrical and Automatic Research Team of Le Havre University (GREAH) in 2005. Since 2009, he teaches electrical engineering and renewable energy

systems at the IST-D (Institut Supérieur de Technologie) and ESPA (Ecole Supérieure Polytechnique), Antsiranana, Madagascar. His research interests are renewable energy systems, e lectrical machines and diagnosis.

Razafimahenina Jean Marie was born in Fianarantsoa in August 1950; he obtained Dipl. of engineering electricity and power electronics in 1979 from the Polytechnic School of Antsiranana. He graduated Comprehensive Study (DEA) in power electronics at the ESPA in 1981, PhD in photovoltaic energy in

1986 and doctorate in electrical networks in 2005. He is a Full Professor at the ESPA and Higher Institute of Technology of Antsiranana, Madagascar.


272

Evaluation of the Quality of Service Parameters for Routing Protocols in Ad-Hoc Networks Zeyad Ghaleb Al-Mekhlafi1, Rosilah Hassan2, Zurina Mohd Hanapi3 1.3Universiti Putra Malaysia (UPM) 43400 UPM Serdang, Selangor, Malaysia

2Universiti Kebangsaan Malaysia, 43600 UKM Bangi, Selangor, Malaysia

[email protected]; [email protected]; [email protected]

Abstract

Recently, many researchers have focused on the Ad-Hoc networks especially the routing protocols which include reactive and proactive routing protocols. The ultimate goal of routing protocols is forwarding data packet from the source to the destination. Consequently, several proactive routing protocols, such as routing information protocol (RIP), and reactive routing protocols, such as Dynamic Source Routing (DSR), are based on exploring, maintenance, and recuperating the route path. The likely problem in the Ad-Hoc networks is how to establish the best routing protocol that assures the requirements of the application concerning about some criteria. This work presents the evaluation of RIP and DSR utilizing the QualNet simulation. Furthermore, the achievement of these routing protocols was assessed based on the throughput, average jitter, average end-to-end delay, and energy consumption metrics. This paper demonstrates that the RIP has superior evaluation performance as compared to DSR in two different scenarios (effect of the number of nodes and effect of packet size).

Keywords

Routing Protocols; Average Jitter; Average End-to-End Delay; Throughput; Energy Consumption

Introduction

The new revolutions in wireless technology have led to the emergence of a new wireless system which is called Ad-Hoc Network. Ad-Hoc Network is a kind of wireless system which allows direct communication with each other. In Ad-Hoc network, each node plays a dual role; a router and a host in the sense at the same time. The process of sending and receiving data packages is controlled by getting some information regarding the surrounding network and dealing with algorithm. This combination between these functions is known as a routing protocol.

A number of studies have recently gained attention in using the routing protocols, particularly, proactive routing

protocol and reactive routing protocol [1, 2]. Proactive routing protocols are those protocols which carry out the function of keeping track of routes for all the destinations in the Ad-Hoc networks. They are supported to be available in the form of tables. Furthermore, proactive routing protocol periodically exchange routing information in the whole network and maintains routes between different nodes dynamically. They have low latency and high overhead, and the routes are reliable. These protocols cannot scale well with the increase in network size. It is stated that one advantage of applying such kinds of protocols is that they facilitate communication to undergo minimal initial delay in the application procedure. However, their disadvantage is represented by the fact that they require additional control traffic to constantly update the entries of the stale route. On the other hand, reactive routing protocols attempt to identify a path to the destination only when a packet of data sent to the destination is received by the network protocol. This is one advantage of such kind of protocols as the degree of uncertainty in the node position is found to be high. They have also proved to be more suitable and more distinguished by their better performance in Ad-Hoc networks. However, taking more time to find a route and requiring more flooding which results into clogging the network are among the disadvantages of such protocols.

Therefore, the arrangement of forwarding data packet from the source to destination is the ultimate aim by utilizing routing protocols. The differences between these protocols are due to the differences in the searching, maintenance and recovering the route path. The decision of choosing the best routing protocol should take into account some considerations such as mobility of nodes, packet size, cost of path, application type, number of nodes, type of traffic, and Quality of Services (QoS).


273

On the whole, QoS shore up in wireless is an extremely demanding issue because of their dynamic character [3, 4]. Diverse techniques, as of physical layer capable of application layer, have been wished-for to supply QoS shore up in wireless Ad-Hoc networking surroundings [5]. Recently, a cross-layer design move toward in QoS conditioning in wireless networks has gained more research interest [6, 7].

Consequently, this paper focuses on the most important factors, namely end-to-end delay, average jitter, throughput and energy consumption. The end-to-end delay is important for the Ad-Hoc networks due to the fact that some of the real-time applications are very sensitive to the delay which means that the data packet sent from the source node should be delivered to the final target node within a specific period of time without any delay. Therefore, the routing protocol will be selected based on the shortest path from the source node to the destination node. The average jitter assesses the variability over time of the packet latency across a network which associated with the delay. The network with constant delay has no jitter. Therefore, the routing protocol that satisfies the constant delay without any variation during the time will be more suitable to be selected for data routing. Moreover, the significance of throughput come from the needs to deliver the more messages to destination nodes during a specific period of time which means that the routing protocols should use some mechanisms to avoid the congestion in some paths which are more frequently used to prevent the packet drops during the data routing. Hence, the reactive routing will be getting a better chance as compared to the proactive routing, to be chosen as it can find alternative paths to be used rather than the congested one. Another mechanism to increase the throughput of routing protocols, in order to be chosen, is how to deal with the failures of the paths during the data delivery; meaning that if the current path used no more available either by the node failure or moving from the current position, the routing which deals with this issue will be more preferred by the user. Beside these, energy consumption is an important factor especially in mobile Ad-Hoc networks which has restricted energy. Therefore, the routing protocol should consider this factor by chosen the paths that consume small energy to extend the lifetime of the node and give the chance to the connectivity of the network to be longer. Moreover, the nodes of paths which routed the data packets will deplete their energy very fast and run-out their batteries. Therefore, the routing protocol must look for new paths to avoid using the same path repeatedly and consuming much energy. Again, the reactive

protocols will be more preferred because of their on-demand property.

Related Works

In [8], an Ad-Hoc routing protocol, namely Ad-Hoc On demand Distance Vector (AODV) has been evaluated. According to this model, the performance of AODV in homogeneous Ad-Hoc was better than heterogeneous one. A performance analysis of proactive and reactive routing protocols for Ad-Hoc networks Dynamic Destination-Sequenced Distance Vector (DSDV), AODV and Dynamic Source Routing (DSR) showed that the performance of AODV was better in dense environment except packet loss [9]. Moreover, it was found that both DSR and AODV performed well, and they proved to be better than DSDV. However, it is not clear which protocol is the best for all scenarios, even though there are rapid growth and development in the field of Ad-Hoc network. A comparison of the parameters of routing protocols between these previous studies is shown in table 1.

TABLE 1 COMPARISON OF THE PARAMETERS OF ROUTING PROTOCOLS BETWEEN PREVIOUS STUDIES

Parameter (Tyagi & Chauhan, 2010)

(Ismail & Hassan, 2010)

Number of nodes

10-200 5,7

Simulation time 1200 sec(20Min) 3000 s

Simulation area 800Х1200 m 500Х500 m, 1000Х1000 m, 1500Х1500 m, 2000Х2000 m,

2500Х2500 m.

Routing protocols

DSDV, AODV, DSR AODV

Transmission range

250 m 250 m

Packet size 512 bytes 100,200,300,400,500,600,700,800,900 and 1000 bytes

MAC protocol 802.11 802.11

Mobility type Random way point Random way point

Type of traffic CBR CBR

Packet rate 54 Mps 54 Mps

Speed (10-100) m/s 2 Mps

Program simulation

NS-2 OMNeT++

A comparative review study on reactive and proactive routing protocols in MANETs provided information about several routing schemes proposed for Ad-Hoc networks [10]. These schemes were classified according to the routing strategy (i.e., Proactive and Reactive). It is


274

shown that each protocol has definite advantages and disadvantages and is well studied for certain situations. Despite of the rapid growth in the field of Ad-Hoc networks, many challenges still exist and need more attention and consideration from researchers so that it is possible for such networks to be used more widely within the next few years. Recently, we have evaluated the routing information protocol and dynamic source routing [11]. According to this model, Routing Information Protocol (RIP) was found to be better as compared to Dynamic Source Routing (DSR).

Performance evaluation of AODV, DSDV, and DSR Routing Protocol in Grid Environment was described in a previous study [12]. According to this model, the AODV, DSR, and DSDV perform very well when the mobility is high. However, simulation results showed that the traditional routing protocols like DSR have a dramatic decrease in performance when the mobility is high. In [13], the performance of routing protocols in mobile Ad-Hoc network was compared for DSDV, AODV, and DSR and showed that DSR outperforms AODV. The DSR has less routing overhead when nodes have high mobility considering the throughput, end-to-end delay and packet delivery ratio metrics while DSDV produces low end-to-end delay compared to AODV and DSR. In [14], the evaluation four Ad-Hoc network protocols (AODV, DSDV, DSR and TORA) in diverse network scales taking into contemplation the mobility factor. Based on this model, the throughput and energy consumption in tiny size networks did not disclose any momentous differences. On the other hand, for medium and huge Ad-Hoc networks the TORA concert proved to be incompetent in this research. Above all, the concert of AODV, DSDV and DSR in tiny size networks was equivalent. Other than in medium and large size networks, the AODV and DSR formed good results and the concert of AODV in terms of throughput is good in all the scenarios that have been investigated.

Thus, our work in this present study is to use the more widely used traditional mobility models and traffic sources to create observations based on more standardized methodology that can be used to evaluate which protocol, proactive routing protocol (RIP) or reactive routing protocol (DSR), is more stabile for Ad-Hoc networks based on some criteria in QualNet simulation.

Ad-Hoc Routing Protocols

The routing protocol resolves the path of a packet from the source to the destination. To forward a packet, the

network protocol requires knowing the next node in the path and the outgoing interface on which to send the packet [15]. A routing protocol computes routing information such as homogeneous and heterogeneous networks [8, 16]. Overall, routing protocols can be classified into two categories: proactive (table driven) routing protocols and reactive (on-demand) routing protocols. Popular proactive routing protocols are (DSDV) [17], Open Shortest Path First (OSPF) [18, 19], and RIP [20], whereas reactive routing protocols include DSR [21] and AODV [22].

Routing Information Protocol

RIP is a routing protocol which is dynamic as OSPF, but it is widely used in both local and wide area networks. It is classified as an Interior Gateway Protocol (IGP) which makes a use of the distance-vector routing algorithm as proposed in 1988 [23]. Since then, RIP Version 1 has been extended and updated to RIP Version 2 in 1998 [20]. It is indicated that both RIP versions are still being used today, but they have been technically supported by more advanced techniques such as OSPF and Open Systems Interconnection (OSI) protocol; Intermediate System to Intermediate System (IS-IS). Moreover, RIP has been updated to IPv6 network which is known as a standard RIP next generation (RIPng).

One of the advantages of employing RIP is that it is simple to understand and easy to configure as it is capable of being supported by all routers, support load balancing, and in general, it is free from loop. However, among the disadvantages, RIP is not efficient, slow when it is used in large networks due to its configuration, supports equal-cost load balancing, its congestion raises a problem and its scalability is limited since it is only measured as 15 hop maximum.

Dynamic Source Routing

Dynamic Source Routing (DSR) is defined by Johnson and Maltz [24] as a routing protocol which is still on demand and in which the sender of data can determine exactly the required sequence of nodes to propagate a packet. This packet header includes a number of intermediate nodes for routing. Each node works to maintain the route cache which cashes the source route being learned. It is stated that “Route Discovery” and “Route Maintenance” are the two main components of DSR which work together to determine and maintain routes to random destinations. The purpose of designing such protocol is to make restrictions to the large consumption of bandwidth caused by control packets in Ad-Hoc wireless networks. This process is done by deleting the messages of the


275

periodic updates required which usually appears in the table-driven approach [25].

The possibility of establishing a route when necessary makes the sender to be able to choose and control routes by reducing the load of data and including routing which is free from loop containing unidirectional links in networks is all the main advantages of DSR. However, DSR may lead to significant overheads because the source route has to be included with each packet. It uses cashing excessively and lacks mechanisms by which it can detect the freshness of the routes which causes delay and reduction; hence, the route mechanism for maintenance is unable to repair a broken link locally. Therefore, this makes the delay of the connection setup higher than that found in table-driven protocols [26].

Metrics for Evaluation

Corson and Macker showed that the evaluation metrics are possible to be made a use of in evaluating the quantitatively Mobile Ad-Hoc Network (MANET) routing protocols [27]. Such quantitative measurement is useful as a prerequisite for assessing or evaluating the performance of network or even to compare the performance using different routing protocols.

Materials and Methods

Simulation Tools

The objective of this QualNet Version 5 simulation is to evaluate the proactive routing protocol and reactive routing protocol in Ad-Hoc networks in two scenarios. In a previous study [11], the effect of the number of nodes was evaluated. Beside this effect, the current study also covered the effects of packet size. It has five experiences with different number of nodes for scenario I (effects the number of nodes), and seven experiences with different packet size for scenario II (effects of packet size). The evaluation metrics used are throughput, end-to-end delay, average jitter, and energy consumption.

a. Average End-To-End Delay

This refers to the interval taking place between the data packet generation time and the time of the arrival of the last bit to the destination i.e. the average amount of time taken by a packet to move from source to destination. The process includes all possible delays which happen due to buffering during route discovery latency, queuing at the interface queue, retransmission delays at the Media Access Control (MAC) and propagation and transfer times [9].

b. Average Jitter

Average Jitter is known as the time variation measured between the arrival of the packets due to the congestion of the network, the drift in timing, or changing of the route [2].

c. Throughput

Throughput is the number of delivered packet per unit of time [28].

d. Energy Consumption

It is defined as the amount of energy consumed in a process or system, or by an organization or society. It is the summation of the idle mode, transmit mode, and receive mode [29].

Simulation Environments

In this paper, the QualNet simulation was implemented; 802.11 MAC [30]. The parameters in the simulation such as number of nodes, time of simulation, packet size, and type of traffic were summarized in Table 2.

TABLE 2 PARAMETERS SETUP

Parameter Scenario I Scenario II

Number of nodes 50,90,130,170,210 7

Simulation Time 1200sec(20Min) 3000s

Simulation area 800Х1200m 500Х500m

Routing protocols RIP and DSR RIP and DSR

Transmission Power 25dBm 25dBm

Transmit Power Consumption

100mW 100mW

Receive Power Consumption

130mW 130mW

Idle Power Consumption 120mW 120mW

Transmission range 270m 270m

Transmission Power 25.0 25.0

Item Size 512bytes 100,200,300,400,500,600 and700 Bytes

PHY 802.11b 802.11b

Type of traffic CBR CBR

Data Rate 11Mbps 11Mps

Speed (10-100) m/s (10-100) m/s

The number of nodes ranges from 50 to 210 nodes which divided into 50, 90, 130, 170, and 210 and the packet size range from 100 bytes to 700 bytes which divided into 100, 200, 300, 400, 500, 600, and 700 bytes. Five reasons

http://www.businessdictionary.com/definition/process.html


276

experiences with different number of nodes and seven reasons experiences with different packet size were implemented in this work.

Evaluation of Results

Results are obtained after the experiments have been conducted. The present paper aims to demonstrate the evaluation performance of each routing protocol with respect to the effects of the number of nodes and effects of packet size. The evaluation metrics considered for average jitter, end-to-end delay, throughput, and energy consumption. The tests highlight the evaluation performance of RIP and DSR in Ad-Hoc network.

Scenario I

Average End-To-End Delay

Data set of the effects of the number of nodes by QualNet simulation of Average End-to-End Delay (scenario I) is shown in Table 3.

TABLE 3 DATA SET OF AVERAGE END-TO-END DELAY

Scenario I

Average End-to-End Delay(s)

No of Nodes DSR RIP

50 0.079186 0.058514

90 0.197886 0.069717

130 0.207281 0.052935

170 0.063845 0.03455

210 0.191009 0.04776

FIG. 1 AVERAGE END-TO-END DELAY BETWEEN RIP AND DSR IN

SCENARIO I

Figure 1 shows the influence of the number of nodes on network average end-to-end delay for two routing protocols. The average end-to-end delay values increased according to the number of nodes for DSR. The maximum average end-to-end delay gained simulation with 130 numbers of nodes from DSR and the minimum average end-to-end delay gained from simulation 170 numbers of nodes from DSR. The increase average end-to-end delay values the increase and the decrease according to the number of nodes for RIP. The maximum average end-to-end delay gained simulation with 90 numbers of nodes from RIP and the minimum average end-to-end delay gained from simulation 170 numbers of nodes from RIP. From the graph, it is clear that RIP out performs DSR for scenario I or II of varying pause time, varying simulation time, varying speed and varying number of nodes. In case of DSR, delay time increased sharply with increasing number of nodes. However, a sharp decrease was noticed when the number of nodes is 170. On the other hand, RIP increased and then decreased with increasing number of nodes. It is important to note that RIP gave a low end-to-end delay as compared to DSR.

Throughput

Data set for the effects of the number of nodes by QualNet simulation of Throughput (scenario I) is demonstrated in Table 4.

TABLE 4 DATA SET OF THROUGHPUT

Scenario I

Throughput (bits/s)

No of Nodes DSR RIP

50 2312 2320

90 3 2301.75

130 6 1532.33

170 14 2285

210 6 2343.25

Figure 2 shows the influence of the number of nodes on network throughput for two routing protocols (RIP and DSR). The throughput values increased according to the number of nodes for RIP while in DSR it first increased when the number of nodes rose to 50 after which it starts to decrease sharply with increasing number of nodes. The maximum throughput was gained from simulation with 210 nodes for RIP and the minimum throughput was


277

gained from simulation with 130 nodes. The maximum throughput was gained from simulation with 50 nodes from DSR and the minimum throughput has gained from simulation with (90,130,170,210) numbers of nodes. RIP have higher throughput value compared to DSR.

FIG. 2 THROUGHPUTS BETWEEN RIP AND DSR IN SCENARIO I

Average Jitter

Data set for the effects of the number of nodes by QualNet simulation of Average Jitter (scenario I) is shown in Table 5.

TABLE 5 DATA SET OF AVERAGE JITTER

Scenario I

Average Jitter (s)

No of Nodes DSR RIP

50 0.0365204 0.015466

90 0 0.036365

130 0.0248375 0.018677

170 0.0143463 0.000938

210 0.0224834 0.01431

The two kinds of routing protocols have different jitter with the increased number of nodes, as shown in Figure 3. Overall, RIP showed a better jitter than DSR when the number of nodes is greater than 50 while DSR showed the better jitter than RIP, when the number of nodes is 90 but when the number of nodes is above 90, the RIP gave a better jitter than DSR.

FIG. 3 AVERAGE JITTER BETWEEN RIP AND DSR IN SCENARIO I

Energy Consumption

In energy consumption, the result was calculated by collecting Idle mode + Transmit mode + Receive mode. The energy consumption was represented in two tables: Table 6 for the Idle mode, Transmit mode and Receive mode and Table 7 for the collected energy consumption (Idle mode + Transmit mode + Receive mode).

TABLE 6 ATA SET FOR ENERGY CONSUMPTION FOR IDLE MODE, TRANSMIT MODE AND RECEIVE MODE

DSR

No of Nodes

50 90 130 170 210

Receive mode

0.066248 26.4599 33.6272 36.8386 29.895

Transmit mode

0.020879 0.008001 0.013737 0.007616 0.007551

Idle mode 39.9387 15.5754 8.95939 5.99503 12.4046

RIP

No of Nodes

50 90 130 170 210

Receive mode

2.25513 2.35914 2.96544 3.51517 4.11252

Transmit mode

0.21928 0.398631 0.577919 0.718834 1.01791

Idle mode 37.9164 37.8188 37.2575 36.7488 36.1947


278

TABLE 7 DATA SET OF THE COLLECTED ENERGY CONSUMPTION

Scenario I

Energy Consumption

No of Nodes DSR RIP

50 40.02583 40.39081

90 42.0433 40.57657

130 42.60033 40.80086

170 42.84125 40.9828

210 42.30715 41.32513

FIG. 4 ENERGY CONSUMPTION BETWEEN RIP AND DSR IN

SCENARIO I

The energy consumption for the two routing protocols increased at the beginning of this work, as shown in Figure 4. DSR has a longer consumption than RIP. Therefore, RIP has the better energy consumption than DSR except when the number of nodes is 50 nodes.

Scenario Ii

Average End-to-End Delay

Data set of the effects of packet size by QualNet simulation of average End-to-End Delay (scenario II) is presented in Table 8.

Figure 5 shows that the average end-to-end delay for two routing protocols decreased; except when the packet size of DSR was higher than 100 bytes. Thus, DSR has longer delay than RIP and RIP exhibits shorter delay than DSR.

TABLE 8 DATA SET OF AVERAGE END-TO-END DELAY

Scenario II

End-to-End Delay(s)

Packet Size DSR RIP

100 6.5376 0.00089

200 6.54139 0.00085

300 6.43877 0.000777

400 6.73125 0.000939

500 6.06969 0.000761

600 6.41203 0.000566

700 6.81644 0.000714

Fig. 5 Average End to End Delay between RIP and DSR in scenario II.

Throughput

Data set of the effects of packet size by QualNet simulation of Throughput (scenario II) is shown in Table 9.

TABLE 9 DATA SET OF THROUGHPUT

Scenario II

Throughput (bits/s)

Packet Size DSR RIP

100 6.5376 0.00089

200 6.54139 0.00085

300 6.43877 0.000777

400 6.73125 0.000939

500 6.06969 0.000761

600 6.41203 0.000566

700 6.81644 0.000714


279

FIG. 6 THROUGHPUTS BETWEEN RIP AND DSR IN SCENARIO II

Figure 6 shows the influence of the packet size on the network throughput for two routing protocols. Overall, the throughput value increased with the packet size for the two routing protocols. The maximum throughput gained from simulation with 700 bytes packet size, while the minimum throughput gained from simulation with 100 bytes packet size. On the other hand, DSR has the maximum throughput values according to increase packet size compared to RIP. Therefore, the DSR has better throughput than RIP.

Average Jitter

Data set of the effects of packet size by QualNet simulation of Average Jitter (scenario II) is presented in Table 10.

TABLE 10 DATA SET OF AVERAGE JITTER

Scenario II

Average Jitter (s)

Packet Size DSR RIP

100 0.956555 0.001107

200 1.03527 0.000909

300 0.997965 0.000897

400 1.04567 0.001143

500 1.03995 0.000736

600 1.04009 0.000409

700 1.05922 0.000677

FIG. 7 AVERAGE JITTER BETWEEN RIP AND DSR IN SCENARIO II

The two kinds of routing protocols have different jitter with increased packet size (Fig 7). In general, RIP had better jitter than DSR while DSR showed longer delay than RIP. Thus, RIP showed the best evaluation performance.

Energy Consumption

There are two tables to show the energy consumption: table 11 for the Idle mode, Transmit mode and Receive mode while table 12 was for the collected result (Idle mode + Transmit mode + Receive mode).

TABLE 11 DATA SET FOR ENERGY CONSUMPTION OF IDLE MODE, TRANSMIT MODE AND RECEIVE MODE

DSR

Packet Size

100 200 300 400 500 600 700

Receive mode

0.01317

0.015901

0.017571

0.016744

0.018613

0.018896

0.0192

Transmit mode

0.045568

0.056014

0.063108

0.060627

0.068709

0.069231

0.071318

Idle mode

149.958

149.948

149.942

149.944

149.937

149.937

149.935

RIP

Packet Size

100 200 300 400 500 600 700

Receive mode

0.008079

0.007198

0.012669

0.00829

0.007753

0.010093

0.009364

Transmit mode

0.029998

0.027191

0.046771

0.031318

0.029584

0.036045

0.03608

Idle mode

149.973

149.975

149.957

149.972

149.973

149.967

149.967


280

TABLE 12 DATA SET OF THE COLLECTED ENERGY CONSUMPTION (IDLE MODE + TRANSMIT MODE + RECEIVE

MODE)

Scenario II

Energy consumption

Packet Size DSR RIP

100 150.0167 150.0111

200 150.0199 150.0094

300 150.0227 150.0164

400 150.0214 150.0116

500 150.0243 150.0103

600 150.0251 150.0131

700 150.0255 150.0124

FIG. 8 ENERGY CONSUMPTION BETWEEN RIP AND DSR IN

SCENARIO II

The two types of routing protocols have different energy consumption with increasing packet size as shown in Figure 8. DSR has longer energy consumption than RIP, while RIP has smaller energy consumption than DSR. As a result, the RIP showed the best evaluation performance in energy consumption.

Conclusion

In the present paper, an evaluation for routing protocols was carried out on acquired simulation results of two routing protocols, RIP and DSR using QualNet V5. RIP and DSR were selected to represent the Proactive routing protocols and Reactive routing protocols, respectively. We found that Routing Information Protocol preformed better than DSR for all evaluation metrics in 2 different

scenarios.

REFERENCES

[1] S.-J. Lee, J. Hsu, R. Hayashida, M. Gerla, R. Bagrodia,

Selecting a routing strategy for your ad hoc network.

Computer Communications 26 (2003) 723-733.

[2] L. Layuan, L. Chunlin, Y. Peiyan, Performance evaluation

and simulations of routing protocols in ad hoc networks.

Computer Communications 30 (2007) 1890-1898.

[3] D. Wu, QoS provisioning in wireless networks. Wireless

communications and mobile computing 5 (2005) 957-969.

[4] A. Duda, C.J. Sreenan, Challenges for quality of service in

next generation mobile networks. parameters 4 (2003) 3G.

[5] P. Mohapatra, J. Li, C. Gui, Qos in mobile a hoc networks.

Wireless Communications, IEEE 10 (2003) 44-52.

[6] Q. Zhang, F. Yang, W. Zhu, Cross-layer QoS support for

multimedia delivery over wireless internet. EURASIP

Journal on Applied Signal Processing 2005 (2005) 207-219.

[7] S.B. Lee , G.S. Ahn, X. Zhang , A.T. Campbell , INSIGNIA :

An IP-based quality of service framework for mobile ad

hoc networks. Journal of Parallel and Distributed

Computing 60 (2000) 374-406.

[8] Z. Ismail, R. Hassan, Evaluation of ad hoc on demand

distance vector routing protocol in HetMAN architecture.

Journal of Computer Science 6 (2010) 830-836.

[9] S. Tyagi, R. Chauhan, Performance Analysis of Proactive

and Reactive Routing Protocols for Ad hoc Networks.

International Journal of Computer Applications IJCA 1

(2010) 31-34.

[10] S. Mohseni, R. Hassan, A. Patel, R. Razali, Comparative

review study of reactive and proactive routing protocols in

MANETs, in, IEEE, 2010, pp. 304-309.

[11] Z.G. Al-Mekhlafi, R. Hassan, Evaluation Study on Routing

Information Protocol and Dynamic Source Routing in Ad-

Hoc Network, in, 7th Conference on IT in Asia

2011(CITA11). Kuching Sarawak Malaysia, 2011,pp.256-259.

[12] N.S.M. Usop, A. Abdullah, A.F.A. Abidin, Performance

evaluation of AODV, DSDV & DSR routing protocol in

grid environment. IJCSNS 9 (2009) 261-268.

[13] P. Manickam, T.G. Baskar, M. Girija, D.D. Manimegalai,

Performance Comparisons of Routing Protocols in Mobile

Ad Hoc Networks. Arxiv preprint arXiv:1103.0658 (2011).

[14] V. Kanakaris, D. Ndzi, D. Azzi, Ad-hoc networks energy


281

consumption: a review of the ad-hoc routing protocols.

Journal of Engineering Science and Technology Review

(JESTR) 3 (2010) 162-167.

[15] M.J. Islam, M.A. Khaer, M.N. Islam, M.M. Islam, Simplified

XOR Based Approach for Reducing Broadcast

Redundancy with Fast Network Coverage in Wireless Ad-

Hoc Networks, in, 5th International Conference on IT in

Asia (CITA'07), 2007.

[16] Q. Guo, X. Xu, J. Zhu, H. Zhang, A QoS-guaranteed cell

selection strategy for heterogeneous cellular systems. ETRI

Journal 28 (2006) 77-83.

[17] C.E. Perkins, P. Bhagwat, Highly dynamic destination-

sequenced distance-vector routing (DSDV) for mobile

computers. ACM SIGCOMM Computer Communication

Review 24 (1994) 234-244.

[18] R. Coltun, D. Ferguson, J. Moy, A. Lindem," OSPF for IPv6,

in, RFC 5340, July, 2008.

[19] J . Moy, RFC2328: OSPF Version 2. RFC Editor United States

(1998).

[20] G. Malkin, RFC2453: RIP Version 2. RFC Editor United

States (1998).

[21] D.B. Johnson, D.A. Maltz, J. Broch, DSR: The dynamic

source routing protocol for multi-hop wireless ad hoc

networks. Ad hoc networking 5 (2001) 139–172.

[22] C. Perkins, E. Belding-Royer, S. Das, Ad hoc on-demand

distance vector (AODV) routing. (2003).

[23] C.L. Hedrick, RFC1058: Routing information protocol. RFC

Editor United States (1988).

[24] D.B. Johnson, D.A. Maltz, Dynamic source routing in ad

hoc wireless networks. Mobile computing (1996) 153-181.

[25] G. Acs, L. Buttyan, I. Vajda, Provably secure on-demand

source routing in mobile ad hoc networks. IEEE

Transactions on Mobile Computing (2006) 1533-1546.

[26] M. Uma, Padmavathi, A Comparative Study and

Performance Evaualtion of Reactive Quality of Service

Routing Protocols Mobile ADHOC Networks. Journal of

Theoretical and Applied Information Technology 6 (2009) 2.

[27] S. Corson, J. Macker, Mobile Ad hoc Networking

(MANET): Routing Protocol Performance Issues and

Evaluation Considerations[S], RFC2501, 1999.

[28] N. Mishra, N.N. AmitPandey, R. Sinha, D.S. Tapaswi,

Selection of Ad Hoc Network Routing Protocols by

Performance Analysis, in, 2008, pp. 113-116.

[29] M.C. Domingo, Packet Size Optimization for Improving

the Energy Efficiency in Body Sensor Networks. ETRI

Journal 33 (2011).

[30] Y. Tay, K.C. Chua, A capacity analysis for the IEEE 802.11

MAC protocol. Wireless networks 7 (2001) 159-171.


282

An Investigation of Power Performance of Small Grid Connected Wind Turbines under Variable Electrical Loads Md. Alimuzzaman1, M.T.Iqbal2, Gerald Giroux3

1,2Faculty of Engineering and Applied Science, Memorial University of Newfoundland, St. John’s, NL Canada A1B 3X5 3Wind Energy Institute of Canada (W EICan), 21741 Route 12, North Cape, PEI, Canada, C0B 2B0 [email protected]; [email protected]; [email protected]

Abstract

In this study, the power performance of two small grid connected wind turbines has been investigated. The objective was to study the impact of load power factor on the wind turbine power curve. Two small wind turbines were tested in a number of load conditions and test data were collected for about two months. A set of resistors, inductors and capacitors were used as load in addition to the grid connection. Every second set of test data was collected for at least two days in each load condition. Data was analysed for active power, power factor and reactive power. Wind turbines’ power curves are plotted with load and without any load connected between the wind turbine and the grid. Results indicate that the type of load does not significantly affect the power curve of a small wind turbine. It was also observed that above 200W the power factor of a small grid connected wind turbine was also constant.

Keywords

Small wind turbine power performance; grid connected small wind turbine; wind turbine under variable electrical loads; reactive power of small wind turbine; power factor of small wind turbine

Introduction

The wind industry has been contributing a significant percentage of electric power generation all over the world. The power performance and power quality of wind turbines and their interaction with the grid is becoming an important issue [1]. Small wind turbines are being widely used to fulfil local demands. They are used for dairy farms, water supplies for small communities, small industry, irrigation and greenhouses. Many of these turbines are grid connected, so when there is excess electricity they can sell the extra power to a grid and when there is a lack of electric power from the wind turbine, they can purchase electricity from the grid.

To produce electric energy, doubly-fed induction generators (DFIG) in large wind turbines and permanent magnet generators (PMG) in small wind turbines have been widely used. Since a PMG has its own permanent magnet, it does not require an external excitation current, so it does not consume reactive power from the grid. Also, it presents high efficiency and a small size compared to a DFIG. That is why PMGs are dominant in small wind turbine systems [2].

Power performance for small wind turbines is very important. The manufacturers provide a power curve for their wind turbines, which is essentially turbine-produced active power versus wind speed. Depending upon the situation, the small wind turbine load may be resistive, inductive or sometimes capacitive in nature. As the small wind turbines are in dispersed locations, they can be used to provide local reactive power consumption. That may decrease the reactive power flow and also decrease the overall power losses [3].

In this article, the power performance of small wind turbines has been investigated. Two wind turbines were tested at the Wind Energy Institute of Canada (WEICan) with a resistive load, inductive load and capacitive load. Recorded data has been compared and analysed for active power, reactive power and power factor. The results and discussion are presented below.

Grid Connected Small Wind Turbines

A small wind turbine system consists of a rotor, generator, rectifier and inverter. After an inverter, system is connected to an AC panel; the local load is also connected to an AC panel. Before connection to the grid, typically, power goes through a transformer


283

to change the voltage level. For our experiment, the system was arranged in a similar way. To convert the wind energy into mechanical energy, a wind rotor is used. Most of the commercially available small wind turbines use furling, flapping, passive pitching and soft stall for their over speed and power control capacity [4]. For electric power generation, various types of generators have been used. Among them, a permanent magnet generator is widely used. As the wind speed and wind direction change every second, so does the total extracted power also change every second. As a result, the produced electric power from the generator is not uniform. The AC power from the generator is first converted to DC power with the help of a rectifier. Then, the DC power is again converted to the desired AC power so that the output can be connected with grid. The inverter plays a vital role in this system. The inverters must produce good quality sine-wave output and must follow the frequency and voltage of the grid. The inverter must observe the phase of the grid, and the inverter output must be controlled voltage and frequency variations [5]. Most of the commercially available grid tie inverters have an active power factor controller to reduce the Total Harmonic Distortion (THD).

Experiments and Results

To investigate the power performance of small wind turbines, two different turbines were used. For our experiment, we can say that they are turbine A and turbine B.

For data logging, a Campbell scientific 1000 data logger was used, collecting the data for every second and storing it in a pc. This creates a file for the whole day (24 hours). Raw data (second data) is later converted to 1 minute average data. Then, one minute data was normalized and the power curve was plotted using the bin method [6]. The bins method was also used to compare the reactive power and power factor.

For the experiment with a motor, another Hioki meter was used. It can measure three phases - active power, power factor and reactive power - every second and store the data in a memory card. Later, collected data can easily be moved from the memory card to a PC.

Experiments with turbine A

Turbine ‘A’ is a 1.1 KW small wind turbine. It has a PMG. It uses stall regulation for its control system. The specifications of this turbine are as follows:

Rated power: 1.1 KW

Rated wind speed: 12.5 m/s

Power Regulation: Non-stop output control

Maximum output power: 4KW

Permanent magnets generator (3 phase, synchronous)

Over Speed Control/Protection: Stall Regulation

Inverter output Voltage: 120V/208V

Voltage Tolerance: ± 5%

Grid Frequency: 60Hz

Frequency Tolerance: ± 0.00083%

Recommended font sizes are shown in Table 1.

1) Power performance without any local load:

The turbine is grid connected. If any load is connected between the grid and turbine, in this article, it will be called local load. For comparison with local load and non-local load, the power performance data of the turbine with non-local load was collected for ten days. Then, 1 minute average values were calculated and normalized. Finally, by using the bin method, the power curve, reactive power curve and power factor curve were plotted.

Fig. 1(a) shows the power curve of the wind turbine. In Fig 1(b), the power factor is plotted and it indicates that, when wind speed is more than 9 m/s, the power factor is close to unity.

Power factor =active power/apparent power

Apparent power2= active power2+ reactive power2

Reactive power is plotted in Fig 1(c) and it can be observed that reactive power is almost linear and it is less 200 var.

(a)


284

(b)

(c)

FIG. 1 POWER PERFORMANCE OF TURBINE A WITHOUT ANY LOCAL LOAD A) POWER CURVE, B) POWER FACTOR, C)

REACTIVE POWER

Active power output depends on wind speed and it increases as the wind speed increases. But with wind speed increase, there is no significant change in reactive power compared to active power change. So, at high wind speed, apparent power is almost equal to active power and thus the power factor improves at high wind speed.

2) Experiment with Heaters

To experiment on whether resistive load can have an effect on the power performance of wind turbines, three heaters with phase b and/or phase c were connected, as shown below in Fig. 2. It is noted that the turbine is connected to phase b and phase c of the transformer. The heaters were adjusted to maximum point. One of the heaters was a turbine shed heater. It was configured for 208 volt and it was connected to phase b and phase c. The other heaters were portable. They were configured for 120V and they were connected to phase b-neutral and phase c-neutral. They were kept outside the shed.

FIG. 2 BLOCK DIAGRAM FOR AN EXPERIMENTAL SETUP

WITH RESISTIVE LOAD (HEATER)

Heaters were connected for 6 days and 1 second data was collected. Data was converted to 1 minute average data. Using the bin method, the power performance curves were plotted and these are shown in Fig. 3.

(a)

(b)

(c)

FIG. 3 POWER PERFORMANCE COMPARISON OF TURBINE A WITH RESISTIVE LOAD A) POWER CURVE, B) POWER

FACTOR, C) REACTIVE POWER


285

From Fig. 3(a), it can be seen that there is no change in power curve. In Fig. 3(b), there is no change in power factor also. In both cases, when wind speed is greater than 9m/s power factor was very close to unity. Fig. 3(c) shows that there is no significant change in reactive power. In both cases, reactive power decreases gradually and it is always less than 200 var. Therefore, we can say that there is no significant effect of heaters on the output power, power factor and reactive power.

3) Experiment with a 5hp induction motor:

For the experiment with inductive load, a 5hp induction motor was used as the load. The motor was connected before the transformer. The motor was in no load mode, so its active power consumption was low but reactive power was high. The motor was a three phase motor. The motor was left for three days and 1 second data was collected for those days. After that, a set of compensation capacitors was connected for two more days and 1 second data was collected again. Data was analysed using the procedure mentioned earlier.

In this case, another meter (Hioki meter) was connected before the transformer. This meter measured the whole building power performance for every second. Here, the whole building includes the wind turbine, inductive motor, data logger, and building lights. Data was analysed using the bin method. The connection diagram is below in Fig. 4 and data is plotted in Fig. 5.

FIG. 4 BLOCK DIAGRAM FOR EXPERIMENTAL SETUP WITH AN INDUCTIVE LOAD (MOTOR) AND CAPACITOR

Experimental result & comparison in the Hioki meter: From Fig .5(b), we can find that, when the inductive load was connected, the whole building power factor was decreased and when compensation capacitors were connected, the power factor improved. Fig. 5(c) shows that the total demand of reactive power for the whole building was more than 1600var, but with compensation capacitor it was similar to reactive power as without a motor.

FIG. 5 (A) POWER FACTOR COMPARISON OF THE WHOLE

BUILDING WITH INDUCTIVE AND CAPACITIVE LOAD

FIG. 5 (B) REACTIVE POWER COMPARISON OF THE WHOLE

BUILDING WITH INDUCTIVE AND CAPACITIVE LOAD

Experimental result & comparison in the turbine transducer:

(a)

(b)


286

(c)

FIG. 6 POWER PERFORMANCE COMPARISON OF TURBINE A WITH INDUCTIVE AND CAPACITIVE LOAD A) POWER CURVE,

B) POWER FACTOR, C) REACTIVE POWER

From Fig. 6(a) it can be observed that the power curve for all the three cases i.e. without motor, with motor and with motor & capacitor are about same. There is no effect of the induction motor and capacitor. Next, two curves indicate that the power factor is almost similar and reactive power is still less than 200var. So, the inductive load and compensation capacitor have no effect on the power performance of this small grid connected wind turbine.

Experiments with Turbine B

Turbine B is a 1.3 KW small wind turbine. It also has a PMG and it uses the stall regulation method for its control system.

The specifications of this turbine are given here:

Rated power: 1.3 KW

Rated wind speed: 12 m/s

Power Regulation: Active – Inverter

Maximum continuous output power: 1.4KW

Utility interconnection voltage and frequency trip limits and trip times: Programmable, Utility specific

Total Harmonic Distortion (current): < 3%

Trip limit and trip time accuracy:<10%

Grid Voltage: Single phase, 208V

Tolerance: ± 5%

Grid Frequency: 60Hz

Tolerance: ± 0.00083%

1) Power performance without any local load:

Fig. 7 below shows the turbine power curve generated from the data collected over a number of days. It indicates that wind turbine B has a nonlinear power

curve and it is not consistent. Sometimes, it goes into stall regulation too early, say at 8m/s or 10 m/s and remains stalled above that wind speed; it may be a synchronizing problem with grid. At that time, it does not produce any power. The following three power curves are for the same conditions (i.e. without any local load, turbine is connected to the grid directly).

FIG. 7 POWER CURVE VARIATION FOR TURBINE B UNDER

THE SAME CONDITIONS

Therefore, it is very difficult to compare power curves with and without local load.

For our experimental comparison, 1 second data for 14 days without any local load was collected and plotted following the same procedure as described above for wind turbine A. The power curve and reactive power against wind speed was plotted. As the active power varies for the same wind speed, here we have plotted the power factor against the active power instead of power factor versus wind speed.

Fig. 8 shows that the power factor is very constant. When output power is more than 200W, the power factor becomes unity. Reactive power is always less than 150 var.

(a)

(b)


287

(c)

FIG. 8 POWER PERFORMANCE OF TURBINE B WITHOUT ANY LOCAL LOAD A) POWER CURVE, B) POWER FACTOR, C)

REACTIVE POWER

2) Experiment with Heaters

FIG. 9 BLOCK DIAGRAM FOR EXPERIMENT SETUP WITH

RESISTIVE LOAD (HEATER) FOR TURBINE B

Experiments were repeated similar to wind turbine A for resistive load. Turbine B was connected to phase A and phase B in the AC panel, so all the heaters were connected to phase A and/or Phase B, as shown in Fig. 9 and data was collected and analysed. Recorded data is plotted in Fig. 10.

From Fig. 10(a) we can see that there is a little change in the power curve. As mentioned earlier, the power curve is not constant for this wind turbine, so it is better to look at the power factor and reactive power. The power factor is also unchanged. From Fig. 10(b), for both cases, the power factor was close to one when active power was greater than 200 watts. Fig. 10(c) also shows that there is no significant change in reactive power. In both cases, reactive power increases gradually and it was always less than 150 var. Therefore, it could be concluded that there is no significant effect of resistive load on the power performance of turbine B.

3) Experiment with 5hp induction motor and capacitors:

The experiment was r epeated for t urbine B w ith inductive and capacitive load. Fig. 11 below shows connection diagram for this set of experiments. As

(a)

(b)

(c)

FIG. 10 POWER PERFORMANCE COMPARISON OF TURBINE B WITH RESISTIVE LOAD A) POWER CURVE, B) POWER FACTOR,

C) REACTIVE POWER mentioned earlier, a Hioki meter was used before the transformer to measure the whole building power performance. Collected data was analysed and it is plotted in Fig. 12.

FIG. 11 BLOCK DIAGRAM FOR THE EXPERIMENTAL SETUP WITH AN INDUCTIVE LOAD (MOTOR) AND CAPACITOR FOR

TURBINE B


288

Experimental result & comparison in the Hioki meter:

FIG. 12 (A) POWER FACTOR COMPARISON OF THE WHOLE BUILDING WITH INDUCTIVE AND CAPACITIVE LOAD

CONNECTED TO TURBINE B

FIG. 12 (B) REACTIVE POWER COMPARISON OF THE

WHOLE BUILDING WITH INDUCTIVE AND CAPACITIVE LOAD CONNECTED TO TURBINE B

In Fig. 12(a), we can see that, when active power is greater than 200W in both directions (from grid or to grid), the power factor is very close to unity for without motor. When we connected the motor power factor decreased, but when compensation capacitors were added, the power factor improved and came back to unity.

From Fig. 12(b), without the motor reactive power is more or less 200 var. But when the motor was connected, the whole building demand for reactive power increased and it was greater than 1600 var. After that, capacitors were connected and the reactive power demand for the whole building was decreased.

More test results are presented in Fig. 14. Turbine transducer data was collected and analysed as described previously for turbine A.

Experimental result & comparison in the turbine transducer:

From the plots in Fig. 14(a), we can see that the power curve varied a little bit. As this turbine power curve is not a constant, it is better to compare the power factor and reactive power. From Fig. 14(b) and Fig. 14(c), it

can be seen that there is no significant effect of the inductive load and compensation capacitor on the turbine power factor and reactive power.

(a)

(b)

(c)

FIG. 14 POWER PERFORMANCE COMPARISON OF TURBINE B WITH INDUCTIVE AND CAPACITIVE LOAD A) POWER CURVE,

B) POWER FACTOR, C) REACTIVE POWER

Conclusions

This paper described the power performance of two small wind turbines under variable load conditions. The active power performance, the power factor condition and the reactive power performance under resistive load, inductive load and compensation capacitor load have been presented in this paper. From the data, it is concluded that there is no significant


289

effect of load type on the power performance of a small grid connected turbine. When experiments were done with a 5hp inductive motor, which consumes about 1600var, it was found that the turbine also had no effect on its reactive power production; thus, the power factor remained unchanged. The motor consumed the reactive power from the grid. Therefore, these experimental results conclude that the small wind turbine cannot produce any reactive power for induction or capacitive load. The wind turbine inverter basically acts as a current source and its current phase angle is very close to the grid voltage phase angle.

Most of the small wind turbines are situated in remote locations isolated from the national power plant. If somebody buys a small wind turbine to run a small motor for his/her business and connects the turbine to the grid, then the system will still consume reactive power from the grid and there will be loss or utility penalty for the reactive power. A control system should be developed so the owner can adjust the reactive power and power factor for an optimal operation and minimal reactive power from the grid.

ACKNOWLEDGEMENTS

The authors would like to thank the National Science and Engineering Research Council (NSERC), Canada for providing financial support for this research. The authors also would like to thank the Wind Energy Institute of Canada (WEICan), for giving all kinds of technical support for the experiments.

REFERENCES

[1] M.Koulouvari, P. Ladakakos, D. Foussis, E.

Morfiadakis, “Measurement Systems Dedicated to

wind turbine power quality applications” 8th

international conference on Harmonics and quality of

power, 1998.

[2] Carlos E. A. Silva Rene T. Bascope, Demercil S. Oliveira

Jr, “Three-phase power factor correction rectifier

applied to wind energy conversion systems” APEC

2008, Twenty –Third Annual IEEE, Page(s): 768-773.

[3] Peiyuan Chaen, Pierluigi Siano, Birgitte Bak-Jensen,

Zhe chen, “Stochastic optimization of wind turbine

power factor using stochastic model of wind power” ,

IEEE transactions on Sustainable Energy ,April 2010.

[4] Md. Arifujjaman, M.Tariq Iqbal, John E. Quaicoe, M.

Jahangir Khan, “Modeling and control of a small wind

turbine” Canadian Conference on Electrical and

computer Engineering, 2005.

[5] Varin Vongmancee, Veerapol Monyakul, “Grid

connected inverter with unity power factor for wind

power applications” industrial Electronics &

Applications, 2009.

[6] A. Llombart, S. J. Watson, D. Llombart and J. M.

Fandos, “Power Curve Characterization I: improving

the bin method” International conference on renewable

energies and power quality, 2005.

international journal of energy science

Documents

washington university

university of washington

university of florida

university of zagreb

curtin university

melbourne university

university of nebraska

china university of