This may be the author’s version of a work that was submitted/acceptedfor publication in the following source:

Kim, Eric, Tan, Andy, Mathew, Joseph, & Yang, Bo-Suk(2010)Development of an online condition monitoring system for slow speed ma-chinery.In Mathew, J, Kiritsis, D, Koronios, A, & Emmanouilidis, C (Eds.) Pro-ceedings of the 4th World Congress of Engineering Assets Management(WCEAM 2009).Springer-Verlag London Ltd, United Kingdom, pp. 743-749.

This file was downloaded from: https://eprints.qut.edu.au/40825/

c© Copyright 2010 Springer-Verlag London Ltd.

This is the author-version of the work. Conference proceed-ings published by Springer Verlag will be available via SpringerLink.http://www.springerlink.com

Notice: Please note that this document may not be the Version of Record(i.e. published version) of the work. Author manuscript versions (as Sub-mitted for peer review or as Accepted for publication after peer review) canbe identified by an absence of publisher branding and/or typeset appear-ance. If there is any doubt, please refer to the published source.

https://doi.org/10.1007/978-0-85729-320-6_86

QUT Digital Repository: http://eprints.qut.edu.au/

This is the accepted version of the following conference paper: 

Kim, Eric, Tan, Andy, Mathew, Joseph, & Yang, Bo‐Suk (2010) Development of an online condition monitoring system for slow speed machinery. In: Proceedings of the 4th World Congress of Engineering Assets Management (WCEAM 2009), 28‐30 September 2009 , Ledra Marriott Hotel, Athens. 

© Copyright 2010 Springer-Verlag London Ltd. This is the author-version of the work. Conference proceedings published by Springer Verlag will be available via SpringerLink. http://www.springerlink.com

DEVELOPMENT OF AN ONLINE CONDITION MONITORING SYSTEM FOR SLOW SPEED MACHINERY

Eric Y Kima, Andy C. C. Tana, Joseph Mathewa and Bo-suk Yangb

a CRC for Integrated Engineering Asset Management, Queensland University of Technology, 2 George St, Brisbane, QLD 4001, Australia

b School of Mechanical Engineering, Pukyong National University,San 100, Yongdang-dong, Nam-gu, Busan, 608-739, South Korea.

One of the main challenges of slow speed machinery condition monitoring is that the energy generated from an incipient defect is too weak to be detected by traditional vibration measurements due to its low impact energy. Acoustic emission (AE) measurement is an alternative for this as it has the ability to detect crack initiations or rubbing between moving surfaces. However, AE measurement requires high sampling frequency and consequently huge amount of data are obtained to be processed. It also requires expensive hardware to capture those data, storage and involves signal processing techniques to retrieve valuable information on the state of the machine. AE signal has been utilised for early detection of defects in bearings and gears. This paper presents an online condition monitoring (CM) system for slow speed machinery, which attempts to overcome those challenges. The system incorporates relevant signal processing techniques for slow speed CM which include noise removal techniques to enhance the signal-to-noise and peak-holding down sampling to reduce the burden of massive data handling. The analysis software works under Labview environment, which enables online remote control of data acquisition, real-time analysis, offline analysis and diagnostic trending. The system has been fully implemented on a site machine and contributing significantly to improve the maintenance efficiency and provide a safer and reliable operation.

Key Words: Condition Monitoring, Low Speed Machinery, Rolling Element Bearing, Acoustic Emission

1 INTRODUCTION

Condition Monitoring (CM) is the process used to determine the operational state and health of a machine for the purpose of detecting potential failures before they turn into functional failures. Condition Monitoring is an integral part of Predictive Maintenance (PdM) or Condition-Based Maintenance (CBM). Typical CM techniques include vibration analysis, oil analysis, wear particle analysis, ultrasonic analysis, thermographic analysis and motor current signature analysis.

Low speed machines (LSM) are usually large and have high rotating inertia. Bearings are arguably the most critical component to be monitored in most LSMs. Extensive research has been done on diagnosing rolling element bearing defects using vibration analysis [1-7]. At low speeds, the impact energy between the rotating elements and the defect is generally low and consequently it is weakly detected through vibration transducer. Theoretically, it is possible to extract a low energy signal using traditional signal processing methods, but in reality it is problematic. The use of wear particle analysis is impractical because low speed units are usually grease lubricated. Conventional techniques involving acceleration vibration may not be able to detect a growing fault due to the low impact energy generated by the relative moving components. This has lead to an increasing use of the Acoustic Emission (AE) technique in fault diagnosis of rotating machinery in low speed machinery condition monitoring [9-13].

AE technique has been revealed to be the most sensitive approach as compared to vibration and ultrasound measurements for detection of incipient faults especially in low speed situations [14]. However, due to the high frequency nature of AE signal, it requires a high sampling rate in the region of a few MHz. This is one of the main shortcomings of the AE application as it requires a highly specialized data acquisition system and experts to interpret the results. An additional problem with AE application in low speed machines is that it requires not only high sampling frequency but also longer recording time in order to cover enough complete shaft revolution. Longer time recording is also necessary to have an enough resolution to observe sidebands of defect frequencies in frequency spectra. Hence, AE data are normally analyzed using the hit-based or continuous

2

time-based and not in the frequency spectrum. This also restricts the application of many advanced signal processing techniques to AE signal analysis. Kim et al [15, 16] have revealed the effectiveness of the AE signals in the frequency range up to 100 kHz for low speed bearing CM/D.

This paper introduces an online CM system which uses the AE-based technique for LSMs. A feasible on-line data acquisition system has been constructed as LSMs are often exposed to harsh environment and hence manual CM data acquisition is has to be minimised in practice. For this reason, the sensors and data acquisition system are installed on the selected machine and the data are collected remotely via web. This enables detection of damages as soon as they appear and minimising the need for regular human interference. Software has been developed using Labview, which enables online remote control of data acquisition setup, online real-time analysis, offline analysis and diagnostic trending.

2 SIGNAL PROCESSING AND FEATURE EXTRACTION

Most of the research on bearing diagnosis based on vibration measurement can be categorized in the time domain and frequency domains. The RMS, crest factor, probability density moments (skewness, kurtosis) are the most popular statistical time domain parameters for bearing defect detection [3]. In the frequency domain, the enveloping method, also known as demodulation or HFRT (High Frequency Resonance Technique), is the most popular technique for detecting of localized defects [4,5]. Wavelet analysis has also been successfully applied to bearing defect detection [6-7].

Recently, AE measurement has been widely applied to assessing the integrity of machines as well as structures. Most of AE applications have been focused on high frequency events (100kHz~1MHz) of AE signals using highly specialized data acquisition and analysis systems. Hit-based features (counts, energy, amplitude, rising time and duration) are typical monitoring features in many applications. However, in rotating machinery the high frequency AE signals are easily overwhelmed by background noise coming from various moving machine parts. A drawback of using the hit-based AE features is the lack of frequency information of the AE events. Frequency of the occurrence of burst-type hit events is by far the most relevant feature for identifying bearing defect frequencies, such as ball passing frequency on outer race (BPFO) or ball passing frequency on inner race (BPFI). It is impractical to get the frequency of AE events in low speed situation because much longer duration of data is required, and consequently a huge amount of data is generated for each revolution of the shaft.

Kim et al [15, 16] have successfully applied low frequency (up to 100kHz) AE signals analysis for early detection of low speed bearing faults. This research utilises those low frequency AE signals with proven processing algorithms, such as time domain features and envelope analysis. Fig. 1 shows the signal processing and feature extraction flow used in the software. Wavelet packet analysis was employed as a band-pass filter. Adaptive line enhancer (ALE) is applied as an option for pre-processing to enhance the signal to noise ratio.

Fig. 1 Flow of signal processing and feature extraction

2.1 Peak-hold down sampling (PHDS)

3

The main challenge in AE application to low speed machinery is its huge amount of data to process due to the high frequency sampling of AE signals and the long duration of time recording in order to obtain the frequency information. Traditional AE-based techniques for slow speed bearings do not involve spectrum analysis though they have been used in bearing diagnosis. To overcome this issue, this research proposed the peak-hold down sampling (PHDS) which keeps the information of high frequency burst AE signals with much reduced sampling frequency.

Assuming the original sampled data is

,,,,,

,,,,,

321

321

i

i

xxxx

tttt (1)

where, the original sampling frequency )/(1 1 iiO ttf .

The down sampled data can be expressed as

,,,,,

,,,,,

321

321

j

j

YYYY

TTTT (2)

where, the new down sampling frequency )/(1 1 jjD TTf .

For the j-th segment of original data, ],,,,[ 21 niiiij xxxxX , where, ji Tt ,

ji Tt 1,

1 jni Tt and 11 jni Tt , the

down sampled data jY is calculated

jj

jj

j

jjj

jjj

j

jj

MINYelse

MAXYthen

MAXif

else

MAXMINYelse

MAXMINYthen

Yif

then

MINMAXif

,

,

0

,

,max,

,max,

0

,

0

1

(3)

where, jj XMIN of valueminimum: and

jj XMAX of valuemaximum: .

In Fig. 2, PHDS signal is compared with normal down-sampled signal in time domain and envelope spectrum. The original signal was from a defective bearing of a gearbox in the aluminum smelter. The result clearly shows that the normal 1kHz down-sampled signal does not provide the bearing defect frequency (BPFO) in the envelope spectrum while the PHDS 1kHz signal give a similar spectrum with that from original 250kHz sampling. In this case the data size was reduced by 250 times giving similar detectability. This is a significant advantage for storing, processing and analyzing the measured data especially from low speed machines due to its longer time span required to get enough resolution in frequency spectrums.

2.2 Noise removal

In low speed machinery, the impact energy generated by the rolling elements on the defective components is insufficient to produce a detectable vibration response. This is aggravated by the inability of general measuring instruments to detect and process the weak signals accurately. Furthermore, the weak incipient signals are often corrupted by background noise along its transmission path to the sensor located on the machine surface.

Kim et al [17] compared two noise removal techniques to enhance the slow speed bearing signals so as to increase the detectability of incipient defects. Blind deconvolution (BD) and adaptive line enhancer (ALE) were used as wide-band notch filter to remove any sinusoidal-type noise. Optimum filter length was chosen to maximise kurtosis and modified peak ratio. The performance of the two algorithms were compared and validated by using simulated bearing signals and signals from test rig with a defective bearing in low speed. Adaptive line enhancer was incorporated in this application owing to its better performance and computation efficiency.

4

(a) Time wave form (b) Envelope spectrum

Fig. 2 Comparison of PHDS with normal down sampling.

3 HARDWARE IMPLEMENTATION

3.1 Target machine

The mixer gearbox in the green carbon production line in an aluminium smelter has been chosen as the target machine on which the developed technique is applied. A continuous supply of carbon anodes is a key factor in the successful operation of the smelter. Effective planning and organisation of maintenance work prevents production backlogs, keeps equipment in top condition and ensures that productivity remains high. Integrated information and technology systems are vital to success of smelters.

Primary aluminium is produced in reduction plants (or "smelters"), where pure aluminium is extracted from alumina by the Hall-Héroult process where calcined alumina is dissolved in molten cryolite at a temperature of 900 degree C. For each ton of aluminium produced about 430 kg of carbon is consumed. Carbon anodes are manufactured in dedicated carbon plants that comprise paste plants, carbon bake furnaces and rodding shops. In the paste plants, carefully crushed and graded fractions of calcined petroleum coke and recycled anode butts are heated and mixed with molten pitch. The hot mixture is then compacted into blocks called green (unbaked) anodes. The green anodes are transferred to the carbon bake furnaces where they are heated in deep brick-lined pits to around 1 100ºC over a period of 21 days. This baking process calcines the binding pitch and ensures that the anodes have good thermal and electrical conductivity. Exhaust manifolds collect waste gases and carry them to the fume treatment centre. After baking, aluminium rods are attached to the anodes and sealed with cast iron. The rod suspends the anode in the pot and acts as an electrical conductor. After the rods are attached, the anodes are delivered to the pot rooms for positioning in the pots.

The role of the gearbox is to produce reciprocating and rotating motion of around 50rpm for the paste mixer. This lateral reciprocating motion causes two evident impact motions when it reaches the maximum stroke due to the significant inertia of the mixer shaft and mixing material. This impact makes the analysis difficult in time domain as well as frequency domain. In time domain, this high level impact signal overwhelms minute useful signals from defects. In frequency domain, these impulsive signals induce a series of harmonics and their side bands in frequency spectrum. Consequently, these eventually veil the minute defective frequencies what the analyst is looking for.

3.2 Data acquisition system

Low speed machines are often exposed to harsh environment. In the green carbon production line, crushed and graded fractions of calcined petroleum coke and recycled anode butts are heated and mixed with molten pitch. A respiratory is compulsory to access to the site as toxic coke powder rides in the air. Hence manual inspection and data acquisition are restricted. Accordingly, online remote monitoring is highly expected for this machine.

0 0.5 1 1.5 2 2.5 3 3.5 4-1.5

-1

-0.5

0

0.5

1

1.5

0 0.5 1 1.5 2 2.5 3 3.5 4-1.5

-1

-0.5

0

0.5

1

1.5

Time (sec)

Am

plitu

de (

mV

)

0 0.5 1 1.5 2 2.5 3 3.5 4-1.5

-1

-0.5

0

0.5

1

1.5

Time (sec)

Am

plitu

de (

mV

)

50 100 150 200 250 300 350 4000

0.5

1

1.5x 10

Am

plitu

de (

mV

)

Frequency (Hz)

50 100 150 200 250 300 350 4000

2

4

6

8x 10

-5 AE1, Env, [32 64]kHz, Origin vs PHDS(250:1kHz)

50 100 150 200 250 300 350 4000

1

2

3

4x 10

-4

Am

plitu

de (

mV

)

Frequency (Hz)

Original 250 kHz Sampling

PHDS 1 kHz Sampling

1 kHz Sampling

BPFO

5

For these reasons, a decision was made to construct online condition monitoring system to implement the AE-based monitoring technique. Suitable hardware systems have been explored. The key requirements for the data acquisition hardware are online monitoring capability, compatibility, expandability and cost. National Instrument’s PXI system (www.ni.com) was chosen as the rugged PC-based platform offers a high performance, low-cost deployment solution for measurement and automation systems. PXI also adds mechanical, electrical, and software features that define complete systems for test and measurement, data acquisition, and manufacturing applications. These systems serve applications such as manufacturing test, military and aerospace, machine monitoring, automotive, and industrial test. The main advantage of NI PXI system over other similar products is the flexible control software. Labview software enables to create fully functional measurement application with analysis and a custom user interface using a variety of hardware.

3.3 AE sensor and installation

There are two types of AE sensors: resonance-type and wideband-type. Resonance-type AE sensors maximise the sensitivity in a specific frequency range near the resonance frequency, while wideband-type one enables covering wide range of frequency normally from 100kHz to 1MHz. In the use of AE sensor for monitoring of rotating machinery resonance-type AE sensors are often used because high frequency (over 100kHz) signals attenuate very short distance and hence, are easily buried by background noise. The authors found the feasibility of AE signal in the range of 50-100kHz [14]. Therefore, we chose resonance-type AE sensor which has its resonance at 60kHz.

A preamplifier is normally used with AE sensor as the signal detected by the AE sensor is too weak and easily buried by the background noise. Integrated-type AE sensor in which a preamplifier is integrated gives often benefits of making the wiring simpler, which is often suitable for industry application without need for a separate preamplifier. This enables driving long cables without interfering by noise. Physical Acoustic Corporation (PAC)’s CH6I is selected as this meets our requirements well. Usually, AE sensors are attached to the object using magnetic holders. However, in this application, the mixer gearbox runs with high vibration due to its reciprocating motion. Also, some sensors need to be attached at the lower side of the bearing housing in order to be close to the loading zone of bearing. Therefore, they were attached in the surface of the gearbox securely and permanently using the epoxy glue which can be detached by using acetone.

The schematic of the data acquisition system and its installation are shown in the Figure 3. Six AE sensors, one accelerometer and one tachometer were installed permanently on the target gearbox. A signal conditioning box for AE sensors is to decouple AE signals from the original signal coupled with 28 VDC. The signals are digitised at the DAQ card and saved in the PXI system. The PXI is connected to internet via local network with security function allowing access from anywhere in the world through internet.

Fig. 3 Schematics of the hardware system and its installation besides the machine

4 SOFTWARE IMPLEMENTATION

The CM software has been developed on the Labview environment which directly controls the data acquisition hardware. The software enables real-time analysis as well as off-line analysis. The software consists of data acquisition control, real-time analysis, off-line analysis, feature extraction and trending of features. The work on incorporating intelligent diagnostics

6

algorithms are undergoing based on acquired historical features. One of preliminary works for this can be found in [19]. Fig. 4 shows some screen shots of the software.

(a) Main window (b) bearing and gear data input window

(c) Analysis window (d) Feature trending window

Fig. 4 Screen shots of the CM software

The system has detected a strong evidence of a bearing defect on the outer race of the input shaft bearing. No evidence for all other components has been detected. AE spectrum clearly indicated the bearing defect frequencies of outer race, while no evidence is shown in the vibration spectrum. As a matter of fact, it has not been detected by the current routine vibration-based monitoring activity on site. The degree of deterioration of the outer race was not significant as it was only detected by high frequency AE ranges. Hence, any maintenance action was not necessary at the stage but a close monitoring was recommended for this bearing.

The software has an option of extracting all relevant diagnostic features for trending. The features include: mean Peak Ratio for outer-race, inner race, ball and train from envelope spectrum; RMS, crest factor, skewness and kurtosis from four levels of wavelet packet analysis from time waveform; amplitude of harmonics of rotating speed and gear mesh frequencies from raw spectrum. Fig. 5 shows a trending of mean Peak Ratio (mPR) [16] of outer race for five bearings. The mPR of bearing 2 is distinctively higher than other bearings, which indicates that its condition is much worse than others. It is also noted from this trending that although a spall was initiated on outer race, its deterioration rate is slow at the moment but careful monitoring is required for this bearing.

7

Fig. 5 Example of mPR trending for the bearings

5 CONCLUSIONS

This paper introduced an AE-based condition monitoring system for low speed machinery. To overcome challenges of traditional AE-based technique to CM of low speed bearings, the peak-hold down sampling was proposed. CM software has been developed in Labview environment, which enables online remote control of data acquisition, online real-time analysis, offline analysis and diagnostic trending. The system is successfully implemented onto site machine and has detected a defect on bearing in early stage of its deterioration. The superiority of the developed AE-based method over traditional vibration-based one lies on earlier detection of defects with an increased sensitivity, which enables not only to prevent a catastrophic failure but to optimize the maintenance schedule having enough leading time until real failure occurs. Further research is being carried out to develop a robust gear diagnostic algorithm and an intelligent diagnostic algorithm based on the data acquired from the developed system.

6 REFERENCES

1. N. Tandon and B. C. Nakra, "Vibration and acoustic monitoring techniques for the detection of defects in rolling element bearings - a review," Shock and Vibration Digest, vol. 24, pp. 3-11, 1992.

2. Tandon N & Nakra BC. (1992) Vibration and acoustic monitoring techniques for the detection of defects in rolling element bearings - a review. Shock and Vibration Digest, 24, 3-11.

3. Tandon N & Choudhury A. (1999) A review of vibration and acoustic measurement methods for the detection of defects in rolling element bearings. Tribology International, 32, 469-480.

4. Alfredson AC & Mathew J. (1985) Time domain methods for monitoring the condition of rolling element bearings. Mechanical Engineering Transactions.

5. McFadden PD & Smith JD. (1984) Vibration monitoring of rolling element bearings by the high frequency resnance technique - a review. Tribology International, 17, 3-10

6. Liu TI & Mengel JM. (1992) Intelligent monitoring of ball bearing conditions, Mechanical Systems and Signal Processing, 6, 419-431.

7. Paya BA, Esat II & Badi MNM. (1997) Artificial neural network based fault diagnostics of rotating machinery using wavelet transforms as a pre-processor, Mechanical Systems and Signal Processing, 11, 751-765.

8. Lin J, Zuo MJ & Fyfe KR. (2004) Mechanical fault detection based on the wavelet de-noising technique, Journal of Vibration and Acoustics, Transactions of the ASME, 126, 9-16.

9. Rao RBKN. Advances in acoustic emission technology (AET) in COMADEM: A state-of-the-art review, COMADEM2003, 1-18.

10. Yoshioka T, Korenaga A, Mano H & Yamamoto T. (1999) Diagnosis of rolling bearing by measuring time interval of AE generation, Journal of Tribology, Transaction of ASME, 121, 468-472.

11. Tandon N & Nakra BC. (1999) Defect detection in rolling element bearings by acoustic emission method, Journal of Acoustic Emission, 9(1), 25-28.

12. Shiroishi J, Li Y, Liang S, Kurfess T & Danyluk S (1997) Bearing condition diagnostics via vibration and acoustic emission measurements, Mechanical Systems and Signal Processing, 11(5), 693-705.

13. Jamuludin N, Mba D & Bannister RH. (2001) Condition monitoring of slow-speed rolling element bearings using stress waves, In Proceedings of the IMechE, 215(4), pp. 245-271, ProQuest Science Journal.

8

14. Shi DF, Wang WJ & Qu LS (2004) Defect detection for bearings using envelope spectra of wavelet transform, Journal of Vibration and Acoustics, Trans. ASME, 126(4), 567-573.

15. Kim Y-H, Tan ACC, Mathew J & Yang B-S. (2006) Condition monitoring of low speed bearings: a comparative study of the ultrasound technique versus vibration measurement, In WCEAM’06, World congress on engineering asset management, paper no. 26, Gold coast, Springer Publisher.

16. Kim Y-H, Tan ACC, Mathew J, Kosse V & Yang B-S. (2007) A comparative study on the application of acoustic emission technique and acceleration measurements for low speed condition monitoring, In APVC’07, Asiapacific vibration conference, Sapporo.

17. Kim EY, Tan ACC, Yang B-S & Kosse V. (2007) Experimental study on condition monitoring of low speed bearings: time domain analysis, In 5th ACAM, Australasian Congress on Applied Mechanics, Brisbane.

18. Kim EY, Tan A & Yang B-S (2008) Fault detection of slow speed rolling element bearing with noise removal techniques, In 15th ICSV, International Congress on Sound and Vibration, pp. 1981-1988, Daejeon,

19. Widodo A, Kim EY, Son JD, Yang Bs & Tan ACC. (2009) Fault diagnosis of low speed bearing based on relevance vector machine and support vector machine, Expert Systems with Applications, 36(3), 7252-7261.

Acknowledgments

The work is supported through a grant from the CRC for Integrated Engineering Asset Management (CIEAM).