artificial intelligence for industial applications

Upload: gmcnultyenergy

Post on 10-Apr-2018

222 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/8/2019 Artificial Intelligence for Industial Applications

    1/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] or 087-6697867 1 06/11/09

    The Uses of Artificial Intelligence in Industry forEnergy savings and Process improvements

    Prepared for Institute of Engineers SeminarJuly 1997

    Gerard McNulty C.Eng.F.I.E.ISystems Optimisation Ltd

  • 8/8/2019 Artificial Intelligence for Industial Applications

    2/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] or 087-6697867 2 06/11/09

    Table of Contents

    INTELLIGENT INDUSTRIAL SYSTEMS 3

    BENEFITS OF ARTIFICIAL INTELLIGENCE 4EXAMPLES OF ARTIFICIAL INTELLIGENCE IN INDUSTRY 5USING ARTIFICIAL INTELLIGENCE TO IMPROVE BUSINESS PERFORMANCE 6USING ARTIFICIAL INTELLIGENCE TO BECOME A WORLD CLASS MANUFACTURE 7THE PRINCIPLES OF ARTIFICIAL INTELLIGENCE NEURAL NETWORKS 8ARTIFICIAL INTELLIGENCE TECHNIQUES 9DATA MINING IN INDUSTRY 10DATA MINING TECHNIQUES 13NEURAL NETWORKS 14

    A THREE LAYER NEURAL NETWORK 14COMBINING THE TECHNIQUES 15

    EXAMPLE OF COMBINING THE TECHNIQUES: 15INTELLIGENT SENSORS 16SOFT SENSORS 17INTELLIGENT SOFT SENSORS - EXAMPLES 18SUPERVISORY CONTROL 19ADVANCED CONTROL 20SUMMARY OF THE BENEFITS OF USING NEURAL NETWORKS FOR 21ADVANCED CONTROL APPLICATIONS 21MONITORING AND DIAGNOSTIC ANALYSIS 22

    OPTIMISATION USING GENETIC ALGORITHMS 23GENETIC ALGORITHMS TO THE RESCUE 24SCHEDULING USING ARTIFICIAL INTELLIGENCE 25PROCESS MODELLING AND SIMULATION 26KNOWLEDGE BASED SYSTEMS 27

    LEARNING FROM DATA: 28

  • 8/8/2019 Artificial Intelligence for Industial Applications

    3/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] or 087-6697867 3 06/11/09

    Intelligent Industrial Systems

    Artificial Intelligence is a general term used to describe various forms ofadvanced computational and control techniques. These techniques utilise

    special analysis processes, the basis of which is the neural network, thatmimic the natural reasoning method of the Human Brain.

    Because neural networks analyse data more wisely then conventionaltechniques, they can provide insights into problem domains previously notpossible to obtained or required vast computational power to do so. The netresult is improved process performance by intelligent control and optimisation,increase sensor reliability through self validation, improved understanding of aparticular are of interest using data mining to discover the contributions ofvarious attributes to an outcome of interest - such as for example finishedproduct quality.

    One benefit of using artificial intelligence includes better decision making andhence increases company profits. Since your success is depending upon youcompanys accumulative intellectual ability compared to the competition,artificial intelligence can provide you with the production edge, the qualityedge, the marketing edge and the ability to make the right decisions at theright time.

    Artificial Intelligence gives companies the power to Prevent problems not justfix them, to Analysis data not just depend on experience, to focus on theProcess not the just the product, to provide for Intelligent localised processcontrol and optimisation not just process monitoring, to provide overallcompany direction by Intelligent Data Analysis not just seat of pantsmanagement. Its the technology that weve all been waiting for-likespreadsheets were to the accountant 10 years ago. What would they dowithout spreadsheets today! - Its unthinkable, isnt it.

    The falling cost of data has encouraged all of to collect more data then we cananalyse. Imagine if you could extract hidden patterns from this data or extractvaluable knowledge from it. Using these new techniques, you wont need asmall band of statisticians to do it.

    The use of Artificial Intelligent techniques in Industry will provide a new era inprocess improvements that will result in a big reductions in process variabilityand hence product quality improvements. To improve a process is to improvea product. Improving process variability will reduce reject rates, cut costs, andincrease market share - the key to a successful business.

  • 8/8/2019 Artificial Intelligence for Industial Applications

    4/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] or 087-6697867 4 06/11/09

    Benefits of Artificial Intelligence

    These are some of the benefits that can be obtained by using artificialintelligence and advanced computational techniques in your business

    Increase Yield through data mining of the factors that effect output andcause production bottlenecks

    Reduce energy usage through process optimisation of set points andoperational schedules along with intelligent control to produce stable andresponsive operation of processes and systems

    Reduced environmental impact by better understanding of processparameters that contribute to emissions excursions

    Increase plant reliability through better understanding of the factors thateffect plant utilisation and the mean time between failures

    Reduce manpower requirements by intelligent controls

    Increase plant safety by ability of plant control system to learn fromexperience and from intelligent operational practices

    Improved product quality through better understanding of factors thateffect quality along with their individual contributions to quality.

    Understand the factors effecting that effect machine downtime

  • 8/8/2019 Artificial Intelligence for Industial Applications

    5/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] or 087-6697867 5 06/11/09

    Examples of Artificial Intelligence in Industry

    Artificial Intelligence or AI is not just for the PhDs of this life; its a technologythat everyone can utilise to his or her benefits. Below are a few examples of

    the type of industries that are using AI to remain on top. Remember that 70 ofthe Fortune top 100 companies use AI. Here are some examples:

    Power station optimisation using genetic algorithms

    Self Turning PID Loops to improve control system performance

    Expert system to model and validate process data

    Improve control of a refrigeration plant to optimise COP

    Steam System Optimisation ( Boilers and Turbines ) to reduce runningcost

    Prediction of product quality using neural networks in glass manufacture

    Furnace scheduling using expert systems to minimise energy costs

    Data Mining to reduce utility plant running costs on energy

    Process simulation to identify opportunities for improvements in energycosts

    Data Mining to improve insight into process operation in chipmanufacture

    Optimisation routine to establish the most economical mix of ingredientsin glass manufacture

    Optimise Air Compressor installation to reduce electrical running costs

  • 8/8/2019 Artificial Intelligence for Industial Applications

    6/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] or 087-6697867 6 06/11/09

    Using artificial Intelligence to Improve Business Performance

    Behavioural Traditional Artificial Intelligence

    Style Approach Approach

    Problems Fixing Preventing

    Problem Solving Expert Based Based on data/systems

    Analysis Experience Data

    Focus Product Process

    Behaviour Reactive Proactive

    Reasoning Experience Based Statistically Based

    Outlook Short Term Long Term

    Decision Making Intuition Data Based

    Direction Seat of pants Benchmarking /Metrics

    Control Centralised Localised with AI

    Improvements Dumb Automation Continuous Optimisation

  • 8/8/2019 Artificial Intelligence for Industial Applications

    7/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] or 087-6697867 7 06/11/09

    Using Artificial Intelligence to become a World ClassManufacture

    Your company survival is depending upon continuously growing your

    business. We all know that growth in business means growth in sales of yourproducts. Growth in product sales is largely determined by your Customersatisfaction. Customer satisfaction is governed by your product quality, price,and delivery times. Quality, price and delivery are not just controlled bychance but by the ability of your process to produce a high quality product,free from defects, at the lowest possible cost and cycle times. Only a processthat is free from variations will be free from defects etc.

    Artificial Intelligence will help your company reduce process variations andhence educe defects, reduce operational costs and cycle times by learning

    from experience. It will help to identify potential problems and provideintelligent control and meaningful management information when you need it.

    Simply put the quality of our products and services are a reflection of howcapable our processes really are. To measure product quality is to measureprocess quality. Process quality is depending upon the control we have overit.

    Artificial Intelligence allows us to work smarter, not harder. This translates intomaking fewer mistakes in all that we do, from the way we manufacture ourproducts to the way we organise our delivery. As we data mine and discover

    and eliminate harmful sources of variation, our quality goes up and ourdefects rates go down, we keep customers and grow our business.

    The table below shows a relative means of benchmarking differentorganisations, processes, and systems based on defects per unit where a unitis any task, product, or physical entity. An invoice, a product made by amachine etc.

    Process Capability Defect per million Opportunity

    2 308,500 ( Worst Eastern)

    3 66,800 (Average Eastern)4 6,200 ( Average Western )

    5 2300 ( Good Western )

    6 3.4 ( World Class )

  • 8/8/2019 Artificial Intelligence for Industial Applications

    8/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] or 087-6697867 8 06/11/09

    The Principles of Artificial Intelligence Neural Networks

    If you hate statistics, then neural networks may be for you. When process

    engineers are faced with complex multi-variable problems they have a numberof options:

    Use rules of thumb and manual intervention - but expect that a largeerrors can result

    Wade through a mass of statistical and non-linear programmingtechniques - if you remember how to do them and hope that the processinteractions you're trying to figure out are all linearly related to oneanother and there only a few parameters that really matter.

    Try Neural Networks

    A neural network takes its design from the current understanding of how thenervous system of living things works. It learns, it can remember and it canadjust, to new situations unlike the present dumb control systems aroundtoday.

  • 8/8/2019 Artificial Intelligence for Industial Applications

    9/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] or 087-6697867 9 06/11/09

    Artificial Intelligence Techniques

    Data Mining for process and product improvements and managementinformation systems

    Intelligent sensors

    Soft Sensors

    Supervisory Control

    Advanced Control

    Monitoring and Diagnostic Analysis

    Optimisation

    Scheduling

    Process Modelling and Simulation

    Knowledge Based systems for product improvements and managementinformation systems

  • 8/8/2019 Artificial Intelligence for Industial Applications

    10/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] or 087-6697867 10 06/11/09

    Data Mining in Industry

    The falling costs of data storage and processing power have encouragedlarge and small organisations world-wide to accumulate vast quantities of

    data. Data files are records of the behaviour or performance of machines,processes, human resources, the environment, etc. Accumulating data fileshas been perceived as gathering information, when in reality it can representinformation overload or clutter unless relationships and patterns in the datacan be derived. As such, data files can be viewed as reservoirs of knowledgethat can be mined to discover relationships, patterns and rules. Theobjective of learning from data is to extract such knowledge from data files.

    Learning from data falls into two categories - symbolic and connectionistlearning. Symbolic learning can generate rules and patterns from data files,while connectionist learning generates networks of processing units from thesame data. By definition, symbolic learning generates results that areunderstandable to the human user.

    Connectionist learning builds numeric computer models from data, with typicalmembers of this latter category being Neural Networks. While the accuracy ofconnectionist models can often be good, they suffer from a lack ofunderstandability - a 'black box solution, whereas the pattern rules anddecision trees can offer both accuracy and clarity.

    There are two modes of learning - supervised and unsupervised learning.

    Unsupervised learning can be used to discover any clustering or patterns indata without specifying an outcome data field of interest using statisticaltechniques that are limited in their power. Supervised learning is used togenerate rules and patterns linking a selected data field to other designateddata fields.

    Learning from data can be considered an alternative knowledge engineeringstrategy if the data represents records of expert decision making.Alternatively, learning from performance data can derive new patterns andrelationships that improve our understanding of a certain process andtherefore enable us to make better decisions in the future.

    Data models (rules and patterns) derived from historical data can be used topredict the outcome of future events. This is called classification if theoutcome of interest is a discrete category such as 'good' or 'bad' in predictingsay glass product quality, while it is called prediction if the outcome of interestis a numeric quantity such as 'price' when predicting the movement of thevalue of stocks.

    Hence, the overall objective is to derive decision trees, patterns rules andneural networks from your data files. This will give you the ability to do morePrediction and Prevention of process problems rather then the traditional

  • 8/8/2019 Artificial Intelligence for Industial Applications

    11/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] or 087-6697867 11 06/11/09

    Detect and React. Data mining aim is to establish the significant inputvariables and than determine how they affect the output capability of theprocess. For example, we can produce from such data a graphical decisiontree to profile any data field in relation to other data fields.

    From this decision tree, we can produce pattern rules - for discrete outcomessuch as product quality were the contribution of various input variables orattributes are determined - for example.

    ( if temp > 300 degc and % lead < 40% and product = A then quality =C(prob 0.7 )

    Data Mining may extract many such pattern rules from a data file. It may onlytake a handful of deduced rules to provide invaluable information about youprocess

  • 8/8/2019 Artificial Intelligence for Industial Applications

    12/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] or 087-6697867 12 06/11/09

    Data Mining in a Brewery

    The Raw Data set used:

  • 8/8/2019 Artificial Intelligence for Industial Applications

    13/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] or 087-6697867 13 06/11/09

    Data Mining Techniques

    Rule Induction

    Analyse data using probabilistic rule induction. This is a statistical technique toproduce a generalised decision tree from a data set. A decision tree can begenerated by repeatedly splitting the given data set according to differentattributes until terminal points (leafs) are reached based on statisticalevidence of best split possible within certain confidence limits.

    Decision Tree

    The parameters or attributes that most effect an area of interest is at the top ofthe tree.

  • 8/8/2019 Artificial Intelligence for Industial Applications

    14/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] or 087-6697867 14 06/11/09

    Neural Networks

    Neural Networks model a process by first using an example set of datawere the expected output is known for a given set of inputs. The network

    weights are adjusted first randomly then many times until the networkproduces the output required for the inputs.

    Then once trained it can be used to predict future outputs for new inputsat high accuracy.

    Neural Networks provide high degree accuracy at the expense ofunderstandability - the black box solution.

    A three Layer Neural Network

  • 8/8/2019 Artificial Intelligence for Industial Applications

    15/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] or 087-6697867 15 06/11/09

    Combining the techniques

    The power of data mining, the modelling and prediction accuracy of neuralnetworks and the optimisation power of genetic algorithms can all be used toprovide an insight in process plant operations that were not easily obtainablebefore.

    Using data mining, industrial systems and processes can be Mined toobtaining important rules and patterns hidden in the data.

    Neural Networks can be set to work to model the process and providepredictions on process outcomes for different input values.

    Genetic Optimisation techniques can use the information learned in the data

    mining exercise to optimise an aspect of the process such as energy usage ,cost of production or production rates

    The patterns rules extracted during the data mining exercise can also be usedto develop an knowledge based system so that plant operators etc. can usethese rules to obtain an indication of the likely effects of process plantmodifications on performance.

    Example of combining the techniques:

    It is required to determine to design a system based on artificial intelligence

    techniques that would enable a plastic manufacture to optimise the productionof a component while minimising rejects rate and operational costs. He hasseveral machines to choose from all with different running costs and nominalproduction capacities.

    The manufacture has collected historical data on the plant including processdata, quality data, energy, and other costs data.

    This data is first transformed and then mined several times to determinewhat factors effect different areas of Interest and to determine if any patternrules exist. This provides a better understanding of say the factors that effect

    quality, running costs and output. Using these cost function is established toinclude into a genetic algorithms to optimise plant running. A model of theplant is built up using a NN system to show how the output may vary if aprocess parameter is varied. The manufacture how has the tools to optimisehis production and minimise cost.

  • 8/8/2019 Artificial Intelligence for Industial Applications

    16/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] or 087-6697867 16 06/11/09

    Intelligent Sensors

    Intelligent Sensors are different then ordinary sensors in that they have in -

    built computers that can condition the data they receive to provide moreaccurate sensor readings, hence provide for better process control. Theadvantages of intelligent sensors are:

    They can linearise , average and smooth readings from the sensingelement

    They can correct and compensate for changes in other variables such aspressure and temperature

    Store information in relation to services and calibrations etc.

    Carry out internal checks

    The benefits of using Intelligent Sensors in your process include:

    Reduce process variability

    Finer control of processes

    Improved product quality

    Reduce running cost due to finer control

    Reduce downtime due to process failure caused by inaccurate sensormeasurements

    Validate sensor networks for faulty sensors.

  • 8/8/2019 Artificial Intelligence for Industial Applications

    17/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] or 087-6697867 17 06/11/09

    Soft Sensors

    Soft Sensors use neural network modelling techniques to infer values thatcannot be measured directly or is uneconomical to measure by conventionalmeans.

    Soft Sensor Example

    A pharmaceutical Manufacture used such as system to provide an indicationof emission levels based on process parameters. This was done by miningprocess past data and emission levels and producing a neural network thatpredicted to 98% accuracy emission levels based on existing plantparameters. So good was the predictions of the predictions that the EPAaccepted this method over expensive field testing and provided better

    availability.

    Soft Sensor Example

    You cant measure paper quality directly. There is usually insufficient

    knowledge to develop theoretical models and simple linear regression can nothandle the non-linearity of the paper making processes

    A neural network soft sensor was developed by data mining of past processperformance to determine what constituted good quality.

  • 8/8/2019 Artificial Intelligence for Industial Applications

    18/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] or 087-6697867 18 06/11/09

    Intelligent Soft Sensors - Examples

    Chemicals:

    Predictive emission monitoring Ph Prediction in chemical manufacture

    Flowability of powder

    Colour

    Moisture Bubble Size

    Food and Pharmaceuticals

    Odour

    Flavour Fermentation Rate

    Bubble size

    Pulp and Paper

    Predictive emission monitoring

    Moisture

    Colour

    Viscosity

    Industrial Utilities

    Predictive emission monitoring

    Ph

    BOD

    COD

    Conductivity

  • 8/8/2019 Artificial Intelligence for Industial Applications

    19/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] or 087-6697867 19 06/11/09

    Supervisory Control

    Artificial Intelligence can be used to provide supervisory type control in anopen loop configuration to a production process or manufacturing system.

    Using advanced reasoning and based on neural networks combined withstatistical analysis such systems can help operations in a variety of ways suchas:

    Detect key features in process trends

    Recognising patterns over time

    Filter out important information

    Manage information flow to operators

    The net benefits of such systems include:

    Help operators resolve production problems such as:

    Debottlenecking production operations

    Ensuring good production logistics

    Managing of resources such as personnel and energy

  • 8/8/2019 Artificial Intelligence for Industial Applications

    20/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] or 087-6697867 20 06/11/09

    Advanced Control

    Artificial Intelligence can be used to provide intelligent real time control ofcomplex production processes using artificial neural networks that can learn

    from previous data or on-line data:

    Process data can be either achieved historical data or on-line data that hasbeen gathered into training sets. Over time, the training set accumulatesexperience about the behaviour of the process during both normal andabnormal operation. The resulting training set can, over time build a neural setmodel that can identify dynamic, non - linear relationships in the data set,hence identify process interactions that the process specialist can find itdifficult to describe analytically or with a set of rules.

    The net result is a system that can:

    Detect non-linear relationships in process operations

    Detect shifts in process performances

    Determine the significance of individual inputs

    Derive probability estimates and other statistics about processperformance

    Validate Process Sensors

    Validate the performance of machinery

    Fault Classification

  • 8/8/2019 Artificial Intelligence for Industial Applications

    21/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] or 087-6697867 21 06/11/09

    Summary of the benefits of using Neural Networks forAdvanced Control Applications

    Neuro-fuzzy solutions can easily be implemented for control of non-linear systems; performance levels comparable with those obtained bymeans of classic solutions can be reached even with a poor knowledgeof the system transfer function.

    Intrinsic characteristics of neural networks are particularly attractive inthose cases where a mathematical model of the plant behaviour is notavailable or when modifications of the system transfer, functions areexpected during the system lifetime.

    Neural network controllers can easily implement fuzzy controls withadvantages when dealing with non-linear or poorly defined problems;their computing power seems to match demanding situations related tocomplex real-time problems.

    Robustness of neural controls can be verified against system parametervariations, which may be a critical factor for some applications.

    One of the most sophisticated uses of neural networks in processengineering is set point optimisation. Neural network models can beused in combination with neural optimisation techniques to advise onbest control settings to, for example reduce energy usage, increaseproduct yield (reduce SEC).

  • 8/8/2019 Artificial Intelligence for Industial Applications

    22/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] or 087-6697867 22 06/11/09

    Monitoring and diagnostic Analysis

    A few years ago an automatic control system with a graphical interface wassufficient to give you a competitive edge. These systems control the processwithin particular parameters but lacked intelligence to turn the vast quantity ofprocess data collected into useful operator information that can iron outprocess variation and minimise costly disruptions thus providing for consistentprocess performance and product quality. Diagnostics involves identifying oneor more causes to a problem out of a large number of possibilities.

    Intelligent Diagnostics can help you make the right decisions by analysingproduction and process data the very only neural networks can. They provideintelligent decision support the way your SCADA and MMI cannot.

    Intelligent Diagnostic systems can provide the following benefits overthe more conventional type approaches:

    Current Situation Improvement Possible WithIntelligent Diagnostics

    Benefits

    Simple Alarms Filter and Correlate alarmsthat separate cause fromeffect

    Focused attention ofoperators , fewer errors

    Manually executed On - line support for executionof operating procedures

    More consistent operations

    Programmed setpointchanges

    Setpoint changes based online diagnosis of process se

    Higher yield andproductivity

    Traditional, Off linestatistical process control

    On-Line SPC coupled withfault diagnosis

    Improved product quality

  • 8/8/2019 Artificial Intelligence for Industial Applications

    23/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] or 087-6697867 23 06/11/09

    Optimisation using Genetic Algorithms

    Traditional Optimisation approaches

    Software applications fall into two categories; analysis and synthesis. Analysisapplications are represented by the traditional input / output model of dataprocessing whereby input data is processed procedurally or heuristically togenerate the output data. Synthesis applications involve the reverse processof deriving the input data required to generate certain desired outputs. This isa difficult task since there are, in most cases, no formulae or rules to deriveinputs from outputs. This is further complicated by constraints imposed on theacceptable values of input data. Optimisation is the process of deriving valuesof input data that satisfy constraints and which results in the desired outputdata. It can be a very difficult task especially in areas were the attributes aremany and the data set incomplete.

    Optimisation problems can be solved by iterative trial and error wherebydifferent combinations of input values are tried in an attempt to arrive at thedesired output value. However, as the number of possible combinationsgrows, it can become impractical to try all combinations to arrive at a solutionin a reasonable time. The problem becomes one of searching a massivespace of solutions looking for the optimal solution.

    For example in the problem of optimising the order with which to manufacture12 products there are (over 40,000,000) possible sequences to consider.Rules of thumb can be used to narrow down the combinations worthconsidering. However, in most cases, good rules are either not available or itis difficult to capture the rule based trial and error strategy of experts.

    Numerical optimisation techniques have been used to solve optimisationproblems and are now available in most advanced spreadsheet programs.These techniques have the following limitations:

    They lend themselves to optimising independent numeric inputs fromwhich a desired output is calculated. They are less capable of optimisingproblems involving sequencing or scheduling.

    They are "exploitation" and not "exploration" techniques. This meansthat given a reasonable starting solution (a set of input values), thenumeric optimisation will converge to a near optimal solution. However,they are not capable of exploring areas of space where good solutionsexist. This is because numeric optimisation techniques can often gettrapped in local optimal solutions.

    They are not suitable if the outcome cannot be explicitly calculated. Forexample, when the outcome is a subjective assessment by an expert oran observed performance.

  • 8/8/2019 Artificial Intelligence for Industial Applications

    24/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] or 087-6697867 24 06/11/09

    Genetic Algorithms to the rescue

    Genetic Algorithms (GA) are techniques for solving optimisation problemsinspired by the theory of evolution and biogenetics. These algorithms are

    excellent at exploring large spaces for optimal or near optimal solutions. Thebasics of a genetic algorithm are:

    Representing possible solutions to problems as a string of parameters(numbers). This string is called a chromosome and the parameterswithin it are called genes.

    Randomly creating a number (generation) of these chromosomes.

    Calculating the effectiveness of each chromosome as a solution to the

    problem then ranking the chromosomes in order of effectiveness (fitnessto survive).

    Repeating steps 3 to 4 for a number of cycles (generations).

    The randomness of the above process allows the effective exploration of thespace of solutions. While the selection of effective solutions (chromosomes)and the mixing of their genes allows the accumulation of good features frompartially good solutions. Therefore, genetic algorithms can explore largedomains and converge on good solutions relatively quickly. GA's also give apowerful trade off between the time taken to reach a solution and the quality

    of the solution. Typically, it is desirable to achieve a "good" solution in a shorttime as opposed to an optimal solution that may require infinite time.

    Genetic Algorithms can out-perform traditional mathematical approaches byfinding better more accurate solutions in fastest times, and without a team ofstatisticians and computer scientists that may be required to solve some of thetypical optimisation problems found in many industries today.

    Typical Optimisation problems in Industry:

    Optimise production

    Minimise energy

    Minimise waste

    Minimise Labour

  • 8/8/2019 Artificial Intelligence for Industial Applications

    25/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] or 087-6697867 25 06/11/09

    Scheduling using Artificial Intelligence

    The object of scheduling is to order activities in time taking into accountconstraints that are imposed on the system.

    Scheduling using Artificial Intelligence can be a very powerful tool and it findsapplications in many industrial areas such as:

    Production Scheduling

    Material Logistics

    Resource Management

    Activity Reviews

    Process Re-Engineering

    Scheduling software based on artificial intelligence can represent dynamicevents and respond based on real time information to process needs.

    Examples of scheduling problems that have been managed with AI include:

    Scheduling production of many different brands of product based ondistribution requirements

    Managing electrical demand by scheduling plant running order

  • 8/8/2019 Artificial Intelligence for Industial Applications

    26/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] or 087-6697867 26 06/11/09

    Process Modelling and Simulation

    Process Modelling and Simulation can provide many benefits over trial anderror real time operations which can both be costly and even dangerous.

    Process modelling is possible using mathematical simulations of plant if theprocess dynamics are fully understood and the equations that define the inputand output variables are know and are well defined. However, manyprocesses are very complex and difficult if not impossible to obtain reasonablemathematical representation of the process dynamics.

    Neural networks can model a process using previous history data of theprocess, which in essence represents the past behaviour of the process.Hence neural networks can model the actual response of a process not justthe theoretical response which may be based on nominal design data.

    Artificial Intelligence provides the answer by analysing previous process datausing neural networks; it is possible to Predict the process behaviour for agiven set of conditions.

    Combining data mining, neural networks, process optimisation and diagnosticprocedures can result in a powerful Intelligent system that can provide acompany with the competitive edge.

    Hence, process modelling can be used in many industrial processes to:

    Perform what if analysis to determine the impact of proposed processchanges in a particular process

    Determine the optimum mix of utility plant to minimise energy costs

    Optimise yields in processes

    Provide alternatives to full scale experiments to determine effects ofparticular process changes

    Provide means of determine safety issues relating to processmodifications.

    Determine environmental impact of process changes

    Help in process re-engineering issues

  • 8/8/2019 Artificial Intelligence for Industial Applications

    27/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] or 087-6697867 27 06/11/09

    Knowledge Based systemsKnowledge Based Systems (KBS) have emerged over the last five years asthe most practical implementation of Artificial Intelligence. Practical because

    they have been used to deliver business solutions on standard hardwareplatforms. A strict definition of a Knowledge Based System is:

    A computer program that:

    Contains human knowledge

    Is able to give advice by inferring from this knowledge

    Can justify the advice given

    The knowledge can be maintained independently of the program

    Experts apply knowledge to solve problems. Generally speaking there are twotypes of knowledge; declarative and procedural knowledge. Declarativeknowledge consists of facts, concepts, and relationships in a particulardomain. Procedural knowledge is information regarding the application of thedeclarative knowledge in problem solving.

    Knowledge can be also classified into surface and deep knowledge. Surfaceknowledge combines declarative and procedural knowledge into problem-solving heuristics (rules of thumb) enabling an expert to solve commonproblems in a domain without formal analysis from first principles. Theseheuristic rules are normally learned by experience without an understanding ofthe underlying reasons. On the other hand, deep knowledge consists offundamental knowledge of a domain, including definitions, axioms, generallaws, principles, and causal relationships.

    The objective of deep Knowledge Based Systems is to model complexdomains which humans find difficult to understand. These Knowledge BasedSystems will assist the experts in making decisions and add to their

    knowledge.

    Systems based on surface knowledge are intended to capture the problemsolving and decision-making skills of a human expert, in order to automatethis expertise.

  • 8/8/2019 Artificial Intelligence for Industial Applications

    28/28

    Use of Artificial Intelligence In Industry

    Systems Optimizations LtdE-Mail: [email protected] 391394 087 6697867 28 06/11/09

    Learning From Data:

    Organisations are increasingly facing situations where a Knowledge BasedSystem is required to carry out a certain task where the expert with the

    expertise is no longer available or where the expert has limited expertise.However, organisations may hold a large database of records representingcase history data of previous decision making or data representing the verdictof history.

    Data Mining can be considered as an alternative knowledge learning strategywhen the data in question contains history of past process performance (fromwhich a model of the process can be formed) or history of past decisionmaking from which valuable information concerning the decision makingprocess can be obtained.