Research On The Techniques Of Intelligent Diagnosis & Condition Prediction For Manufacturing Equipment

This dissertation concentrates on the study of the basic theory, methods andapplication of manufacturing equipment fault diagnosis, condition monitoring andpredicting techniques, combined closely with the practice of equipment managementand predictive maintenance. Especially, the techniques and methods of fault diagnosisbased on neural network adaptive pattern recognition, feature extraction using waveletand waveletpackets analysis, condition prediction methods based on the theories ofgray system and neural network are studied systematically. The main achievementsare as follows:In the study of neural network adaptive pattern recognition techniques:A diagnostic model of functional-link net is presented. Using this model, someshortcomings of BP model, such as low rate of convergence and local minima, areovercome. Additional, the structure of this model is simpler and its algorithm iseasier.A fuzzy-neural network diagnostic model is built by combining neural networktheory with fuzzy theory. In this model, the training samples can represent eachpattern better. Because of replacement of two-value mode with Membership values,condition of equipment can be described more exactly.By applying fuzzy theory to functional-link net model, a fuzzy-extensionenhancement model, which combines the advantages of the two models mentionedabove, is presented.Effectiveness and practicality of the three diagnostic models above are verifiedby practical examples.In the study of the wavelet transformation theory, including algorithms ofdecomposition and reconstruction for wavelet and waveletpackets, some signalextracting methods based on wavelet and waveletpackets are presented, then severalinstances are given to verify them.In the study of mechanical equipment condition monitoring and predicting:Standards for condition judgement are summarized, and several methods ofbuild-up for threshold of fault-alarm are presented.Predictive models based on grey theory and based on neural networks are builtand verified using computer-simulated data. Predictive modeling problems about timeseries in random interval, which is very common in equipment condition monitoring,are researched, furthermore, a modeling method for such a time series based onfunctional-link net model is presented and verified using computer-simulated data. Inaddition, this method can reduce noise in time series, so it can be used to improveaccuracy in any predictive method.On the above theoretical research work, Intelligent Diagnosis and PredictiveMaintenance System (for Windows'95), a practical tool with functions of equipmentmanagement, dynamic data management, data collecting, data analysis, faultdiagnosis, condition monitoring and predicting, etc., is designed and developed. Thenew fault diagnostic and predictive methods are proved effective and satisfactoryresults have achieved in several practical cases.