Based On Support Vector Machines Fault Diagnosis Methods For Complex Product Manufacturing Processes

The thesis addresses the problems of fault diagnosis in complex product manufacturing process using the data of TE process and automotive parts machining processes as experimental data source.It combines data-driven methods such as independent component analysis,support vector machine and particle swarm optimization algorithm,to build more effective models of fault diagnosis in high dimension processes and multistage processes.This study is helpful for troubleshooting and maintenance in time,improving the product quality.It is very important to enhance the enterprise's competition ability.The main contents of this thesis include:Based on FastICA fault feature extraction method.The Research focuses on feature extraction method in high dimension manufacturing process,which can extract potential variable characteristics from the strong correlation measurement variables.As an important data pre-processing technology,it can effectively enhance the high dimension manufacturing process fault diagnosis ability.Independent components need to determine in advance in the FastICA algorithm.This thesis takes use of a automatic selection criteria based on negative entropy threshold to deal with this problem.Based on support vector machine(SVM)fault diagnosis method has advantage on over-fitting problem and the small-samples cases.For improving the classifier's performance,extraction feature is required.The fast independent component analysis(FastICA)is applied for feature extraction.The parameters of SVM are optimized by particle swarm optimization(PSO).In this thesis,an integrated framework of FastICA and PSO-SVM algorithm for fault diagnosis is presented.Compared with other predictors,this model has greater generality ability and higher accuracy.Based on optimized SVM fault diagnosis mothed is researched in multistage manufacturing process.The quality of the final product or service produced by a multistage system is determined by complex interactions among multiple stages.It is challenging to accumulate the process error and its propagation along a series of stages.The state space model for a multistage system is introduced.Combined with the idea of pattern matching,the thesis takes the pattern vectors as the input of the fault classifier instead of measurement vectors.The results of experiment show that the model can effectively diagnose faults in multistage manufacturing processes.