Research On Process Monitoring Method Based On Blind Source Separation Algorithm

Process monitoring is a very important measure to guarantee the safety and stability of the production.The present production process generates a large amount of data,which provides the basis for process monitoring of data.But there are many redundant information in the large data and too much data will cause the problems such as slow operation and so on.After researching on monitoring method based on blind source separation algorithm,this thesis mainly focuses on extracting the main information from the large scale data by the nature of entropy and giving a new sample selection which is based on the original sample selection method of the entropy.What is more,this thesis aims at the drawbacks of the fixed point algorithm of kurtosis and negative entropy and gives a combining method of self-adaption weighting.The main contents of this thesis are as follows:Firstly this thesis analysis the theory of principal component analysis(PCA),blind source separation and independent component analysis(ICA)as well as the application of process monitoring.According to entropy which can measure the information richness,this thesis discusses the application and necessity of sample selection method in data reduction.By researching a sample selection method based on information entropy and entropy means,it can select small sample data.By utilizing PCA and ICA as well as comparison the simulation results,it can verify the accuracy of new sample selection method of the information entropy and entropy means.Secondly this thesis also studies on the fixed point algorithm of kurtosis and negative entropy to analyze the differences of two algorithms and their advantages and disadvantages.Accentuating the positive and eliminating the negative,what is more,introducing the idea of weight adjustment,this thesis gives a method of combining with kurtosis and negative entropy based on self-adaption weight.By simulation,the correctness of the weights adjustment and the process monitoring performance of the self-adaption weights are verified.By comparing combine method and single method,the method which is based on the self-adaption weights of the kurtosis and negative entropy combination has more extensive applicability.Thirdly in the context of the grinding process,researches the new sample selection method based on information entropy and entropy means.In the simulation,compare with the self-adaption weights of the kurtosis and negative entropy combination and ICA method respectively.This simulation results show the good effectiveness of the method for data reduction and the separation method for self-adaption weights of kurtosis and negative entropy combination.