Research On Industrial Process Fault Classification Based On Hybrid Method

In the actual industrial production process,maintaining the stable and efficient operation of the system is the basis for completing the production tasks safely and on time,and this is inseparable from the state detection and fault diagnosis of the industrial process system.The techniques of fault classification can determine the fault category and is helpful to locate the fault.It is an extremely important step in the fault diagnosis process.However,as the functions and structures of the industrial processes become more diverse and complex,the fault classification tasks become more difficulties.It is difficult to complete the fault classification task of increasingly complex industrial processes by a single method.This has a very negative impact on the timely completion of fault diagnosis to ensure the safe and stable operation of the system.In order to solve this problem,a hybrid method for fault classification is proposed in this thesis.The main contributions of this thesis are as follows:(1)A fast K-nearest neighbor method based on hybrid feature generation is proposed for fault classification.Firstly,a hybrid feature generation method combining Relief F algorithm and linear discriminant analysis is proposed to improve the quality of the training samples.Then,to address the shortcoming of high computation cost of the K-nearest neighbor classifier,the K-medoids clustering algorithm is used to select a small number of representative training samples to reduce the computational cost.The proposed method not only guarantees the classification accuracy of K-nearest neighbor classifier but also improves the classification efficiency.(2)An improved voting method based on the validity of the classifier is proposed.Although the hybrid fault classification method in(1)can provide higher classification accuracy than a single K-nearest neighbor method,it is still hard to deal with the classification problem of all kinds of the process faults using only one classifier.In view of this,another hybrid method is proposed for fault classification,which is called multi-classifier system.In order to improve the effect of multi-classifier system,an improved voting method is proposed to integrate the classification results of the base classifier more effectively.The proposed method firstly introduces the concept of the classifier validity,and then highlights or suppresses the influence of different base classifiers on voting results according to this concept.The simulation illustrates that the improved voting method has better integration effect than the original majority voting method.(3)An improved voting method based on combination weight is proposed.Considering that the performance of the base classifier in multi-classifier systems is different under different conditions and different performance evaluation indicators,multiple performance evaluation indicators are used to measure the performance of the base classifier and determine the voting weight of the base classifier.To avoid the possible deviation of using only subjective weight or objective weight,a combined weight determination method is proposed.The simulation results show that the proposed improved voting method can effectively integrate the classification results of the base classifier,and the performance of the multi-classifier system is improved.