Research Of Fault Detection And Diagnosis Based On PCA And K-NN

With the scale of modern industrial production expanding, the number of industrial equipment is growing and more equipment tends to be more intelligent. The industry system is gradually developed into intelligent manufacturing system, which makes the whole industrial process more complex. Thus, a wide range of fault is difficult to avoid. If fault occurs while this complex industry system is running, it will affect the normal operation, seriously, it may lead to the collapse of the system. Therefore, it is an indispensable part of the modern industrial process to give real-time monitoring, make timely detection and diagnosis and judge the cause of the fault accurately. This will provide strong guarantee for the secure and stable operation of the entire industry system.Previous study objects on fault detection and diagnosis generally focus on the large complex equipment, mainly research in fault detection and the diagnosis of fault cause, or only research in the fault classification. However, with the development of information technology, all the equipment and systems involved in industrial processes form a large and complex system through the network. In this complex system, the types of fault increase significantly, more variables are involved in, fault detection and diagnosis become more difficult. Only to detect and diagnose the fault has been unable to provide sufficient reference for further fault solving. There is a need to research in fault detection, fault diagnosis and fault type identification systematically, making the process of fault detection and diagnosis unified and complete. Therefore, for the increasingly complex industrial system, studying on how to achieve fault type recognition effectively is very necessary. The introduction of fault type recognition on the basis of fault detection and fault diagnosis can make the process of fault detection and fault diagnosis better, providing a valid reference and strong basis for further fault solving.On the basis of deep analysis of PCA and k-NN, this paper makes systematic research in fault detection based on PCA, fault cause diagnosis based on contribution plot and fault type recognition based on k-NN. In complex industrial processes, making fault detection and diagnosis for the equipment always needs to consider all the related variables of the entire production process, complex industrial systems formed by network involves more variables, which creates a larger difficulty for statistical analysis. Principal component analysis(PCA), as a multivariate statistical analysis method, can efficiently identify the most "major" structure of variables in data, remove noises and redundancy, reduce the dimension of the original complex data and find the brief structure hidden in the complex data, which greatly simplify the entire analysis process.Therefore, this paper first uses PCA to detect fault. After finding the fault, using the contribution plot to diagnose fault causes, identifying the specific data variable which causes the fault. Then, introducing the k-nearest neighbor(k-NN), a method of machine learning, which is widely used in automatic text classification because of its high classification accuracy, in the field of fault type recognition. Making k-NN as the basis for the research in fault type identification of complex industrial systems. For the main problems of miscarriage of justice due to uneven distribution of samples, influence of classification accuracy because of k value in k-NN, on the basis of an optimization algorithm LMC, this paper proposes N-LMC. This method uses the cosine distance which focused more on the difference of the dimension to measure the similarity between samples, acquiring better classification results than LMC. The N-LMC is more suitable for multidimensional data classification in industrial process. Then, for the fault data of large amount and complexity, this paper first uses PCA to reduce the dimension of fault data set, then applies N-LMC in identifying the fault type of fault data.Finally, achieving fault detection and fault diagnosis in specific data set by MATLAB simulation and testing the fault type recognition based on the N-LMC proposed in this paper. The results show that the classification accuracy of N-LMC is higher than k-NN and LMC. And it is more suitable for classification of complex multidimensional data. Then verifying the validity of the proposed method that applied in identifying the fault type of fault data after combined with PCA. Compared with the existing method, LMC、NBC and SVM, the results show that the method proposed in this paper is simpler, the recognition speed is faster, the accuracy is relatively higher and the overall performance of type recognition is better. It can provide a valid reference and strong basis for further fault solving.