Study On Technologies Of Fuzzy Pattern Recognition And The Applications In Fault Diagnosis

This dissertation mainly studies fuzzy pattern recognition and its application to fault diagnosis in systems.To begin with, the research history, the current situation and development trend of fault diagnosis and intelligent information processing techniques are described. Then, according to the connatural fuzziness in the field of fault diagnosis and the widely used of pattern recognition technologies in fault diagnosis, this dissertation focuses on the investigation and improvement of fuzzy pattern recognition. We develop our research on fuzzy pattern recognition from two aspects: Firstly, the fuzzy C-means clustering algorithm based on kernel function (FKCM) is proposed. With the development of science and technology, the systems in industrial process become more and more complex. It is more and more difficult to acquire accurate fault information. It is not very clear of the difference among data which belong to different class. If the data were cross in the origin space, it is difficult for classical fuzzy C-means clustering (FCM) algorithm to clustering correctly. By using kernel technique, the data in the original space are mapped to a high-dimensional feature space. The data are optimized in feature space and thus the difference among data is expanded. Performing FCM clustering algorithm in feature space, the data, whose difference is small, can be clustered accurately. Secondly, fuzzy support vector machine (FSVM) is proposed. Support vector machine (SVM) is a powerful classification machine in pattern recognition, especially when data are not enough. So SVM can solve the problems in the field of fault diagnosis, when it is difficulty to acquire fault data. Because of the connatural fuzziness the fuzzy theory is incorporated into SVM, a fuzzy parameter is introduced into SVM, thus the SVM becomes the FSVM. FSVM adequately takes advantage of two theories. To confirm the membership reasonably, the kernel hyper-ball based on kernel technologies is presented. It can evaluate reasonably the effect of every data.