Technical Studies, Remote Fault Diagnosis Based On Self-organizing Neural Network

The technology of remote fault diagnosis is very important in equipment management and in support of fault maintenance. It can discover and deal with the fault of equipment immediately; it can also improve the work-efficiency of machine, and reduce the cost of maintenance in order to avoid the phenomenon of diseconomy and irrationality caused by "excess servicing". Fault diagnosis requires more intelligence than before because the large, complicated, automatic, intelligent, mechanotronics equipment has appeared. Neural networks (NNs), which are a typical representation of computational intelligence, provide a powerful way to diagnose remote faults. The aim of this paper is to improve the recognition rate of fault pattern. The research on the technology of remote fault diagnosis is based on the characteristics of Self-Organizing Map (SOM) including parallelism, fault-tolerance, and self-organizing learning, and is based on the technology of information integration, and follows the data processing in fault diagnosis. The main contents of the dissertation are as follows:1. Remote fault diagnosis based on Multilayer SOM and data preprocessing. The noise of fault data is removed by mean; PCA and ICA are adopted to deal with dimension reduction, the effect of which is compared and analyzed; multilayer SOM is used to converge mode clustering regions so as to improve the recognition rate of fault pattern.2. Pattern recognition method of Remote Fault Diagnosis based on synchronous enlargement of Self-Organizing Feature Map. The input vectors are processed by SOM NNs to search for clustering regions, which extracts their central seeds from SOM. The synchronous enlargement of the central seeds partitions mode clustering regions to recognize the fault patterns.3. Pattern recognition method of Remote Fault Diagnosis based on gravitation field of Self-Organizing Feature Map. The input vectors are processed by SOM to search for clustering regions, which extracts their central seeds from SOM. The fault pattern is recognized according to the convergence of each point in the gravitation field with those central seeds as its gravitation origin. 4. Ensemble Neural Network based on Analytical Hierarchy Process (AHP). According to the basic principle of the information fusion, measurements between input fault mode and typical failure mode are calculated to build pair-wise comparison matrices. AHP is used to integrate the results which are identified by different neural network subsystem. AHP can improve the recognition rate of fault pattern.Emulators are built up through Matlab to deal with the fault data of a certain airplane's undercarriage, which proves effective.