Improved BP-neural Network of the Particle Swarm Optimization in the Research on Engine Fault Diagnosis

Abstract:

In the process of using BP-neural network in fault diagnosis, there will be “dimension tragedy” as the input variable increases, which causes the lower training effective. Besides, traditional BP algorithm tends to fall in local optimization. The reduction based on the rough set (RS) is the conventional “reduce dimension” method, but it is NP-hard problem, whose computing will gradually augment as the information increases. Therefore, a heuristic algorithm was used for attribute reduction based on the importance of attribute value to reduce attribute, a fault diagnosis approach was formed combining the fuzzy information system knowledge method with BP-neural network of the particle swarm optimization (PSO) algorithm to diagnose the fault of engine. The experiments show that comparing with the conventional method, it can not only require fault diagnosis rule, but also reduce net input dimensions effectively, avoid falling in local optimization and increase the efficiency of fault diagnosis.