ANN-based fault diagnosis method with a combined BP algorithm

This paper presents a knowledge acquisition approach for an ANN-based fault diagnosis expert system. A hierarchical classification diagnostic model is used in this system for the problems of complicated system. The model is implemented by a multiple structure neural network composed of some subnetworks. A improved backpropagation (BP) learning algorithm combined with conjugate gradient method and adaptive gradient method is used to train these subnetworks independently. The paper discusses the system model, implement method and algorithm improvement, and also gives an example of fault diagnosis of a marine diesel engine.