Hierarchical genetic neural network for fault diagnosis

A genetic neural network was developed for fault diagnosis in rotating machinery. A new hierarchical structure is presented using a neural network learning algorithm which combines a genetic algorithm and evolutionary programming. The shooting method is used to optimize the network structure and train the connection weights. Rotating-machinely fault-classification data was used to compare the shooting method and the traditional backwards propagation (BP) algorithm. The result proves that the hierarchical genetic neural network converges faster than the BP training algorithm, that it avoids falling into local minima, and that it provides a much simpler classification neural network for the faults.