Fault Recognition Study Based On Improved Wavelet Neural Network

Wavelet Neural Networks (WNN) is one of the most widely used Neural Network (NN) model. The NN has the merits of self-learning and self-organization, whereas wavelet analysis has the ability of local characteristic, combining their merits, WNN was put forward. Therefore, some improvement of the convergence rate and fault tolerance of the network is achieved through tuning the wavelet basis adaptively. However, WNN based on back-propagation (BP) learning algorithm is easy to fall into local minimum value and the convergence speed is slow. Genetic algorithm (GA) is a global search algorithm. Combining GA with WNN could overcome the drawback of WNN and improve the global search ability of the WNN.The theory of artificial neural network and wavelet analysis is introduced firstly, then the WNN is constructed and the algorithm is deduced. The original parameters of the WNN is obtained by making use of the GA to localize a good region, after this, the gradient descent algorithm is employed to find an optimal solution in that region. Finally, the improved WNN is applied to the fault diagnosis of bearing, gearbox and piston compressor, the results showed that the improved WNN is feasible and effective.