Pruning Elman neural network and its application in bolt defects classification

Abstract

The Elman model is developed based on BP neural network. Because Elman neural network has the advantages of low training complexity, high stability, quick convergence, and simple construction, it can be well applied in the area of classification. The application of Elman has a bottleneck problem, that is, network structure is difficult to determine. This paper proposed an improved Elman neural network model using pruning algorithm for the network structure optimization. Firstly, we choose the bolt defect parameters which are relatively easy to detect as the classification sample, select the initial large-scale network structure, analysis the contribution of hidden nodes in the network, and reduce the dynamic threshold of hidden nodes in combination with the network to obtain the optimal network structure. Simulation results show that pruning Elman neural network (P-Elman) can adaptively obtain the optimal network structure, and achieve a high classification accuracy, which is much higher than other methods.