Improved probabilistic neural network PNN and its application to defect recognition in rock bolts

Abstract

The probabilistic neural network (PNN) model is developed based on Bayesian criteria of multivariate pattern classification. Because PNN has the advantages of low training complexity, high stability, quick convergence, and simple construction, it can be well applied in the area of defect recognition. The application of PNN has a bottleneck problem; that is, the selection of smoothing parameters will seriously affect PNN recognition accuracy. In this paper, we propose an improved PNN model that employs a differential evolution algorithm to optimize the smoothing parameters. Furthermore, a defect recognition approach is proposed based on the improved PNN, which is successfully applied to an anchoring testing field. The proposed approach includes two key steps. The first step involves extracting the energy eigenvector of the defect signal with normalization based on a wavelet packet. The second step involves recognizing defects based on the improved PNN. Simulation results show that our improved PNN model is superior to the traditional PNN model. Thus, the improved PNN can provide useful references for recognizing the defect type of rock bolts in engineering.