Corrosion life prediction of glass fiber reinforced plastics by optimized BP neural network

Abstract

• The annealing particle swarm optimization algorithm is improved. The improved algorithm(im SAPSO) has stronger global search ability. • im SAPSO is used to replace the traditional neural network training method(most of these algorithms are based on gradient descent algorithm, so there will inevitably be local extremum problems), the effect is better. • The optimized neural network is used to predict the corrosion life prediction of glass fiber reinforced plastics. The output results of the network are in good agreement with the measured results, which proves the reliability of the algorithm. The retention rate of bending strength (RRBS) of glass fiber reinforced plastics (GFRP) in service under corrosive conditions is responsible for structural safety. However, it is very difficult to measure the RRBS by conventional measurement methods. In order to predict the service life of GFRP under corrosive conditions, a back-propagation (BP) neural network was establised. A series of experiments were carried out, considering three factors including temperature, time and corrosive medium concentration. 60 groups of RRBS were obtained and used as samples for neural network training. The network was optimized by improved simulated annealing particle swarm optimization algorithm (imSAPSO). The optimized network was then used to predict the other six groups of test data. The results show that the predicted values compare well to the measured, with maximum relative error of −0.615417% and minimum 0.015934%, and standard deviation of 0.00254567.