Optimization of backpropagation neural network models for reliability forecasting using the boxing match algorithm: electro-mechanical case

Abstract

Presenting a robust intelligent model capable of making accurate reliability forecasts has been an attractive topic to most industries. This study mainly aims to develop an approach by utilizing back propagation neural network (BPNN) to predict the reliability of engineering systems, such as industrial robot systems and turbochargers, with reasonable computing speed and high accuracy. Boxing Match Algorithm (BMA), as an evolutionary meta-heuristic algorithm with a new weight update strategy, is proposed to bring about performance improvements of the ANN in reliability forecast. Consequently, the hybrid model of BMA-BPNN has been provided to gain a significant level of accuracy in optimizing the weight and bias of BPNN using three sets of function approximation data to benchmark the proposed approach's performance. Then, the BMA is utilized to improve reliability forecasting accuracy in engineering problems. The obtained results reveal that the presented algorithm delivers exceptional performance in function approximation, and its performance in forecasting engineering systems' reliability is about 20% better than further compared algorithms. Similarly, rapid convergence rate, reasonable computing time, and well-performing are additional characteristics of the presented algorithm. Given the BMA-BPNN characteristics and the acquired findings, we can conclude that the proposed algorithm can be applicable in forecasting engineering problems' reliability.