Comparing the performance of neural networks developed by using Levenberg–Marquardt and Quasi-Newton with the gradient descent algorithm for modelling a multiple response grinding process

Abstract

Monitoring and control of multiple process quality characteristics (responses) in grinding plays a critical role in precision parts manufacturing industries. Precise and accurate mathematical modelling of multiple response process behaviour holds the key for a better quality product with minimum variability in the process. Artificial neural network (ANN)-based nonlinear grinding process model using backpropagation weight adjustment algorithm (BPNN) is used extensively by researchers and practitioners. However, suitability and systematic approach to implement Levenberg–Marquardt (L–M) and Boyden, Fletcher, Goldfarb and Shanno (BFGS) update Quasi-Newton (Q-N) algorithm for modelling and control of grinding process is seldom explored. This paper provides L–M and BFGS algorithm-based BPNN models for grinding process, and verified their effectiveness by using a real life industrial situation. Based on the real life data, the performance of L–M and BFGS update Q-N are compared with an adaptive learning (A-L) and gradient descent algorithm-based BPNN model. The results clearly indicate that L–M and BFGS-based networks converge faster and can predict the nonlinear behaviour of multiple response grinding process with same level of accuracy as A-L based network.