Near Optimal Grinding Process Design Using Neural Network and Real Coded Genetic Algorithm

Abstract

A typical grinding process is an essential manufacturing operation and has been considered to be a precise and economical means of shaping the parts into the final products with required surface finish and high dimensional accuracy. The need to economically process hard and tough materials which can withstand varying stress conditions to ensure prolonged service life of parts has become a real challenge for researchers and practitioners. In this context, with the advance development and automation of grinding processes, use of appropriate modelling and optimization techniques has been continually emphasized. In view different types of end product and process requirements in grinding processes, optimization often becomes non-linear, multiple response constrained problem with multi-modal distribution of response quality characteristics. The objective of this study is to apply back propagation neural network modelling technique for prediction of a computer numeric-controlled (CNC) rough grinding process behaviour, and thereby determine overall near optimal process design using real coded genetic algorithm. The study proposes an integrated approach using back propagation neural network algorithm, composite desirability function, and real-coded genetic algorithm. The effectiveness and suitability of the approach is determined based on data analysis of a single-pass 6-cylinder engine liner CNC rough grinding (honing) operation in a leading automotive manufacturing unit in India.