Genetic Programming for Modeling Vibratory Finishing Process: Role of Experimental Designs and Fitness Functions

Abstract

Manufacturers seek to improve efficiency of vibratory finishing process while meeting increasingly stringent cost and product requirements. To serve this purpose, mathematical models have been formulated using soft computing methods such as artificial neural network and genetic programming (GP). Among these methods, GP evolves model structure and its coefficients automatically. There is extensive literature on ways to improve the performance of GP but less attention has been paid to the selection of appropriate experimental designs and fitness functions. The evolution of fitter models depends on the experimental design used to sample the problem (system) domain, as well as on the appropriate fitness function used for improving the evolutionary search. This paper presents quantitative analysis of two experimental designs and four fitness functions used in GP for the modeling of vibratory finishing process. The results conclude that fitness function SRM and PRESS evolves GP models of higher generalization ability, which may then be deployed by experts for optimization of the finishing process.