The nonlinear multi-response grinding process is too difficult to optimize due to a large number of interacting process variables. Conventional experimentation techniques such as factorial designs, evolutionary operations, response surface methodology (RSM), and the Taguchi method may not be implementable or economical for many types of mass production lines involving grinding, boring, turning and other necessary operations. Particular, for the grinding processes, in the absence of a reliable and generalized mechanistic (analytical) model that is applicable in varied situations, researchers and practitioners in general prefer empirical modeling (static or dynamic) technique(s) to understand the inherent complex characteristics of the grinding processes. In this context, a simple and easy-to-implement modeling and optimization technique for grinding processes in mass production lines is a prime necessity. The potential of the nonlinear multivariate Artificial Neural Network (ANN) technique and metaheuristic search strategies needs to be explored, either in their original form or in the form of their variants. In this paper, an integrated approach of ANN, the composite desirability function, and the metaheuristic strategy is proposed for modeling and optimizing the parameters of the grinding processes. Independent computational run results based on two different metaheuristicsreal-valued genetic algorithm (RGA) and simulated annealing (SA)show that RGA is more efficient to determine multiple near-optimal (approximate) solutions [in terms of sample mean of the overall fitness (objective) measure], that is less likely to be trapped in local optimal, and which requires comparatively less computational time as compared to SA.
In today’s manufacturing context, there is a growing need of hard and tough materials that can withstand varying stress conditions to ensure prolonged service life of parts. The need to economically process these materials so as to meet stringent product
quality requirements has become a real challenge for a manufacturing unit. Grinding,
being a primary metal-cutting operation, has been considered to be an accurate and
economical means of shaping the parts into the final products with the required surface
finish and precise dimensionality.
In many mass manufacturing lines, systematic design of experiment based on
one time or sequential experimentations (such as factorial, fractional factorial,
evolutionary operations, response surface design methodology, and Taguchi
method) are too difficult to implement, if not impossible. Laboratory-type offline
experimentation in such situations can be uneconomical, both in terms of cost
and time, and lead to delay or stopping of the entire manufacturing lines. Moreover,
discrete levels or factors selection for experimentations are generally based on
expert(s) knowledge or perception, and may be biased towards certain procedures,
certain operators or may not include the effects of actual manufacturing
environments [6]. |