Optimization of Machining Processes Parameters

Abstract

Optimization of machining parameters is irreplaceable in modern productions since it is essential not only to achieve a high level of precision but also to increase productivity. The inherently bounded and non-linear nature of machining processes, their combining integer, discrete and continuum variables, further complicates this effort. Secondly, mathematical formulations of these systems are often discontinuous, implicit, or characterized by a non-differentiable (with regard to the design variables) nature. As a result, gradient-based non- linear optimization methods have rather limited usage. Genetic Algorithms (GA) have thus become an interesting choice, which can easily deal with complex and highly non-linear optimization issues of machining. In comparison with the other traditional approaches, GAs operate on a population of possible solutions and use both stochastic and deterministic search strategies to guide the search to solutions that meet feasible solution criteria and optimal solutions. Despite the strengths that GAs provide there are still some limitations that are inherent in the structure of the algorithm: (1) inefficiency of the way they encode continuous variables using ones and zeros, (2) lack of local-search functionality, (3) lack of self-adaptation and (4) poor handling of conveys. The article describes a new, structured evolutionary algorithm based on canonical genetic algorithms that was specifically designed to overcome the problems detected in current implementations. Discriminating improvements encompass a more efficient characterization of the nature of the problem-issues, the incorporation of selection procedures with an eye to population-level goals, specific genetic operators appropriate to variable groupings, competent constraint management measures and judicious initial population-setting regimes. Its efficacy and efficiency as provided by the proposed framework are empirically demonstrated with the help of two machining case studies, in which the framework proves better than the other established ones in terms of efficacy and efficiency.