Generalized approach for multi-response machining process optimization using machine learning and evolutionary algorithms

Abstract

Contemporary manufacturing processes are substantially complex due to the involvement of a sizable number of correlated process variables. Uncovering the correlations among these variables would be the most demanding task in this scenario, which require exclusive tools and techniques. Data-driven surrogate-assisted optimization is an ideal modeling approach, which eliminates the necessity of resource driven mathematical or simulation paradigms for the manufacturing process optimization. In this paper, a data-driven evolutionary algorithm is introduced, which is based on the improved Non-dominated Sorting Genetic Algorithm (NSGA-III). For objective approximation, the Gaussian Kernel Regression is selected. The multi-response manufacturing process data are employed to train this model. The proposed data-driven approach is generic, which could be evaluated for any type of manufacturing process. In order to verify the proposed methodology, a comprehensive number of cases are considered from the past literature. The proposed data-driven NSGA-III is compared with the Multi-Objective Evolutionary Algorithm based on Decomposition (MOEA/D) and shown to attain improved solutions within the imposed boundary conditions. Both the algorithms are shown to perform well using statistical analysis. The obtained results could be utilized to improve the machining conditions and performances. The novelty of this research is twofold, first, the surrogate-assisted NSGA III is implemented and second, the proposed approach is adopted for the multi-response manufacturing process optimization.

Keywords

• Machining process optimization
• Data-driven surrogate model
• NSGA-III
• Many-response parametric design
• MOEA/D