Machine learning models-based optimization of burnishing operationin terms of power efficiency and surface properties

Abstract

In the current work, process parameters of the roller burnishing process, including the spindle speed (S), feed rate (f), depth of penetration (D), and flow rate of lubricant (Q) are optimized to maximize the power efficiency (PE) as well as micro-hardness (MH) and minimize energy consumption (EC) as well as mean roughness (Rz). Predictive models of burnishing responses are proposed using Gaussian process regression (GPR) and adaptive-network-based fuzzy inference system (ANFIS) approaches. An efficient algorithm entitled multi-objective non-sorted dragonfly algorithm (MONSDA) is developed to produce optimal solutions. The MEREC and VIKOR are utilized to calculate weights and select the best data. The findings presented that the optimal S, f, D, and Q were 1168 rpm, 0.05 mm/z, 0.06 mm, and 4 L/min, respectively. At the optimal point, the PE and MH were enhanced by 39.1% and 13.0%, while the EC and Rz were reduced by 5.1% and 47.3%, respectively. The PE model was significantly affected by the S, D, f, and Q, respectively. The EC model was significantly affected by the f, S, D, and Q, respectively. The Rz model was significantly affected by the f, D, Q, and S, respectively. The MH model was significantly affected by the D, f, Q, and S, respectively. The GPR models presented better precision for the predictive aim, as compared to ANFIS correlations. The GPR-MEREC-MONSDA-VIKOR was a prominent solution to deal with complicated optimization issues, as compared to the conventional one. The outcomes can be applied to enhance energy efficiency and surface properties.