Surface roughness prediction of AISI D2 tool steel during powder mixed EDM using supervised machine learning

Abstract

Abstract Surface integrity is one of the key elements used to judge the quality of machined surfaces, and surface roughness is one such quality parameter that determines the pass level of the machined product. In the present study, AISI D2 steel was machined with electric discharge at different process parameters using Jatropha and EDM oil. Titanium dioxide (TiO2) nanopowder was added to the dielectric to improve surface integrity. Experiments were performed using the one variable at a time (OVAT) approach for EDM oil and Jatropha oil as dielectric media. From the experimental results, it was observed that response trends of surface roughness (SR) using Jatropha oil are similar to those of commercially available EDM oil, which proves that Jatropha oil is a technically and operationally feasible dielectric and can be efficiently replaced as dielectric fluid in the EDM process. The lowest value of S.R. (i.e., 4.5 microns) for EDM and Jatropha oil was achieved at current = 9 A, Ton = 30 μs, Toff = 12 μs, and Gap voltage = 50 V. As the values of current and pulse on time increase, the S.R. also increases. Current and pulse-on-time were the most significant parameters affecting S.R. Machine learning methods like linear regression, decision trees, and random forests were used to predict the surface roughness. Random forest modeling is highly accurate, with an R2 value of 0.89 and an MSE of 1.36% among all methods. Random forest models have better predictive capabilities and may be one of the best options for modeling complex EDM processes.

Keywords

• Surface roughness
• Jatropha oil
• Linear regression
• Decision tree
• Random forest