Analysis and Prediction of Grind-Hardening Surface Roughness Based on Response Surface Methodology-BP Neural Network

Abstract

Surface morphology and surface roughness are very important properties used to assess the quality of grind-hardening surfaces. In this study, grind-hardening tests for 42CrMo steel were designed using the response surface methodology to reveal the surface morphological characteristics of the grind-hardening surface and the effects of grinding parameters on its roughness. The results showed considerable grinding damage in both the cutting-in and cutting-out areas of the grind-hardened surface, while the middle area was more stable. More specifically, the cutting-in area showed much bonding and damage, while the cutting-out area showed more microcracks. Under the conditions of this test, the surface roughness tended to increase with the increase in cutting depth and workpiece feed speed. The effect of grinding line speed on the grind-hardening surface roughness was not significant. The significance of the effects of grinding parameters on surface roughness ranked as: cutting depth > workpiece feed speed > grinding speed. In turn, a response surface methodology-BP neural network prediction model for the surface roughness of grind-hardening was developed, whose feasibility and validity were confirmed by the experimental results. The model achieved surface roughness prediction of the grind-hardening process with a mean relative error of 2.86%.