Residual stress prediction in hard machining: A comparative study of ANN, ANFIS, SVM and GPR models

In the present investigation, techniques based on learning are applied to predict longitudinal and circumferential residual stresses during the hard turning of AISI 52100 steel by a CBN cutting tool. Residual stresses are one of the most commonly variables which evaluate the machined surface integrity. Predicting this last is a major objective related to the quality and life of manufactured products. In this context, we use four models to estimate residual stresses: Artificial Neural Network (ANN), Adaptive Neuro-Fuzzy Inference System (ANFIS), Support Vector Machine (SVM), and Gaussian Process Regression (GPR). The analysis is based on experimental data structured in 34 combinations using work material J-C rheological properties (A, B and n) and cutting parameters (Vc, f and ap). These rheological properties are related to the hardness and microstructure, which depend respectively on the heat treatment and carbide inclusion. For the developed models, ANFIS gives globally the best performances, achieving high value of R² and minimal MSE; it shows the most promise of prediction. This underscores the effectiveness of learning techniques in estimating residual stresses.