Conception de modèles prédictifs de performances liées au comportement d’un usinage dur

Abstract

This thesis concerns the design of predictive models for flank wear of a mixed alumina insert during dry longitudinal turning of AISI D2 steel. The developed models utilize ML techniques (ANN, ANFIS, SVM, and GPR), polynomial fitting by a genetic algorithm (GAPOLYFITN), and numerical simulation via the FEM. Upon the obtained results, the GPR technique emerges as the most effective model for predicting flank wear; however, the GAPOLYFITN model is more practical as it consists of a simple mathematical formulation allowing for its integration, for example, into an adaptive numerical control loop aimed at optimizing the machining operation. Furthermore, by combining geometric modeling and numerical simulation with experimental validation, two models were developed for predicting flank wear and its rate via the modified Usui wear model. The geometric modeling allows for the selection of the inclination angle of the wear surface, which will be validated by a 3D numerical simulation of chip formation and a measurement of the flank wear.