Machine learning for process parameter selection in intelligent machining

Abstract

In a product life-cycle, CAD/CAM integration speeds up the design and manufacturing process of a product significantly. Despite the fact that the correct process parameters have significant influence on the quality of the final product, few attempts have been made to incorporate the systematic selection of the process parameters of a metal-cutting operation in a facility-specific manner. It is believed that incorporating the determination of machine-specific process parameters in the process planning phase will help to further increase productivity in the manufacturing phase of the product cycle. In addition, it is believed that the quality of products will improve, and costs for follow-up treatment of parts produced in a cutting process will decrease. Ideally, the evaluation criteria for a machining operation (such as surface finish, accuracy, cutting forces and temperature at tool tip) should be controlled in a closed-loop control system. The controller of such a system would use the deviation of the measured values of the evaluation criteria (as produced by the process) from their desired values and would then produce a control signal to adjust the process parameters. Narendra and Parthasarathy (1990) propose the use of neural networks for the intelligent controller of an arbitrary system which does not require an underlying mathematical model of the controlled process.