Model Learning in a Multistage Machining Process: Online Identification of Force Coefficients and Model Use in the Manufacturing Enterprise

This work presents a systems approach in machining process control. Traditional force-based machining process control has been focused on single machine-single operation. The force or power sensor is used to measure the instantaneous force/power, and control action is taken by changing the feedrate in real time to follow a given force setpoint. The application of such control has successfully been implemented to prevent chatter and to elongate tool life by minimizing tool wear. This research seeks to extend the application of control algorithms to learn about the machining system (comprised in this context of a workpiece being operated on in progressive machining), and how knowledge generated by the process can be passed on to the next process for optimization. To demonstrate this, turning of a partially hardened bar is explored. A nonlinear mechanistic force model-based control framework attempts to control the cutting force at a designated setpoint, with material properties changing over the cut. The force coefficients for the material are calculated offline using experimental data and Bayesian inference methods. Since the hardened part of the bar will shift the force coefficient values, an online estimation strategy (Bayesian Recursive Least Square estimator) is used to learn the new coefficients as well as satisfying the control objective. With the newly learned coefficients passed downstream, the subsequent operation experiences no compromise of control objective as well reduces the maximum values of force encountered. Numerical analyses presented show the adaptation and control scheme performance.