Closing the loop of model predictive force control in milling with ensemble Kalman filtering

  • Schließen des Regelkreises einer modellprädiktiven Kraftregelung beim Fräsen mit einem Ensemble Kalman Filter

Schwenzer, Max; Bergs, Thomas (Thesis advisor); Abel, Dirk (Thesis advisor)

1. Auflage. - Aachen : Apprimus Verlag (2022)
Book, Dissertation / PhD Thesis

In: Ergebnisse aus der Produktionstechnik 13/2022
Page(s)/Article-Nr.: 1 Online-Ressource : Illustrationen

Dissertation, RWTH Aachen University, 2021


In milling, the force determines productivity, quality, and safety. Fast-changing engagement conditions between tool and workpiece provoke abrupt changes in the force causing classic, reactive control to fail. Such failure is commonly avoided by using predictive control methods that anticipate process behavior on the basis of process models. One representative, which had been successfully applied to the milling process, is model-based predictive control (MPC). MPC for force control in milling requires an online identification of a process model. A process model consists of a mechanistic force model and of a model of the radial deviation adjusting the undeformed chip thickness. The state-of-the-art identification method is a repeated curve fit over larger intervals, essentially opening the force control loop in the meantime. The results of these curve fits are uncorrelated and have to be smoothed in order to producea mathematically continuous feedback signal to avoid instability of the controller. For the identification, the presented work introduces a recursive method based on the idea of a disturbance observer. The observer is a particle filter known as ensemble K ALMAN filter (EnKF). EnKF is mathematically proven to converge correctly if the search space is large enough so that it contains the solution. The search space can be narrowed by additional knowledge limitingit to realistic solutions. There exist a large number of experimental reports on mechanistic force models making them well-suited for an EnKF-based identification. The filter states the originality of this work.After the theoretical background is derived and after the adaption of the filter to the problem case in simulations, the EnKF is studied on real measurements. It shows a remarkable robustness against measurement noise. With regardto online capability, design and setup of the identification, of the controller itself, and of the overall system are put on a proper basis. The system is tested against a conventionally designed process with constant feed, as wellas against previous implementations of a model predictive force controller (MPFC) in milling. The contribution to the state of the research are first and foremost robustness as well as the ability to permanently close the feedback loop via the EnKF. The advantage to state-of-the-industry solutions is the reduction of the manufacturing time by 40 - 60 % compared to a process with constant feed. So far, these contributions do not address a potentially better utilization of the tool life as a result of a more homogeneous load on the tool. The thesis describes the continuous adaptation of a model in the MPFC and thus enables a truly adaptive, optimal manufacturing process at its technological limit and an increase in workpiece quality.