The application of neural networks in self-tuning constant force control

The constant force control gradually becomes an important technique of modern manufacturing processes. For example, the constant turning or cutting force is a useful approach for increasing the metal removal rate and increasing the tool life. The variation of machining condition may cause the robustness of a classical control theory (PID) to become ineffective, even make a control system unstable. The pole placement self-tuning control (PSTC) theory with a recursive least square parameters estimator is proposed to adapt the environmental variety. Unfortunately, the adaptability and the robustness of a self-tuning control system cannot be maintained in good condition simultaneously all the time. In this paper, a self-tuning controller equips with a neural network parameter classifier in conjunction with a least square estimator is developed to improve the adaptability and the robustness in suffering the obvious environmental variation. In order to verify the applicability of this control method, a prototype system is designed and constructed to resemble the feed rate mechanism of lathe. The dynamic responses of this force control system with different estimators are compared based upon the experimental data. The contact force is measured from a load cell and adjusted by regulating the feed rate.