International Journal of Advanced Manufacturing Technology (02683768)129(7-8)pp. 2949-2968
The mechanistic force model is one of the most common methods used to predict cutting forces in milling processes. In this model, the cutting forces are considered a function of cutting geometry and the material properties of the tool and workpiece, generally known as cutting coefficients. These coefficients are commonly identified by performing special calibration tests and are applied to predict cutting forces for other conditions. Although the mechanistic model is a powerful tool for predicting cutting forces, its accuracy decreases as the cutter-workpiece engagement geometry differs from the calibration tests. Thus, it is necessary to update the cutting coefficients to preserve the accuracy of the model. This paper proposes a real-time intelligent method, named “the mechanistic network,” for identifying and updating the cutting coefficients. To this end, an analogy between the mechanistic force model and artificial neural networks is identified, in which the weight coefficients of the artificial neural networks have been replaced with the cutting coefficients. To identify and update the cutting coefficients, an algorithm is proposed using stochastic gradient descent, which updates the coefficients in each iteration. In addition, some other important parameters in milling processes, such as the phase shift between the measured and predicted forces and run-out parameters, are calculated using stochastic gradient descent. The good performance of the proposed network is shown through case studies by utilizing reliable data existing in the literature and also by performing ball-end milling experimental tests. The results show that the proposed network can predict the cutting forces with an error of less than 10% and update the cutting coefficients with a calculation speed of 125k iterations per second. The robustness of the network against noise that may arise in real machining conditions is also shown. The proposed mechanistic network is a reliable and efficient tool that can be applied to real-time applications such as cyber-physical manufacturing systems and condition monitoring of machining processes. © 2023, The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature.
