Deep learning-physical hybrid model-driven in-situ cutting force monitoring of ultra-precision diamond turning

Abstract Ultra-precision diamond turning (UPDT) can manufacture a high-quality surface at a nanometric surface roughness and submicron form error. Cutting force is a good indicator for monitoring surface quality in the UPDT process. However, the traditional monitoring methods of cutting force suffer from problems such as data noise interference and delayed monitoring response in UPDT. Motivated by this, based on the fusion of deep learning and physical cutting force modelling, an in-situ monitoring method is proposed for UPDT. Firstly, a hybrid cutting force model (named PCLA model) is established, integrating a physical cutting force model with a Convolutional Long short-term memory Attention neural network (CLA). Secondly, an in-situ monitoring model for UPDT is established based on an autoencoder. The in-situ monitoring model is trained using cutting force data generated by the PCLA model, and simulation verification is performed using the physical cutting force model. Finally, the proposed method is experimentally verified by the cutting force of UPDT and the machined surfaces. The overall accuracy of the proposed model can achieve 96.30% through simultaneous analysis of the designed defects, corresponding cutting force, and monitoring errors. And the results verify the monitoring ability of the proposed method for anomalous cutting force in the time domain.

Keywords: Deep learning; In-situ monitoring; Cutting force; Ultra-precision diamond turning