Superresolution Restoration of Cutting Forces in Thin-Walled Milling With a Diffusion Model

Accurate acquisition of the instantaneous cutting force appears to play a fundamental role in understanding most of the machining process. It is usually hard to fully take into account the nonlinear effects and dynamic response in the cutting force prediction with mechanistic force model in thin-walled milling. Besides, direct measurement can be difficult due to the dynamic behavior of the machining system and the bandwidth limitation of the dynamometer especially for the high-frequency part. To solve those problems, this article aims to propose a novel online approach to achieve accurate and efficient restoration of the cutting forces based on the diffusion model with two processes. The forward process is defined by a fixed Markov chain, and the reverse process is constructed based on a neural network, which can recover the high resolution force signal from a Gaussian noise. The cutting conditions and the measured low-frequency force signals are encoded respectively and concatenated as the conditioner of the network. The efficacy and the prediction accuracy of the developed model are demonstrated in the milling experiments on various parts, and a better performance is found in the prediction of surface roughness with the restored cutting force.