Uncovering nonequilibrium dynamics in machining via mechanism-assisted machine learning

Developing lightweight models with clear underlying mechanisms and robust generalization capabilities is one of the means to advance machining process monitoring technology. Current models predominantly rely on traditional feature extraction techniques, significantly increasing model complexity and lacking clear mechanisms. This paper presents a mechanism-assisted Gaussian process, grounded in the principles of the Boltzmann distribution and entropy theory in statistical mechanics, for uncovering dynamics from machining data. Employing non-equilibrium statistical mechanics theory to establish the dynamics of the machining process facilitates accurate estimation of the processing state without resorting to overly complex models. Experimental evidence demonstrates a strong correlation between the established entropy and partition features associated with machining signals and the actual state. Furthermore, the model performs remarkably well in tasks encompassing the progressive wear prediction in milling, the chatter detection in turning, and the wear condition identification in drilling.