Bayesian updating with SC-MCMC for PMSM digital twin models

The fidelity of permanent magnet synchronous motor (PMSM) digital twins tends to degrade during operation due to system non-stationarity, which arises from time-varying operating conditions, parameter drift, and environmental disturbances. This degradation significantly limits their effectiveness in predictive maintenance and intelligent operation. This paper presents a chain-splitting self-correcting Markov chain Monte Carlo (SC-MCMC) method for autonomous Bayesian updating of PMSM digital twins. The method employs a residual-variance-driven auxiliary chain to explore inefficient sampling regions and adaptively correct the main chain’s proposal distribution, enhancing sampling efficiency and posterior convergence. A physics-driven PMSM digital twin is constructed by integrating finite element analysis with eddy-current-equivalent reduced-order modelling. Sobol global sensitivity analysis identifies key parameters, incorporated into a Bayesian framework for online updating. Experimental results show the updated digital twin significantly improves phase current prediction, reducing root-mean-square and mean absolute errors by 48.65% and 44.65%, respectively, providing a practical solution for high-fidelity modelling and predictive operation in smart manufacturing.