Physics-informed hybrid digital twin framework integrating fractal contact modeling and edge-cloud artificial intelligence for dynamic thermal contact conductance prediction

Abstract

Accurate characterization of thermal contact conductance (TCC) at tapered roller/groove interfaces is essential for predicting thermal behavior in precision machine tools. Conventional analytical models fail to capture the nonlinear, time-varying effects arising from fluctuating loads, curved geometry, and evolving micro-asperities. To address the above challenges, in this study, a physics-informed digital-twin framework is proposed for real-time TCC prediction. The fractal contact mechanics, a Simulink-based transient thermal network, and edge-cloud artificial intelligence are integrated in this framework. Specifically, a fractal-Monte Carlo model is used to reconstruct multi-scale rough surfaces and a geometric coincidence factor is introduced to quantify curvature-dependent contact conformity. The variable-resistance Simulink model is proposed to generate physics-consistent datasets for training. At the data-intelligence layer, a physics-informed neural network (PINN) and Transformer predictor are used to embed heat-conduction constraints and learn temporal dependencies in dynamic TCC evolution, respectively. The results show that the proposed hybrid model significantly outperforms standalone Simulink, PINN, and Transformer models. Under dynamic loading, hysteresis-dependent TCC behavior is accurately reconstructed and tracking errors are reduced by 72–89% across 0.5–2 Hz load cycles. With edge-cloud deployment, real-time inference achieves < 150 ms latency, maintaining synchronization within ±0.3 °C. The proposed method provides a physically interpretable, data-efficient, and real-time deployable solution for TCC characterization, offering a new paradigm for artificial intelligence (AI)-enhanced thermal behavior modeling in machine tools.

Topics & Concepts