Mechanism-informed modeling and compensation of thermal errors in fully closed-loop feed drive systems with fused spatiotemporal feature interactions

The dynamic variation of the temperature field induces thermally-induced positioning errors (TPE) in fully closed-loop feed drive systems (FCL-FDS). Neglecting such errors will lead to a decline in the positioning accuracy of the machine tool and compromise the overall machining quality. Since the mechanisms underlying TPE in FCL-FDS have not been clearly elucidated and are subject to complex multi-factor coupling, existing modeling and compensation methods suffer from limited robustness and generalization capability. To address this challenge, this study theoretically elucidates the formation mechanism and influencing factors of TPE in FCL-FDS from both the forward command and backward feedback perspectives. On this basis, a SA-IWOA-CALA algorithm is proposed, which fuses GoogLeNet-CNN and LSTM networks in a parallel architecture, and introduces a hierarchical three-level attention mechanism. This framework enables the effective representation and accurate modeling of the nonlinear spatiotemporal evolution process of TPE driven by multiple thermal sources in FCL-FDS. To further enhance the model performance, an improved SA-IWOA intelligent optimization method is introduced for hyperparameter tuning. Interval compensation experiments demonstrate that the proposed approach achieves a maximum TPE reduction of 81.88 %, maintaining TPE within ±2 μm after compensation. These results validate the effectiveness of the proposed modeling and compensation framework. The findings of this study offer a robust theoretical basis and practical methodology for mechanism analysis, modeling, and thermal error compensation in FCL-FDS.