YLYu Liu

Yu Liu

*This is an algorithmically generated author record based on publicly available information.

Yu Liu has a total of 4 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Thermal error prediction of high-speed motorized spindles using TC-VIF-based sensor optimization and hybrid polynomial regression–LSTM modeling

    by Peng Li, Yu Liu, Lixiu Zhang, Songhua Li, Gang Wang

    Precision manufacturing heavily relies on high-speed motorized spindle equipment. Correspondingly, thermal errors generated by spindles severely curb achievable machining accuracy and stand out as a critical restrictive element. Existing thermal error models often rely on redundant sensor configurations and suffer from degraded prediction performance under varying speed conditions. This study presents a two-stage approach for motorized spindle thermal error prediction. First, we construct a thermomechanical coupled finite element model for characterizing temperature distribution and guiding the layout of measuring sensors. The Temporal Correlation–Variance Inflation Factor (TC-VIF) method is then utilized for optimization of temperature measuring positions, cutting the number of sensors from seven to four and removing multicollinearity. For thermal error modeling, a hybrid framework is developed in which second-order polynomial regression with ridge regularization captures the fundamental temperature–deformation mapping, while a two-layer LSTM models the residuals to compensate for dynamic nonlinearity. For cross-condition validation, the model is trained on three speed conditions and tested on two unseen conditions. Results show that the proposed model achieves the lowest MAE on the test set, demonstrating superior average prediction accuracy. Compared to PSO-SVM, SSA-KELM, WOA-BP, and TCN benchmarks, the hybrid model reduces MAE and RMSE by 44.8% and 40.0% on average, respectively, and achieves a coefficient of determination R² of 0.95 on the training conditions. This enhanced cross-condition generalization is attributed to the effective integration of physics-informed modeling with data-driven learning.