Mechatronic digital twin monitoring of hydrostatic rotary tables using multi-point thermal sensing and spatial reconstruction

Hydrostatic rotary tables are key mechatronic components in high-precision manufacturing equipment, and their thermal evolution affects operating stability and machining-related accuracy. This study presents a sensing-driven digital twin monitoring framework that integrates multi-point thermal sensing, future temperature prediction, spatial thermal-state reconstruction, and virtual–physical visualization. Experimental temperature data from fifteen structural monitoring positions are organized using a chronological file-level split to avoid temporal leakage in time-series evaluation. Five forecasting models are compared, and a causal temporal convolutional network is selected as the predictive module. The final model uses the preceding 30 min temperature sequences of T1–T15 to forecast their temperatures 5 min ahead, achieving an RMSE of 0.0673 °C and an MAE of 0.0461 °C on the independent test subset. Measurement-constrained region-aware reconstruction reduces RMSE from 0.1816 to 0.1628 °C; strict sensor omission validates spatial recovery, while Unity updates 64,000 voxels in 13.42 ms.