Hybrid-Mechanism Deep Learning Modeling for Machine Tool Thermal Error: Robust Prediction via Few-Sample Learning
Abstract
Spindle thermal error significantly limits high-precision machining. To address the limitations of mechanism models (low accuracy) and data-driven approaches (high data dependence), this study proposes a hybrid modeling method. It synthesizes a linear thermal expansion mechanism model and a GAT-LSTM-based residual data-driven model. Experimental validation under complex operating conditions demonstrates superior performance: RMSE = 3.81 um (a 31% reduction vs GAT-LSTM and 50% vs MLR). Crucially, using only 17% of the training data (single-case set), the model maintains RMSE = 6.16 um, validating robust few-sample adaptability for real-time compensation.
Fields of Study & Subjects
Metallurgy and Material Forming
Advanced machining processes and optimization
Advanced Measurement and Metrology Techniques
Publication details
- Journal / source
- Machines
- Publisher
- MDPI AG
- Volume
- 14
- Issue
- 4
- Pages
- 399–399
- Publication date
- 6 April 2026
- Language
- EN