XGBoost-Based Digital Twin Model for Predicting Trajectory Errors in a Hexapod Coordinated Machining System Using...

Area of Science:

• Robotics and Manufacturing
• Data-Driven Modeling
• Metrology

Background:

• Dynamic errors in robotic machining reduce part quality, especially in flexible systems prone to disturbances.
• Accurate prediction of these errors is crucial for maintaining high-precision manufacturing.

Purpose of the Study:

• To develop a data-driven digital twin for predicting circular trajectory errors in hexapod machining cells.
• To utilize a compact sensor configuration combining ballbar and vibration data for error prediction.

Main Methods:

• Synchronized processing of ballbar deviations, acceleration data, and CMM profiles.
• Feature engineering using sliding-window vibration statistics and ballbar path error.
• Machine learning models (XGBoost, MLP, Random Forest) for pointwise error prediction.

Main Results:

• XGBoost model achieved micrometer-level accuracy (RMSE ~5 µm, R² > 0.80) on unseen data.
• The model demonstrated near-complete coverage within a ±20 µm tolerance band.
• Hybrid feature sets combining ballbar data and vibration descriptors maintained high predictive accuracy.

Conclusions:

• The data-driven digital twin effectively predicts circular trajectory errors in robotic machining.
• Ballbar path error is key for geometric error, while vibration data captures dynamic aspects.
• The approach enables practical offline compensation for enhanced part quality.