Dynamic Error Modeling and Predictive Compensation for Direct-Drive Turntables Based on CEEMDAN-TPE-LightGBM-APC Algorithm

TL;DR

This paper introduces a new algorithm to improve the accuracy of direct-drive turntables by predicting and compensating for positioning errors.

Contribution

A novel dynamic error compensation model using CEEMDAN-TPE-LightGBM-APC for precise prediction and correction of turntable positioning errors.

Findings

Positioning error ranges were significantly reduced after compensation, with standard deviation reductions of 71.2% and 61.6%.

The proposed model demonstrated improved prediction stability and operational efficiency in precision mechanical systems.

The method effectively integrates decomposition, prediction, and correction stages for enhanced accuracy.

Abstract

The direct-drive turntable serves as the core actuator in high-precision macro-micro drive systems, where its positioning accuracy fundamentally determines overall system performance. Accurate error prediction and compensation technology represent a critical prerequisite for achieving continuous error compensation and predictive control in direct-drive turntables, making research on positioning error modeling, prediction, and compensation of vital importance. This study presents a dynamic continuous error compensation model for direct-drive turntables, based on an analysis of positioning error mechanisms and the implementation of a “decomposition-modeling-integration-correction” strategy, which features high flexibility, adaptability, and online prediction-correction capabilities. Our methodology comprises four key stages: Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN)-based decomposition of historical error data, development of component-specific prediction models using Tree-structured Parzen Estimator (TPE)-optimized Light Gradient Boosting Machine (LightGBM) algorithms for each Intrinsic Mode Function (IMF), integration of component predictions to generate initial values, and application of the Adaptive Prediction Correction (APC) module to produce final predictions. Validation results demonstrate substantial performance improvements, with compensated positioning error ranges reduced from [−31.83″, 41.59″] to [−15.09″, 12.07″] (test set) and from [−22.50″, 9.15″] to [−8.15″, 8.56″] (extrapolation test set), corresponding to standard deviation reductions of 71.2% and 61.6%, respectively. These findings conclusively establish the method’s effectiveness in significantly enhancing accuracy while maintaining prediction stability and operational efficiency, underscoring its considerable theoretical and practical value for error compensation in precision mechanical systems.

Figures