A dual compensation strategy based on multi-model support vector regression for tool wear monitoring

Abstract Tool wear monitoring is essential in precision machining and helps to ensure processing quality. Although numerous data-driven methods have been proposed for tool wear monitoring, most of them build global models that ignore local wear characteristics and the changeable working conditions. This study proposes a dual-compensation (DC) strategy based on multi-model support vector regression (MSVR) to improve estimation accuracy and model maintenance. By dividing the original samples into two clusters with the K-means algorithm, the DC strategy develops a separate SVR model for each cluster. Test samples are classified using the decision function and input into the corresponding model. Finally, the predicted value obtained by MSVR is corrected by system error compensation and nearest bias updating, which is improved to adapt to milling environments. In addition, the database is updated after each process. Two groups of milling experiments were conducted to validate the improved strategy in comparison with other methods. The results show that multi-model SVR improves performance by more than 28.7% and has better generalization ability. The root mean square error value decreases from 0.1109 to 0.0392, a 64.7% reduction, with our complete strategy. The DC strategy based on multi-model support vector regression (DCMSVR) can achieve high prediction accuracy in variable working conditions due to its high estimation accuracy and adaptability.