Research On Tool Condition Monitoring Technology Based On Heterogeneous Ensemble Learning Model

With the development of aerospace technology, more and more materials which are difficult to cut have been used widely. When these materials are machining, the tool will wear severely. If the tool condition could be known in time, the quality of workpiece can be improved and the processing efficiency will increase. Therefore, it is meaningful to monitor the tool wear condition. However, nowadays the single classifiers in the monitoring system can't get best performance because of the complex tool wear style and diversity of the machining ways, so it's necessary to build a new classifier to realize the tool condition monitoring for complex machining process.In this paper, a heterogeneous ensemble learning model was proposed, and the tool wear monitoring system was built based on the model. In the model, the SVM, RBF and HMM models were selected as base classifiers depend on the base classifier selection criterion. Meantime, in order to show the advantage of heterogeneous ensemble, the homogeneous ensemble learning and single classifiers were constructed to make a comparison and the majority voting and stacking strategy were used to choose the better combination ways. In order to test the performance of the monitoring system, the titanium alloy milling experiment and CFRP drilling experiment were carried out. In the milling and drilling experiment, the force signals were collected. Then the milling force signal was used to extract harmonic features, and the mRMR algorithm was used to realize the feature selection. But the feature extraction technology in time domain was used for force signal in drilling, then the LPP algorithm was used to select features. Based on these datasets, the heterogeneous ensemble learning model was trained and tested. By the comparison with homogeneous ensemble learning and single classifiers, it's proved that the heterogeneous ensemble learning has better accuracy and stability. By the comparison between the two strategies, the stacking strategy outperforms majoring voting both in accuracy and stability. Therefore, the heterogeneous ensemble learning with stacking strategy has the best performance.According to the research in the paper, it's proved that the heterogeneous ensemble learning model has better generalization ability and the model complexity is reduced, so the heterogeneous ensemble model has better performance. In addition, the tool condition monitoring system based on the heterogeneous ensemble learning model is meaningful in practice.