Study On Cutting Tool State Monitoring Based On Feature Fusion

Sensitive signals to cutting tool state, such as cutting force and vibration, is selected to serve as monitoring foundation in this research, and the problem on collecting sensitive signals, signal transformation, data collecting, signal denoising, feature extraction, feature fusion and cutting tool state recognition based on fusion feature is systematically studied. Aim at unsteady feature of state signal of cutting tool and noise interference causing difficulty to feature extraction and state recognition, this paper make use of orthogonal wavelet transform to denoise collect signal of cutting tool state, propose denoising method based on relativity of wavelet transform domain coefficient. Aim at sampling array incorporating excessive redundant information and irrelevant information, this paper make use of low order statistical method to analyze diversity of time domain parameter and frequency spectrum of sampling array along cutting tool state, extracting low order statistical feature of cutting tool state. Aim at no Gauss signal constituent of sampling array sensitive to cutting tool state, paper make use of high order statistical method to analyze diversity of high order cumulative statistic and double spectrum of sampling array along cutting tool state, extracting high order statistical feature of cutting tool state. Aim at local unsteady property of sampling array, this paper make use of wavelet method to analyze diversity of each wavelet reconstructive component of sampling array along cutting tool state, extracting wavelet feature of cutting tool state. Aim at the discrepancy of classified capacity and stability of feature, this paper make use of BP neural net to remark and select above feature, pick out basic feature of cutting tool state. Aim at the limitation that single feature exist classified blind area, only providing limit cutting tool state information, the paper make use of statistical analysis, linear dominical element analysis, mutual information entropy analysis to fuse basic feature, in other word taking full advantage of multiple basic feature including tool state information to reasonably combine the redundant information and complementary informationof basic feature according fusion standard, so that fusion feature of cutting tool state can much completely and accurately illustrate cutting tool state. Aim at classified difficulty of feature sample of distinct cutting tool state kind existing overlap region, the paper propose the model and method of cutting tool state recognition based on fuzzy decision. Aim at feature decision edge existing certain nonlinearity, the paper propse the model and method of cutting tool state recognition based on BP neural net. Then two above recognition model are fused at decisional level, the model and method of cutting tool state recognition based on classifier fusion is proposed. The efficiency and accuracy of cutting tool state recognition is remarkably raised by the fusion of two above recognition model based on different feature. The experiment shows that cutting tool state monitoring based on feature fusion has the advantage of strong resistance to interference, high monitoring precision, excellent reliability and high recognition accuracy.