Monitoring tool wear using wavelet package decomposition and a novel gravitational search algorithm–least square support vector machine model

Monitoring tool wear has drawn much attention recently since tool failure will make it hard to guarantee the surface integrity of workpieces and the stability of manufacturing process. In this paper, the integrated approach that combines wavelet package decomposition, least square support vector machine, and the gravitational search algorithm is proposed for monitoring the tool wear in turning process. Firstly, the wavelet package decomposition is utilized to decompose the original cutting force signals into multiple sub-bands. Root mean square of the wavelet packet coefficients in each sub-band are extracted as the monitoring features. Then, the gravitational search algorithm–least square support vector machine model is constructed by using the extracted wavelet–domain features so as to identify the tool wear states. Eight sets of cutting experiments are conducted to prove the superiority of the proposed integrated approach. The experimental results show that the wavelet–domain features can help to ameliorate the performance of the gravitational search algorithm–least square support vector machine model. Besides, gravitational search algorithm–least square support vector machine performs better than gravitational search algorithm–support vector machine in prediction accuracy of tool wear states even in the case of small-sized training data set and the time consumption of parameters optimization in gravitational search algorithm–least square support vector machine is less than that of gravitational search algorithm–support vector machine under large-sized training data set. What's more, the gravitational search algorithm–least square support vector machine model outperforms some other related methods for tool wear estimation, such as k-NN, feedforward neural network, classification and regression tree, and linear discriminant analysis.