Abstract: Tool wear has a great influence on the machining quality，production efficiency and cost of parts in the process of machining. The reasonable and accurate noise reduction for collected tool wear signal is the core technology for tool wear detection. Based on the weights of the signal-to-noise ratio，the smoothness and the mean square root error which are constructed by the entropy method，the fusion evaluation index of noise reduction quality is proposed. Comparison of the denoised simulated signal with the real signal shows that the fusion evaluation index is of the feasibility and superiority. And then an optimization method of noise reduction parameters is presented with the objective of maximum fusion evaluation index. For the actual vibration signal of tool wear after wavelet threshold denoising，the noise reduction parameters optimized by the fusion evaluation index proposed in this paper can be used to remove the noise signals at high frequency \[6 kHz，12 kHz\] and also retain completely the real signals at low frequency \[0 kHz，6 kHz\] compared with the traditional single evaluation index. Finally，a neural network prediction model is established from the extracted feature values of tool wear for describing the relationship between the tool wear and the cutting parameters. Experimental results show that the tool wear signal features can be accurately extracted using the optimal noise reduction parameters based on multi-index fusion evaluation. The maximum error between the predicted values and the measured values is no more than 6.0%.

Key words: toolwear, noisereductionparameter, entropymethod, waveletthresholddenoising, evaluationindex