In recent years, the development of models for tool condition monitoring and predicting mechanical system failures attract more and more attentions. Although different cutting signals have been studied for accurately predicting tool wear, it is found that the cutting signals are easily affected by processing parameters, tool geometry, etc.. Therefore, in most studies, experiments were only based on specific processing parameters.

    In order to understand the influence of the clamping force of the vise on cutting signal, a load cell was installed on the vise and set the clamping forces to be 600 kgf and 1400 kgf. The difference in the time domain and frequency domain signals of the three-axis accelerometer and the acoustic emission under these two clamping forces were observed. These signal characteristics were used to build up a neural network model for tool condition monitoring. This study first built up a model with cutting signals based on 600 kgf clamping force, and then used the cutting signals under different clamping forces as the test data set to test the accuracy of the model.

    The results showed that the signal characteristics are affected by the clamping force, resulting in prediction error of the neural network model. Through the improvement of feature engineering, the accuracy of the model based on X-direction vibration signal and acoustic emission signal have been increased to 90.4% and 83.2%, respectively.

Keywords:  Vise clamping force; Tool wear; Vibration signal; Acoustic emission; Neural network