Force-based tool wear estimation for milling process using Gaussian mixture hidden Markov models

Tool wear monitoring system is of vital importance for the guarantee of surface integrity and manufacturing effectiveness. To overcome the weaknesses of neural networks, a new tool wear estimation model based on Gaussian mixture hidden Markov models (GMHMM) is presented. Nine types of time-domain features are extracted from the milling force signals which are obtained under four sorts of tool wear state. Besides, the sensitive features which can indicate the tool wear states accurately are selected out by correlation analysis. To test the effectiveness of the presented model, the selected sensitive features serve to identify the tool wear states by utilizing GMHMM and back-propagation neural network (BPNN), respectively. Moreover, the identification performance of GMHMM under the combinations of various numbers of Gaussian mixtures and various lengths of observation sequence is analyzed to verify the practicability of the presented tool wear model. The experimental results show that the GMHMM-based model can identify the tool wear states effectively and GMHMM outperforms the BPNN model in accuracy and stability. This method lays the foundation on tool wear monitoring in real industrial settings.