Study On Cutting Process Monitoring System Based On Hidden Markov Model

In manufacturing, tools are turn worn during cutting. It does not satisfy the developing request on the automation and intelligence machines to check the wearing condition of cutting tools by experience or stopping machine in the process of manufacturing. So, how to monitor the tools effectively is becoming a hot issue in the field. The Hidden Markov Model (HMM), which is a time dynamic recognition model, is employed to monitor the drilling process based on the vibration signals of the machine worktable.Three fundamental algorithms of HMM are introduced in this paper. The basic methods modified in practice and model initial parameters chosen are described too. Meanwhile, the structure of HMM and the monitoring principle based on HMM are also given.Signals are processed for feature extract respectively by RMS&PEAK amplitude, FFT, Wavelet packet, AR model and so on. Former three features are chosen according to the correlation between features and tool wear condition. In vector quantization system which is formed to resolve the coding problem after the feature extract in HMM, typical LBG algorithm is introduced, another system called SOFM network is adopted as well.HMM is used for recognition, following the feature extraction and coding. So multi-method based on HMM are studied, the result indicates that every method has a high rate in recognition, especially for the worn tool, is above 95%. HMM can effectively monitor the tool wear condition, and compared to BP network widely used in recognition, HMM has some merit as short-time training, strong time-dynamic, adaptive in unreasonable signal.