Research On Tool Fault Diagnosis Based On Wavelet Packet And Autoregressive Spectrum Analysis

With the development of computer technology, we are now applying more and more unmanned manufacturing systems. The most important requirement of an automatic manufacturing system is that the system can automatically carry out on-line monitoring and adjustment for the breakdowns appearing during the production process. Tool is one of the most important factors in the machining process. Tool wear can affect the quality of the product, increase the cost, reduce the production rate, and cause the accident. Therefore, the on-line monitoring of tool seems especially important. And the most important and key problem is the feature extraction of the test signal of the tool. In a sense, the feature extraction of the tool can be considered as the bottle-neck problem in the research of the monitoring and fault diagnosis, because of the diversity of the machining condition, the variety of the machining parameter, the random property and complexity of the wear and tear etc. It directly relates to the accuracy and credibility of the tool wear state monitoring and fault diagnosis, and is the key of the success of the monitoring and fault diagnosis process.Wavelet analysis has the good time-frequency property for the treatment of the non-stationary signal. Wavelet transform observes the signal with the different dimensions, and decomposes the signal into different frequency band. Through this, the general view as well as the details of the signal can be revealed. So the relationship between the time and frequency of the signal can be accurately reflected. Autoregressive model can reflect the running state of the system and can also forecast the future state and trend of the system.Wavelet packet analysis can get decomposable sequence distributing in the different frequency band through the pass frequency band, and the information is intact. So the wavelet packet analysis has a localized analysis function for the signal. Autoregressive spectrum has the function of extrapolation forecast and is good at handling short data. Its resolution power of frequency can be raised...