Non-intrusive Monitoring of Milling Process

Fault diagnosis of industrial equipments is more and more important for improving the quality of products and avoiding a waste of money and resources, especially for expensive precision machines, like the milling machine used in Berg Propulsion. If the milling process is continued when the cutting tool is worn out, one may have a failure product. This will cost the company a lot of money. And what’s more, with the time going on, the worn out tool may cause the damage to the milling machine. So developing a fast and reliable diagnosis system is necessary for these industrial equipments. In this thesis, we will analyze the working conditions of cutting tools by collecting the sound signals from the machine using Wavelab and using signal processing and case based reasoning methods to detect the faults during the milling process. Analyses in the time and frequency domains are conducted first on the collected sound signals in order to illustrate their main features and then extraction of features is made. Features extracted can be built as a case library for case-based reasoning (CBR) to give a recommendation later on when the machine should stop when the tool damage is just beyond the specified tolerance of damaging degree. This recommendation is based on the previously identified and classified cases in a case library. However, as there are some limitations, such as the precision of microphone used to collect the data, the exact features are difficult to find and thus not given in this thesis, and building the case library will be one of the future researches. This thesis is mainly focused on analysis of signals, extraction of features for a few cases using signal processing and establishing a few simple cases, e.g. analyzing the features of each sound signal detected during the process, illustrating a few examples and some ideas, and making a few suggestions on improvement in the future research. Key Words: signal processing, feature extraction, case-based reasonin