Problem: At present, there are more than two thousand rotating machines in the studied gas separation plant that need to be continuously inspected in routine monitoring process. The current inspection routine is done by collecting vibration signals from all of the machines and saving them in the central database. The data would then be analysed by experts to identify possible fault types of the machines. However, the company was struggling with a strongly decreasing degree of expertise due to job transfer and retirement, and the scarcity of young experts with relevant training.

Solution: Generally, if a set of labelled faulty data is available, a supervised-based fault classification could be applied. However, the data recorded in industry for rotating machine health monitoring are often a large number and unlabelled. It is impractical to label these data manually. Our solution is to develop of an expert system that merges supervised and unsupervised techniques. Unsupervised methods have been used to facilitate the supervised learning by performing the data clustering. Once the groups are found, the identification of the fault type of each group can be performed by experts.