A lean model for performance assessment of machinery using second generation wavelet packet transform and Fisher criterion

The development of efficient on-line data processing and decision support algorithms is one of future trends of expert systems for machine condition monitoring research. This paper contributes to a lean model for machine performance assessment by combining an efficient signal processing algorithm, an effective feature selection criterion, and an intelligent assessment method. In the proposed model, firstly, a second generation wavelet packet transform is used to project raw signals into the wavelet domain; secondly, the Fisher criterion is applied to reduce redundant dimensions; eventually, a fuzzy c-means clustering method is used to assess and classify the performance of mechanical systems. The vibration signals from a rolling element bearing experiment has been used to verify both efficiency and effectiveness of the lean model. Compared with conventional methods, the lean model can reduce the time consumption of feature extraction by 49.7% and storage space or data transfer load related to the feature dimensionality by 97.7%, which indicates a great improvement in efficiency.