Multi-Class Classification Approach based on Fuzzy-Filtering for Condition Monitoring

摘要

Analytical model-based methods have been developed during the last decades to achieve the goals of fault diagnosis of systems. One of the drawbacks of these methods, is the necessity of precise models of the considered system in order to design an appropriate fault detection/diagnosis system. This strong assumption can not be fulfilled for all cases. Additionally, different models have to be defined for distinguishing different states of machines operation. Qualitative model-based methods and also signal-based methods avoid this problem due to their different principle concepts of modeling. This contribution deals with the idea of combining qualitative model-based methods using fuzzy logic and statistical methods describing signal properties. The desired goal is to design a condition monitoring system based on suitable and available signals, related measurements experiments, and classifying information of the system to be monitored. The key idea of this contribution is the generation of a set of features to distinguish related different states of the system. For validation of the developed method, experimental data are used from an experimental studyed friction and wear processes of a metal surface allowing the distinction of different wear states. The developed method shows good ability to distinct the related states of wear.