Machine degradation analysis using fuzzy CMAC neural network approach

Traditionally, machine condition is described as a dichotomous problem: normal or failure. In fact, machines go through a series of degrading states until failure occurs. Degrading states do not make machines breakdown, but they do decrease performance reliability and increase the potential for faults and failures. In this paper, a neural network approach based on fuzzy cerebellar model articulation controller (FCMAC) is proposed to analyze machine degradation severity. Two kinds of situations are discussed: (1) machine signature from different levels of degradation severity are available to train the network; (2) machine signature from only normal perfect state are available to train the network. A degradation index is developed to reflect the degradation degree. The FCMAC network works as a 'classifier' or a 'condition discriminator' in these two situations. An example is presented to demonstrate this method and the results show that the FCMAC network is capable of ranking machine degradations quantitatively in both situations. With degradation analysis, maintenance activities can be implemented before failure eventually occurs, thus preventing fatal breakdowns more effectively.