Real-time multivariate performance degradation assessment based on a cerebellar model articulation controller

With the development of the sensor technology, it is easy to acquire the multivariate information. However, how to fuse multiple performance data to track component performance degradation process becomes more complex. In this paper a real-time multivariate performance degradation assessment method is presented. Firstly, a cerebellar model articulation controller (CMAC) is constructed, and it is used to map multivariate input of the state space into univariate output of the feature space. Secondly, a pattern recognition model based on CMAC is presented to analyze the machine condition quantitatively. In this model, the good pattern is defined as "0" and the fault pattern is defined as "1". We use CMAC to learn weighted table from the normal pattern and the fault pattern respectively. When a new data is measured, a real-time indicator of the relative performance degradation rate is proposed by calculating its similarity metric with the normal and fault pattern in the feature space. Furthermore, by analyzing the degradation data from high pressure water descaling pump in the process of failure, this method is proved to be able to analyze machine degradation quantitatively. This method could help equipment maintenance personnel make correct decision to decrease unnecessary downtime.