Intermittent fault detection based on feature ensemble

Intermittent faults are usually difficult to be detected, which may impact the performance and reliability of systems. Small magnitude, short duration, random occurrence, etc. are the main reasons lead to that it is intractable to effectually detect intermittent fault. Therefore, it is essential to develop competent techniques for this issue. Although many methods have been proposed to detect intermittent faults in the past few years, these methods may rely on some data assumptions, and may not fully mine hidden information and complex relationships in the data. Different approach tends to extract and utilize different process information. By ensembling the features extracted by these methods, the limitations of a single method can be overcome and the generalization ability can be improved. How to combine features extracted from different methods to improve detection performance has become a new challenge. In this paper, an optimal weighted and moving average feature ensemble model (OWMAFEM) is proposed. Firstly, the principal component analysis (PCA), dynamic PCA (DPCA), moving average PCA (MAPCA), exponentially weighted PCA (EWMPCA) and optimally weighted moving average T ² control chart (OWMA-TCC) are adopted as base detectors. Then an optimal weighted and moving average is developed to ensemble features. In addition, the superiority of the proposed approach is verified by numerical simulation. The final results show that the OWMAFEM performs better than OWMA-TCC, MAPCA, EWMPCA, PCA and DPCA.