Abstract
With the rapid advancement of science and technology, the modern intelligent power plant with large capacity and low energy consumption gradually take place of the traditional power plants. Vibration signal is widely used for fault diagnosis in modern power plant. However, because of nonstationarity, nonlinearity and complexity of vibration signal, it is difficult to analyze the vibration signal directly. To solve the problem, a novel method using vibration signal is proposed in this paper to develop the fault diagnosis model. First, ensemble empirical mode decomposition (EEMD) is employed to decompose the original signal into serval intrinsic mode functions (IMF). To derive rich faulty information, both time domain and frequency domain statistical features are calculated for each IMF. Then fault related key features will be selected to reduce the feature redundancy. In feature selection, we find that faulty information in nonstationary part which may be neglected still plays an important role in fault diagnosis to reveal the trend of the original signal. So Augmented DickyFuller test is utilized to divide the IMFs into stationary part and nonstationary part, and then feature selection is performed in stationary part and nonstationary part respectively. Finally, the key features of both stationary part and nonstationary part are used to develop fault diagnosis model. The efficacy of the proposed method is illustrated using the dataset from the intelligent power plant and the superiorities are shown in comparison with other two method.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call