An EEMD and convolutional neural network based fault diagnosis method in intelligent power plant

Abstract

Listen

Translate article

Rotating machinery is common equipment in intelligent power plant, such as steam turbine, generator and pump. What’s more, some rotating machinery plays an important role in plants, such as steam turbine, generator and pump. Once these equipment breaks down, the whole power generation process will be paralysed. As the result of nonstationarity and nonlinearity, rotary machinery is complex and difficult to monitor and diagnosis. Traditional methods usually preprocess the vibration signal of rotary machinery firstly and then extract statistical features for fault diagnosis. However, the statistical features are hard to choose. Thus features reduction is appended to the features extraction. But it wastes a lot of human power. In this paper, to reduce the human labor for feature extraction and improve the efficiency of modeling, an EEMD and convolutional network based method is proposed to develop the fault diagnosis model using vibration signal. First, intrinsic mode functions (IMF) are obtained by the ensemble empirical mode decomposition (EEMD) which is a good way to decompose the complex signal. Next, a 1D convolution network using the IMFs is built to extract features automatically. Each IMF is regarded as a channel signal to input to the convolution network. The whole feature extract process requires little prior knowledge and is under the supervision of fault category. The efficacy of the proposed method is illustrated using the dataset collected from a real intelligent power plant.