Study of diagnosis for rotating machinery in advanced nuclear reactor based on deep learning model

Abstract

Many types of rotating mechanical equipment, such as the primary pump, turbine, and fans, are key components of fourth-generation (Gen IV) advanced reactors. Given that these machines operate in challenging environments with high temperatures and liquid metal corrosion, accurate problem identification and health management are essential for keeping these machines in good working order. This study proposes a deep learning (DL)-based intelligent diagnosis model for the rotating machinery used in fast reactors. The diagnosis model is tested by identifying the faults of bearings and gears. Normalization, augmentation, and splitting of data are applied to prepare the datasets for classification of faults. Multiple diagnosis models containing the multi-layer perceptron (MLP), convolutional neural network (CNN), recurrent neural network (RNN), and residual network (RESNET) are compared and investigated with the Case Western Reserve University datasets. An improved Transformer model is proposed, and an enhanced embeddings generator is designed to combine the strengths of the CNN and transformer. The effects of the size of the training samples and the domain of data preprocessing, such as the time domain, frequency domain, time-frequency domain, and wavelet domain, are investigated, and it is found that the time-frequency domain is most effective, and the improved Transformer model is appropriate for the fault diagnosis of rotating mechanical equipment. Because of the low probability of the occurrence of a fault, the imbalanced learning method should be improved in future studies.