Session: 01-08: Nuclear Plant Operation, Modification, Life Extension, Maintenance and Life Cycle - VIII

Paper Number: 135905

135905 - Research on the Reactor Coolant Pump Fault Diagnosis Based on Typical Fault Mode Test and Deep Learning Algorithm Model 

Abstract: 

The reactor coolant pump (RCP) is the key equipment of the reactor coolant system and is part of the pressure boundary of the system. equipment structure of RCP is also rotating equipment, its structure is complex, technical difficulty, including machine, electricity, water, gas, oil system. The normal operation of reactor coolant pump is directly related to the safety of nuclear power plant, and the identification of its failure mode requires the expert knowledge of operation and maintenance personnel. At present, the RCP relies on the threshold alarm, and the fault logic judgment is single. In the case of failure without early warning, there is not enough time to analyze or judge when the fault occurs, which brings great pressure to the power plant operator.

Through investigation, most of the forced shutdown and planned delay events caused by RCP failure in nuclear power plants domestic are caused by the failure of RCP shaft seal and thrust bearing. Shaft seal assembly and bearing assembly are extremely expensive and frequently replaced parts of RCP. At present, it uses the way of planned maintenance, and there are unreasonable maintenance situations such as premature replacement or replacement not in time. Therefore, it is necessary to adopt appropriate simulation test and deep learning algorithm model for fault diagnosis and life prediction of key components of RCP, which can lay the foundation for upgrading key components of RCP from planned maintenance to preventive maintenance, so as to improve the operation and maintenance economy.

Due to the few reactor years of operation of nuclear power plants in China, no large database related to RCP operation and maintenance has been formed. However, the failure mode test with all components and the full flow is costly and the cycle of the test is very long. According to the sensor type, measuring point distribution and acquisition frequency of RCP in the nuclear power plant, the typical failure mode of the key vulnerable parts of RCP is investigated and sorted out. Then, the test bench of bearing and shaft seal components is set up respectively to test the operation of bearing and shaft seal components in several typical failure mode, and collect their multi-channel fault sample data.

In order to solve the problems of poor prediction and diagnosis accuracy, poor anti-noise performance and weak generalization ability of traditional neural network, a multi-information fusion method based on wavelet decomposition, gated circulation unit (WD-GRU) and improved attention mechanism is proposed in this paper.

Firstly, the original data of each channel is denoised by wavelet decomposition, and then the timing features of each sensor signal are extracted by GRU, and a multi-channel attention mechanism is introduced to adaptively fuse the features of different channels. Finally, fault diagnosis is realized by classifier.

The test results based on the multi-channel data set (including temperature, acoustic emission signal and wear amount) of RCP's key components (bearing and shaft seal) test bench show that: compared with the single sensor model, the diagnosis rate is significantly improved; Compared with the model without the introduction of attention mechanism, the diagnosis rate is improved to some extent. Compared with classical machine learning, deep learning and improved algorithms based on deep learning in recent years (CNN-LSTM, AUTO-ARIMA, KNN and SVM, etc.), the diagnostic model designed in this paper has the highest diagnostic rate and stronger stability. The achievement of the research can play an important role in improving the operation and maintenance economy and intelligence level of RCP in nuclear power plants.

Keywords: Reactor coolant pump; Fault diagnosis; Operation and maintenance; multi-information fusion ; improved attention mechanism ; Nuclear power plant.

Presenting Author: Cui Huaiming Nuclear Power Institute of China

Presenting Author Biography: I have been working in China Nuclear Power Research and Design Institute since 2019, mainly engaged in the design of the coolant pump and valves for nuclear power reactors. I have a certain understanding of the application of PHM technology in the field of nuclear power. During my work, I participated in the design and development of the intelligent monitoring and fault diagnosis system of the the reactor coolant pump, as well as some of the design work of the health management and intelligent operation and maintenance system of the reactor and primary circuit of the nuclear power plant.

Authors:

Cui Huaiming Nuclear Power Institute of China
Kuang Chengxiao Nuclear Power Institute of China