Research On Several Key Technologies Of PHM For Pump Machinery In Nuclear Power Plant

As typical active equipment,rotating machinery such as pumps are widely used in nuclear power plants(NPPs),such as the reactor coolant pump in the primary loop,the main feed water pump,circulating water pump and condensate pump in the secondary loop.The main function of pumps in NPPs is to convert the mechanical energy provided by the prime mover into the energy of the conveyed fluid,generally into kinetic energy and pressure energy.The realization of its function depends on the rotating parts and their rotation.Although the typical components and mechanism of NPPs pumps are similar to conventional pumps,their safety and reliability requirements are higher in complex operating environments.Once significant performance degradation or failure occurs,it may cause huge safety hazards and economic losses.Generally,the digital instrumentation and control system is equipped with a large number of sensors for critical equipment to ensure normal operation.However,the preventive maintenance and fault maintenance strategies still have some problems,such as relying on manual experience,inflexible maintenance and high cost.Prognostics and health management(PHM)technology can provide operation support based on equipment condition assessment and prediction,to realize the strategy transformation towards condition-based maintenance and predictive maintenance.PHM technology can reasonably improve the economy of operation and maintenance while ensuring the safety.In this paper,the research on intelligent diagnosis and fault prediction technology of pumps is investigated in combination with PHM technology,including intelligent diagnosis and fault prediction scheme,abnormal state detection method,fault diagnosis method and fault prediction method.The main research contents include the following aspects.(1)To improve the problems of difficult state assessment,low efficiency and insufficient intelligence level in the operation and maintenance of pumps in NPPs,an effective intelligent diagnosis and fault prediction scheme is proposed with the analysis of main structure and abnormal mode characteristics.The system functions and information flow are described,and corresponding solutions are proposed around the key technologies of state detection,intelligent fault diagnosis and fault prediction,to promote the transformation towards condition-based and predictive maintenance strategies.(2)State detection is the basis of intelligent diagnosis and fault prediction scheme.Considering that there are many detection parameters of pumps and the state feature extraction is very important,a detection model DDAE based on deep denoising autoencoder network is constructed.DDAE can achieve early anomaly detection and positioning while ensuring a low false alarm rate.To alleviate the time-varying effect of parameters in long-term detection,equidistant sampling and DDAE model are combined to enhance detection efficiency.Case tests are carried out using simulation data and actual data,and the detection and positioning capabilities of the proposed scheme is verified under normal and abnormal conditions.(3)Aiming at the problems of relying on empirical knowledge,low accuracy and nonunique results in the NPPs fault diagnosis,an intelligent model DCNN is proposed,which combines multi-sensor fusion strategy with deep convolutional neural network.The continuous wavelet transform is introduced to process multi-sensor signals to avoid the limitation of single monitoring.Deep residual neural network is applied to enhance the feature extraction and classification performance.The advantages of the model in multi-sensor fusion and feature extraction are verified by using the experimental data of rotating shafts and motors.(4)In order to enhance the model generalization performance of pumps under variable speed and unlabeled data,a fault diagnosis model HDAL based on deep transfer learning is proposed to achieve a breakthrough in the intelligent fault diagnosis model of NPPs.The model uses multiple sensors to comprehensively identify the operating status,and combines the convolutional neural network to adaptively extract fault features.More importantly,the HDAL model adopts a transfer strategy of generative adversarial network driven by subdomain adaptation to carry out fault diagnosis tasks at different speeds.Case tests show that the HDAL model can significantly reduce feature distribution difference,and show strong fault diagnosis and anti-noise performance on the basis of multi-sensor information fusion.(5)To solve the problem of fault diagnosis performance degradation under imbalanced samples,a deep convolutional conditional generative adversarial network(DCCGAN)is proposed.To improve the imbalanced sample augmentation performance of the model,the structure and training method of the traditional model are improved with the help of conditional generation adversarial strategy,thereby alleviating the constraint problem of sample generation.In addition,a deep convolutional network is constructed to extract the features of the vibration signals to achieve fault classification.Finally,combined with the proposed comprehensive evaluation scheme and case tests,the advantages of the method in training performance and imbalanced sample diagnosis are verified.(6)In the fault prediction of pumps,a hybrid prediction model EWT-GRU combining empirical wavelet transform(EWT)and gate recurrent unit(GRU)is proposed to solve the problems of low prediction accuracy,insufficient extraction of time-related information and multi-step prediction.The model can overcome the problem of low prediction accuracy of single model and significantly enhance the performance of signal prediction.In the multi-step prediction strategy,the sliding time window technology is combined to perform multi-step recursive prediction,to alleviate the error accumulation.Through case tests such as single-step and multi-step prediction,strong prediction performance is shown,which lays the foundation for life prediction and predictive maintenance strategy.In summary,through the above method research,theoretical modeling,strategy optimization and case analysis,combined with PHM technology,the research results of this paper provide an effective solution for operation state detection,fault diagnosis and prediction of pumps.This has important application reference significance for improving the operation and maintenance efficiency of key equipment in NPPs,and helps to promote the process of condition-based maintenance and predictive maintenance of equipment.