Research On Fault Diagnosis Method Of CFETR Blanket Remote Hydraulic Drive System Based On Deep Learning

The blanket components of the China Fusion Engineering Test Reactor(CFETR)get ruined due to high thermal loads,strong electromagnetic forces and neutron irradiation,requiring maintenance by remote handling.In the blanket remote handling system,the transfer device is responsible for the blanket transfering task,and its stable operation is an important guarantee for the normal execution of maintenance tasks.It is found that the main reason for the unstable operation of the transfer device is the hydraulic drive system failure.Therefore,it is urgently necessary to carry out research on the fault diagnosis method of the transfer device’s hydraulic drive system.To address issues such as lack of effective fault samples,difficulty in extracting fault features and fault diagnosis under variable conditions,a series of studies have been conducted,which are detailed below:1.To address the problem of lacking sufficient and effective fault samples,a method based on simulation model for obtaining fault samples is proposed.Firstly,the working principle and fault mechanism of the hydraulic drive system of the transfer device are analyzed in depth,and four typical faults and the required state signals to be collected are determined.Then,a simulation model is established to simulate the typical faults,collect the state signals,and establish a simulated fault data set.2.To address the problem of difficulty in automatically extracting fault features of the hydraulic drive system,a fault diagnosis model,MSCNN-LSTM,based on deep learning is proposed.The model integrates the advantages of convolutional neural networks and recurrent neural networks,and introduces a multi-level attention mechanism to integrate the extracted features,achieving the function of"multi-angle" feature extraction of faults.In ten fault diagnosis experiments,the average accuracy of the model is as high as 99.53%,with a standard deviation of only 0.13%,which reflects the effectiveness of the MSCNN-LSTM fault diagnosis model.3.To address the issue of potential sudden changes in operating conditions in actual operating conditions,leading to sudden changes in system state data distribution and sparse samples,a diagnosis method combining transfer learning is proposed.By using a small amount of new task data samples to fine-tune the pre-trained model,the correlation between the new task and the learned task is explored,and knowledge transfer and reuse are achieved.This enables the diagnosis model to quickly adjust and adapt to new tasks,even in situations with sparse samples.In several operating condition experiments,the diagnosis models adjusted by this method efficiently completed the diagnostic tasks,demonstrating the potential application value of this method in practical applications.4.Finally,the feasibility and effectiveness of the deep learning-based method for fault diagnosis in the transfer device’s hydraulic drive system are validated by conducting fault diagnosis experiments in the testing device.