Research on Measurement and Analysis Method of Radiation Test Section Based on Neural Network

Abstract In the irradiation test equipment, frequent operations such as material replacement, irradiation device surface replacement, and entry and exit of the reactor are carried out. These factors may lead to sensor failures, resulting in inaccurate test results. During the irradiation test process, the fault diagnosis is usually carried out by operators by observing the parameters during operation and combining their own knowledge. The advantage of using neural networks for fault diagnosis lies in their direct and effective statistical analysis and information extraction of massive, multi-source, and high-dimensional data. This method collects different types of data as the basis, uses the implicit effective information to characterize the operating mode and status of sensors, and thus achieves the purpose of detection and diagnosis. In this study, a neural network fault diagnosis model was established to detect and diagnose the faults of sensors. The fault diagnostic model is trained based on experimental data, and the model is synchronously trained online during the irradiation test process to continuously improve diagnostic efficiency. The fault diagnosis model adopts an unsupervised autoencoder, which takes experimental data as the target for learning. The model trained using obtained from irradiation experiments performs well in simulation tests, and can rapidly and accurately determine the data and process related experimental data when sensor faults occur, providing the location and type of sensor faults, and intuitively and quickly prompt operators with effective information.