Multi-physics Coupling Prediction of Gas-cooled Reactor Based on POD-ANN

The gas-cooled reactor has much advantages such as high power density, compact core, high system efficiency, and strong adaptability, which is an important candidate solution for breaking through the energy bottleneck in the future. Considering the interaction between different physical fields in the core of the gas-cooled reactor, there are certain challenges in the safety design of gas-cooled reactors. Therefore, it is necessary to construct coupling between multiple physical fields such as nuclear, thermal and mechanical fields in the gas-cooled reactor, to analyze the behaviors in the operation process. However, in conventional core multi-physics coupling calculations, the efficiency of data exchange and grid mapping is low, consuming a large amount of computing resources. Artificial neural network, as a method with strong nonlinear fitting ability, is a reliable choice for quickly obtaining multi-physics coupling results. This work focuses on modeling a small gas-cooled reactor and uses RMC and ANSYS Mechanical for nuclear-thermal-mechanical coupling to analyze the characteristics of the core. A core nuclear-thermal-mechanical coupling surrogate model based on proper orthogonal decomposition and artificial neural network (POD-ANN) was proposed. Based on the results of nuclear-thermal-mechanical coupling calculations, the main features of the data were extracted using the POD method, and the data was mapped to a low dimensional space to achieve dimensionality reduction. A fully connected feedforward neural network model was built and trained using reduced dimensional data to obtain surrogate models of temperature field, stress field, and displacement field, which achieves rapid reconstruction and prediction of the nuclear-thermal-mechanical coupling characteristics of the core. The results indicate that the constructed nuclear-thermal-mechanical coupling surrogate model captures the main characteristics of the coupled physical field. When a certain proportion of accumulated energy is required and the number of selected feature modes is small, the model’s prediction performance in the test set is better. Compared with the stress field and displacement field, the temperature field surrogate model has better prediction performance. The surrogate models are used for physical field prediction, where the peak temperature error of the core is only 5.50 K, the average error is only 1.75 K, the peak stress error and peak displacement error are 12.22 MPa and 0.009 mm and the average errors are 1.47 MPa and 0.026 mm, respectively. The average relative errors are all less than 3%, just a small error, which indicates that the model has certain effectiveness and feasibility. This method significantly saves the time and computational resources required to obtain the results of multi-physics field coupling in the core, and provides new ideas for the application of multi-physics field coupling in transient analysis.