Using a Neural Network Model to Analyze Uncertainties in Estimating the Stressed State of Metal in the Main Coolant Pipeline of an NPP with VVER Reactors under Seismic Loads

The problem of using the uncertainty analysis methodology when performing strength calculations is addressed. This issue is relevant for substantiating reliable operation of nuclear power plants with pressurized water reactors. As a consequence of measurement errors and the empirical nature of the models, the results obtained by using computer codes contain uncertainty. The use of these codes requires significant time and large hardware resources of computation stations. The use of neural network (NN) models can reduce significantly the computation time while maintaining the correctness of the results. A mode of seismic impact on a pipeline with a postulated defect is analyzed. Seismic load calculations were carried out in compliance with the requirements of the regulatory documentation (STO, NP) adopted by the State Atomic Energy Corporation Rosatom. Based on the calculation results, a stress variation time history was obtained for the entire period of seismic impact. Based on the results obtained, a neural network model was developed using special software, followed by testing it. The computation results have shown that the absolute error does not exceed 1 MPa, and the relative error is within 1%. The obtained results have demonstrated the relevance of using NN models not only for seismic load, but also for other impacts on the pipeline within the probabilistic safety assessment framework.