Deep Learning Based Structural Reliability Finite Element Analysis Surrogate for Hydro-Generator Lower Frame

Abstract

The lower frame is one of the most important components of hydro-generator. Its reliability directly affects the service life of hydro-generator, thus further impacts the power generation, which is related to the national economy and people's livelihood. Currently, finite element analysis (FEA) is the most used method to perform structural reliability analysis. However, due to the huge size and complex structure of hydro-generators, it usually takes minutes or even days to complete one FEA simulation. In this paper, a novel deep learning based structural finite element analysis surrogate model is developed to speed up the structural analysis process. The results show that for the same structural analysis problem, our surrogate model can predict the elastic strain and stress distributions with accuracy of 93.5% and94.7%, respectively. Besides, the inference time of the model is less than 1 millisecond which is much faster than the FEA method. With the outstanding computing speed, our surrogate model can also be deployed in the hydropower station to perform real-time structural monitoring.