An Improved Fast Computational Method for Vibration Characteristic Analysis of Dry‐Type Iron Core Reactor

ABSTRACT Structural mechanics field simulation plays a critical role in the vibration characteristic analysis of electrical equipment. Existing structural field analysis of equipment based on numerical simulation faces challenges such as convergence difficulties and prolonged computational time, failing to meet the demand for real‐time prediction of equipment status. In the rapid structural mechanics analysis of dry‐type iron core reactor, the sparsity of snapshot matrix will compromise the computational efficiency of proper orthogonal decomposition (POD) algorithm and introduce substantial approximation errors in reduced‐order models (ROMs). An improved POD method based on stochastic gradient descent (SGD) is proposed to realise the fast decomposition and calculation of sparse snapshot matrix in the paper. First, the SGD method is applied to the optimisation of snapshot matrix constructed from structural field simulation. Then, POD method is employed to extract the dominant reduced‐order modes and the corresponding modal coefficient. Finally, a graph neural networks‐based surrogate model is constructed to realise the rapid prediction of reactor vibration. Compared with conventional model reduction methods, the proposed method significantly improves the efficiency in handling rapid vibration simulation of dry‐type iron core reactor, and the calculation error also decreases by 14.1%.