구조동역학에서 iIRS방법을 활용한 효율적인 인공신경망 접근 방안

Abstract

Artificial neural network approaches are used to efficiently generate meta-prediction fields for structural dynamics problems. However, these approaches exert heavy computational burden, which makes it difficult to improve the quality of the prediction fields. Therefore, we propose an artificial neural network strategy for structural dynamics problems using the iterated improved reduced system (iIRS) method. In the proposed method, characteristics of structural data are first extracted using the transformation matrix of the iIRS method. Next, the neural network (NN) is trained using only the extracted features. The prediction fields are restored by combining the trained NN results with the transformation matrix in the iIRS method. As a result, the quality of NN for structural dynamics problems is significantly improved owing to the efficient computational procedure. The performance of the proposed method is verified using the gearbox-housing model in an electric vehicle.