Random vibration load identification of a cylindrical structure using data Deep Recurrent Neural Network

Abstract

A novel data-driven Deep Recurrent Neural Network (data-DRNN) method is proposed for the time-domain load identification of a cylindrical structure subjected to random vibration excitation by an electromagnetic shaker. The data-DRNN model comprises of two Long Short-Term Memory (LSTM) layers and one Bidirectional LSTM (BLSTM) layer, and is trained using a large dataset of loads and corresponding responses data. The effectiveness and accuracy of the proposed method have been validated by analyzing the trapezoidal spectrum excited response data under various temperature conditions. Furthermore, the model's generalization capability has been evaluated by examining different testing target spectrums. The results indicate that data-DRNN has excellent accuracy and generalization ability, making it a promising choice for load identification of random vibration.