基于LSTM-CNN的结构固有频率激励下正弦载荷识别方法研究

Research on sinusoidal load identification method under structural natural frequency excitation based on LSTM-CNN

当外载荷频率达到或接近结构固有频率时,传统载荷识别方法(比如截断奇异值分解法)的识别精度会降低。为此,通过卷积网络的特征提取和长短期记忆网络的长时记忆功能建立LSTM-CNN载荷识别模型,提出一种基于LSTM-CNN模型的载荷识别方法,对GARTEUR飞机模型开展载荷时域波形识别研究。通过采集结构的响应数据和激励数据进行模型训练和载荷识别,并与截断奇异值分解(TSVD)方法、长短期记忆网络(LSTM)方法和深度卷积神经网络(DCNN)方法的识别结果进行对比分析。结果表明:基于LSTM-CNN模型的载荷识别方法可以有效应用于结构固有频率激励下正弦载荷识别问题,具有较高的识别精度和抗噪能力。

Addressing the challenge of low identification accuracy in traditional load identification methods based on the truncated singular value decomposition(TSVD)method,especially when the external load frequency approach-es or reaches the natural frequency of the structure,the LSTM-CNN load identification model is proposed in this paper.This model combines the feature extraction capabilities of the convolutional neural network(CNN)with the long-term memory function of the long short-term memory network(LSTM).The load identification method based on the LSTM-CNN model is then applied to research load time domain waveform identification on the GAR-TEUR aircraft model.For model training and load identification,the response data and excitation data from the structure are corrected.The identification results are compared with the TSVD method,LSTM method,and DCNN method.Results show that the load identification method based on the LSTM-CNN model proves effective for sinusoidal load identification problems,especially under the natural frequency excitation of the structure.The method exhibits high identification accuracy and robust noise resistance capabilities.