首页 >

平稳随机载荷的信号时频特征提取与深度神经网络识别

杨特1,杨智春v2,梁舒雅1,康在飞1,贾有3   

  1. 1. 西北工业大学航空学院结构动力学与控制研究所
    2. 西北工业大学航空学院
    3. 西北工业大学
  • 收稿日期:2021-06-15 修回日期:2021-09-08 出版日期:2021-09-22 发布日期:2021-09-22
  • 通讯作者: 杨智春v

Time-frequency feature extraction and identification of stationary random dynamic load using deep neural network

  • Received:2021-06-15 Revised:2021-09-08 Online:2021-09-22 Published:2021-09-22

摘要: 本文针对线性时不变结构的平稳随机载荷识别问题,从结构的动力学响应求解原理出发,利用小波变换对于信号时频特征的提取能力与长短期记忆神经网络(Long-Short Term Memory, LSTM)对于序列问题的强大建模与映射能力,提出了一种针对平稳随机载荷的特征信号识别方法,通过对作用于三自由度振动系统数值模型上的平稳随机动载荷识别,证明了方法的可行性。对一个受两点平稳随机载荷作用的加筋壁板结构模型进行动载荷识别实验,结果表明,用本文提出的方法识别的动载荷均方根相对误差均小于5%,该动载荷识别方法具有良好的识别能力。

关键词: 平稳随机载荷, 小波变换, 深度神经网络, 动载荷识别

Abstract: Aiming at the problem of stationary random dynamic load identification for linear time-invariant structures, a feature signal identification method for stationary random dynamic load is proposed based on the dynamic principle of structures, using the ability of wavelet transform to extract the time-frequency characteristics of signals and the powerful modeling and mapping ability of Long-Short Term Memory ( LSTM ) to sequence problems. The feasibility of the method is proved by the identification of stationary random dynamic loads acting on a three-degree-of-freedom vibration system. The dynamic load identification experiment was carried out on a stiffened panel structure model under two-point stationary random loads. The results show that the root mean square error of dynamic load identified by the proposed method is less than 5%, and the dynamic load identification method has good practical ability.

Key words: Stationary random dynamic load, wavelet transform, deep neural network, dynamic load identification