DYNAMIC LOAD LOCALIZATION BASED ON DEEP NEURAL NETWORK

Accurate identification and localization of the dynamic load has significant influence on the safe and reliable operation of actual engineering structures. In order to ensure the reliability of structure design, the designers need to accurately know the location, amplitude and duration of the dynamic load acting on the structure. Only in this way can they effectively reduce or even eliminate the adverse effects of dynamic load on the structure. This article is devoted to studying the application of deep learning method in dynamic load localization. The recurrent neural network, which is widely used in the field of deep learning, is introduced into the research of dynamic load localization. The purpose of dynamic load localization has been achieved by combining the "memory" characteristics of recurrent neural network and the solving principle of vibration response. On the basis of establishing an identification model for the time history identification of impact load using Long-Short-Term Memory(LSTM) neural network, which improved from traditional recurrent neural network, further improvement was made to the network model. Bi-LSTM neural network was applied to study the localization of single point and multi-point dynamic loads. Bi-LSTM is composed of two LSTM layers stacked up and down, and its output parameters are determined by the states of these two LSTM layers. The regression layer at the end of the network is changed to a classification layer, and the label at the location where the dynamic load may appear is outputted. A simplifying the experimental model of wing structure was uesd to locate both the single point and multi-point dynamic loads. The identification results showed that the Bi-LSTM neural network model established in this paper can effectively identify the location of dynamic loads acting on the wing model, and its localization accuracy can reach over 99%.