Research On Structural Damage Identification Based On Piezoelectric Sensing Technology And Deep Learnin

In the operation of civil engineering structures,the structures are subject to the combined effects of the surrounding environment,external loads and natural disasters,as well as the inevitable aging of materials,etc.,and the structures can suffer damage manifested as unhealthy conditions.In the context of big data,with the advancement of sensor technology and innovation of artificial intelligence,structural health monitoring technologies have emerged,providing more possibilities to realize automated and intelligent monitoring of structural health.In the field of structural health monitoring,how to combine sensor technology and artificial intelligence to work together is still a popular research direction.In order to accurately assess the structural health status,based on the detection method combining piezoelectric ceramic sensing technology and deep learning,an improved damage identification method of parallel convolutional neural network fusing 1D and 2D structural response data is proposed,which is studied in this paper as follows:1.By describing the cases of structural damage accidents at home and abroad,the necessity of real-time structural damage detection is illustrated.According to the existing structural damage detection methods,the shortcomings of the current research results are pointed out,according to which the research work ideas and methods of this paper are proposed.2.The development process of neural network is described,highlighting the structure and working principle of each layer of convolutional neural network(CNN),briefly explaining the traditional one-dimensional and two-dimensional convolutional neural network recognition methods,and finally proposing the framework of one-dimensional and two-dimensional parallel convolutional neural network based on vibration response and damage recognition process: proposing An improved network structure,the 1D and 2D parallel convolutional neural network model,extracts the features of the 1D time-domain signal and 2D time-frequency map,respectively,where the 1D time-domain signal retains the most original response information of the structure and the data providing the 2D features is the time-frequency map obtained by continuous wavelet transformation(CWT),and the network model transforms the damage The network model transforms the damage recognition process into a damage classification problem to achieve the recognition of structural damage.3.Using the finite element software Ansys,a steel beam with different damage levels,different damage locations,single damage and multi-damage were simulated for a total of 8damage conditions,and the piezoelectric ceramics were used as sensors to obtain the original response signals under different damage conditions.4.Based on the simulated original response signals of the structure,it is demonstrated that the detection means of combining response signals with deep learning is more advantageous in damage identification than time domain and frequency domain signal analysis methods;meanwhile,in order to verify the damage identification effect of the proposed parallel convolutional neural network incorporating 1D and 2D multi-domain signals,it is compared with the traditional 1D and 2D convolutional neural networks.The experiments show that the piezoelectric sensing technology can be used together with deep learning to accomplish the damage recognition work;the parallel convolutional neural network has stronger information feature extraction ability and better recognition effect compared with the traditional network,and the recognition accuracy of each damage condition under the same number of iterations reaches more than 95%;meanwhile,the three types of network models trained under different noise levels(1D-CNN,2D-CNN and At the same time,the improved parallel neural networks were more effective in the structural damage recognition problem,and the noise resistance and robustness of the model were significantly improved for the three types of network models(1D-CNN,2D-CNN and P-CNN)trained under different noise levels.The above study fully illustrates that sensor technology and deep learning can well solve the health detection problem of structures,while the proposed improved parallel convolutional neural network recognition effect waits for significant improvement,which can better solve the actual engineering cases.