Data Compressive Sensing And Reconstruction Based On Group Sparse Optimization For Structural Health Monitoring

With more and more large civil infrastructures have been installed the structural health monitoring（SHM） systems, the cost of data collection, storage and transmission gradually attracted people’s attention. Otherwise, due to various factors, a lot of SHM data has some errors. Compressive sensing（CS） theory provides a good idea for compressive sampling and data reconstruction of SHM. Traditional CS data reconstruction algorithm is only for single sensor measured signal, using the sparsity in the frequency domain to reconstruct the data from the incomplete measurements. But in the actual SHM system, there are multiple sensors on the same structure. These measurements of the multiple sensors have spatio-temporal correlation and similar sparsity in the frequency domain which is called as group sparsity. In order to improve the reconstruction precision of the CS data and the error data of SHM, this thesis propose a method of CS data reconstruction based on group sparse optimization.The main contents are included as follows:A new method of CS data reconstruction based on group sparse optimization is proposed. The monitoring data of the multiple sensors have spatio-temporal correlation, that is the structural vibration response data have similar sparsity in the frequency domain which is called as group sparsity. Take advantage of this feature, the study introduced the group sparse constraint(minimize the （?）_p,q norm, which p（28）2, q（28）1) into CS data reconstruction process and improved data reconstruction precision. According to the augmented Lagrange multiplier method, an iterative algorithm is established.To investigate the CS data reconstruction accuracy of proposed group sparse optimization algorithm, an field test on Xiamen Haicang bridge wireless sensor is carried out. Using wireless sensor collect vibration data of Xiamen Haicang Bridge, first the CS process is simulated; then the group sparse optimization algorithm and single sparse optimization algorithm are used to reconstruct the CS data, respectively. The comparative study of the reconstruction accuracy between these two optimization algorithms are implemented. The influences of the data reconstruction error on modal parameters identification are aslo studied.A new method of the error data reconstruction of SHM based group sparse optimization is proposed. To investigate the data error modes, a large number of actual monitoring temperature, humidity, acceleration, strain, GPS displacement data are statistical analyzed. Based on the understanding of error data, with considering the group correlation and non-uniform of error data, the group sparse optimization algorithm is employed to reconstruct the error data. First, the perfect actual SHM data is employed to simulate three types of data errors, such as random outliers, small pieces of continuous data anomalies and the integration of two kinds of abnormal types. Then the group sparse optimization algorithm is used to reconstruct these three kinds of simulative error data. Finally, comparison of the reconstruction result with the original data are implemented to verify the validity of the proposed method.