Research On Compressive Sensing Method For Dynamic Structural Monitoring Data

Structural health monitoring(SHM)techniques are widely used in modern civil engineering to ensure structural safety.The structural damage and degradation can be identified based on the timely measurements of structural load and response,which are obtained by various sensing devices deployed on the structures.However,the sampling frequency and lifetime of SHM systems are limited by energy resources,transmission bandwidth,and communication costs,etc.Compressive sensing(CS)is an efficient and novel sensing technology that can reconstruct the SHM data based on a few samples exploiting the sparsity.CS can help improve the sampling efficiency of the SHM system and reduce the energy and transmission bandwidth requirements.Therefore,in this paper,several novel CS methods are proposed for dynamic data in SHM(vibration response data,wind field data)which is often acquired with high sampling frequency.The proposed methods are verified by numerical simulation,experiments and engineering practice,respectively.CS technology is embedded in the SHM system for use in engineering practice applications.The main research contents are as follows.1.A novel Bayesian CS method is proposed for the cooperative reconstruction of the vibration response data acquired from multiple measurement points.The correlation between vibration response data is exploited to achieve cooperative CS of multiple data sets using the multitask Bayesian compressive sensing(MT-BCS)method.By separating the real and imaginary parts of the sparse projection,the original MT-BCS method is improved to solve the ”realcomplex” conversion problem caused by the discrete Fourier transform(DFT).The data reconstruction accuracy of the original MT-BCS method is improved by sharing a sparse prior for the real and imaginary parts.By sharing the measurement matrix among the sensing devices,the computation time of the original MT-BCS method is reduced to the single-task level,which greatly improves the computational efficiency.The improved MT-BCS model was tested through a shaking table test and the SHM application of Guangzhou Tower.The results show that the improved MT-BCS model has the advantages of both high accuracy and high efficiency compared with the traditional model.2.A changeable dictionary-based method for CS of wind field monitoring data is proposed.A novel changeable dictionary based on sliding window is proposed to realize the sparse representation of wind field monitoring data.The changeable dictionary is designed by exploiting the correlation and lag between the data from different measurement points.The changeable dictionary is used with LASSO to reconstruct the wind field monitoring data.The proposed method is validated using wind speed data collected from the roofs of Hangzhou East Railway Station and Zhejiang University Stadium.The results show that a high compression ratio can be achieved by using this changeable dictionary.The statistical characteristics of the reconstructed wind speed signal are consistent with the original ones,which meets the requirements of SHM applications.3.A deep convolutional generative adversarial network(DCGAN)based method is proposed for the reconstruction of data.The DCGAN is the first deep network dedicated to the CS of SHM data,which is composed of a 1D U-net with shortcuts(generator)and a 1D convolutional classification network(discriminator).To improve the reconstruction accuracy,a composite adversarial loss function is proposed,which includes reconstruction errors in the time and frequency domains as joint optimization objectives.The DCGAN is trained and tested with simulation data and experimental data collected from a steel grandstand.The results show that compared with the sparsity-based CS method,DCGAN has the advantages of high computational accuracy and efficiency.Moreover,DCGAN does not require sparsity and random measurement.4.A CS platform for SHM application is developed.The scheme of CS based on random demodulation,including pre-storage of pseudo-random chipping sequences,demodulation by complementary codes,and accumulate-and-dump filtering,and is designed and embedded into smart wireless sensors.The sampling experiment using the improved sensor shows that CS can effectively reduce the energy consumption and transmission bandwidth requirements of wireless sensors.A cross-platform software is developed for the application of CS in SHM.The developed software has a beautiful graphical user interface and supports a variety of functions such as signal reconstruction,data management,sensor control,data visualization,etc.,which can meet the complicated needs of SHM practice.