Sequential Bayesian Comoressed Sampling And Applications In Structural Health Monitoring Signals And Images

Since the adoption of reform and opening-up policy,Infrastructure construction of China was facing great opportunities.However,the service environment of large-scale infrastructure is complicated,which not only suffers from extreme effects such as earthquake and typhoon,but also inevitably suffers long-term effects such as environmental erosion.Therefore,it is an important task and goal of civil engineering to implement structural health monitoring(SHM)techniques,to ensure and improve the safety performance of civil structures.It is known that civil structures are complex and large scale with many degrees of freedom,and therefore a large number of sensors are usually installed for a SHM system.The continuous monitoring system of the large-scale infrastructure often collects massive data,resulting in inefficient signal transmission and storage.In addition,because of the limitation of battery energy,wireless sensor networks cannot collect,store and transmit large amounts of data.Efficient signal compression algorithms thereby are important for SHM.Compressed sampling,as a brand-new signal compression and processing technique,exploit the sparseness of a signals with respect to some domains for accurate reconstructing it.The compressive sampling maeasurements can be far fewer than required by the Shannon-Nyquist sampling theorem.In the last decade,the technique has shown application prospects in many fields,including structural health monitoring,however,how to achieve accurate data decompression with high compression rate is still a challenging issue.It is known that the structural state is relatively stable during the service life and the structural response in adjacent time periods tends to be smooth,that is,the signal changes with time also be sparse.Based on this,sequential Bayesian compression sampling is investigated in this study.The main research contents are as follows:In the second Chapter,based on the application requirements of compressive sampling and signal reconstruction of continuous change signals,a hierarchical Bayesian learning model is established and the constraints of signal dynamic sparseness are embedded effectively,that is,the original signal be sparse and,in addition,the changes of the signals between adjacent time instants also be sparse.Based on the Kalman filter equations,a sequential Bayesian compression sampling method for dynamic sparse systems is proposed.The posterior mean and covariance matrix of the reconstructed signal at each time instant are computed.At the same time,the real-time estimation of the noise parameters of the Kalman filter equations for each time instant is investigated to avoid the inaccuracy problem of Kalman filter estimation.Therefore,real-time or nearreal-time signal reconstruction at time instant is realized.In the third Chapter,the sequential Bayesian compressive sensing algorithm is established from the theoretical analysis in the previous chapter and several acceleration strategies are also investigated.Based on the synthetic signals,a comparison study is performed with the traditional Bayesian compression sampling method for different cases.The results show that the established algorithm can achieve accurate signal reconstruction with higher compression rate for each time instant.In the last Chapter,by using the real data of one-dimensional health monitoring signals and two-dimensional images,the superior performance of the proposed algorithm compared with the traditional BCS methods is also verified.The results show that for both the highly and approximately sparse signals,the improvement of signal reconstruction performance can be achieved.Therefore,the proposed method has broad application prospects in data compression and reconstruction of structural health monitoring signals,the recovery of lost data in wireless sensor networks and the image acquisition and compression of computer vision.