Bridge Health Monitoring Data Preprocessing Based On Data-driven

The bridge health monitoring technology has been formed a relatively perfectsystem. It is because of the rapid development of sensor technology, communicationtechnology and computer technology. The vast amounts of data that collected by thebridge health monitoring system is the premise and basis of the evaluation of bridgehealth status. However, in the study of the theory of the bridge health monitoring, moreand more research is focused on the optimal sensor layout, the design of the sensornetwork architecture and the algorithm of the bridge structure safety assessment. Theyalways ignore the importance of the bridge health monitoring data preprocessing. It islikely to lead to the diagnosis of “garbage in garbage out” results. So the bridge healthmonitoring data preprocessing is very important to the bridge health monitoring.Considering the data of bridge health monitoring has Characteristics such as largevolume, complicated structure, etc. The paper use data-driven method. The methodrefers to a class which don’t need accurate mathematical model of a system known inadvance, it based on its huge amounts of data that can realize the fault diagnosis andcontrol, etc. The technology, which including much theory knowledge. This paper chosethree of theoretical knowledge for research and application: wavelet analysis, waveletpacket decomposition and principal component analysis. In this paper, A fault diagnosismodel is established. The model contains three algorithms, wavelet threshold noisecancellation algorithm, fault detection based on wavelet packet energy curvaturedifference algorithm and the algorithm of fault diagnosis based on principal componentanalysis. Three algorithms cooperate with each other to complete the pretreatment workof bridge health monitoring. This paper’s main work includes the following aspects:First, the study of the importance of data preprocessing in bridge health monitoring.Comparison in data preprocessing methods. Highlights the importance of data-driven.Second, described the theory of knowledge, including the principle of waveletanalysis and wavelet packet analysis, wavelet packet energy curvature calculationprinciple, principle of principal component analysis.Third, design an improved wavelet threshold noise cancellation algorithm, andcomputer simulation has been accomplished.Fourth, design a fault diagnosis model based on data-driven method. It combined with the proposed improved wavelet threshold noise cancellation algorithm, using analgorithm which combined with wavelet packet energy curvature difference method andprincipal component analysis, fault detection and diagnosis faults form huge amounts ofdata of sensors and fault degree.Fifth, a computer simulation has been accomplished to design model of datasimulation experiments.