Data fusion for compensation of temperature variations in Lamb-wave based SHM systems

Abstract

ABSTRACT Temperature variationsa ectLamb wave propagation and therefore inthis way theycan severely limitapplicationof baseline signals in SHM systems. Various techniques are proposed in the paper to solve this problem. Newmethod based on an interpretation of multiple signals acquired in distinct points of the structure is introducedand compared with other widely used approaches. Data fusion is used to merge a number of methods into onewith a substantially increased eciency.Keywords: Lamb waves, Temperature Compensation, instantaneous phase 1. INTRODUCTION Recently, Structural Health Monitoring (SHM) has received much attention of research community. Autonomousdetection and localization of damage in structures allows for considerable reduction of expenses related to theiroperation but also an increase in safety. A SHM system can involve various diagnostic techniques among thoseappliction of Lamb wave (LW) plays an important role. LWs' sensitivity to structural changes and their abilityto travel over large distances enable their use for monitoring of large structures. However, their multi-modalnature and dispersive character complicates data interpretation that becomes a dicult task.Typical approach to guided-wave based SHM is to acquire a set of signals for a structure in its intact state(baseline) and to use them as a reference in the damage detection/classi cation process. A prede ned measureof distance between the signals acquired in the reference and in an unknown state can be used as a DamageIndex (DI). This approach may appear to be ine ective in cases where measurements are noisy or changes inthe system environment occur. Measurement conduced under temperature, humidity or loading di erent fromthe reference one may result in a unjusti ed high value of DI and thus cause false positive detection. This issueseverely limits application area of the GW-based algorithms. Methods, that can handle this problem can beroughly divided into two groups: the data driven and signal driven techniques.In typical SHM application temperature or operational changes appear periodically. If structure's behaviorcould be observed under di erent operational and environmental states, one could build a database containingknowledge of the states that are normal for that structure. If such a library would be present, one coulddetermine whether the signal acquired in unknown state falls into a normal or outlier category. This idea,introduced to SHM by K. Worden