A temperature compensation method to consider the effects of thermal inertia in anomaly detection schemes

The effects of materials’ thermal inertia need to be considered when deploying temperature compensation methods for anomaly detection in structural health monitoring. This article presents a method based on multiple linear regression between sequences of temperature and structural response that addresses this problem by taking into account the time delay effects induced by thermal inertia. Such method is first thoroughly evaluated on two numerical datasets, each consisting of temperature measurements acquired at a single structural location paired with either a “global” structural parameter (i.e., a natural frequency of vibration) or a “local” one (i.e., strain). Both datasets are generated via a finite element model of an aluminum beam-like structure in which heat transfer and structural analyses are coupled. Two types of damage are considered: a local reduction of the Young’s modulus of the material and a differential settlement. The effects of thermal inertia on the measured global and local structural responses are discussed, and the algorithm’s ability to capture these effects is evaluated. Then, the method is applied to some open-access experimental modal data acquired from the well-known KW51 railway bridge. In all considered scenarios, the method strongly outperforms the widely used linear regression between “instantaneous” pairs of temperature and structural measurements. Notably, the results demonstrate that by considering sufficiently long histories of temperature measurements, the information from even a single thermocouple can accurately predict both global and local structural responses when only thermal loads are present. • MLR is used on sequential temperature data to predict the structural response. • This is first tested on two synthetic scenarios using strain or modal parameters. • Such scenarios are simulated via a thermo-mechanical FEM of an aluminum beam. • The proposed method is then tested over the KW51 bridge’s open-source dataset. • The method offers an excellent temperature compensation performance on all datasets.