Bayesian Regression Trees as Surrogate Models for the Health Monitoring of Civil Infrastructure

Abstract

Structural Health Monitoring (SHM) of civil infrastructure often depends on algorithms which detect changes with time in measured data. However, such algorithms can be susceptible to raise false alarms if the structure changes as the result of benign operational or environmental variations. In the case of a bridge, such variations are often caused by temperature effects or traffic loading. One approach to alleviate this problem is to model the effects of the latent (environmental and operational) variables on the measured quantities and subtract them out before change detection algorithms are applied. Recently, it has been shown that Gaussian Process (GP) models are an effective means of developing appropriate surrogate models. However, the GP approach runs into difficulties if changes in the latent variables cause the structure of interest to abruptly switch between regimes. A good example here is given by the Z24 Bridge in Switzerland which completely changed its dynamical behaviour when it cooled below zero degrees Celsius as the asphalt of the deck froze. The solution proposed here is to adopt the recently proposed Treed Gaussian Process (TGP) as an alternative. The approach is illustrated here on data from the Tamar Bridge in the UK which shows marked switching behaviour in certain dynamical characteristics. It is shown that a treed linear model is in fact sufficient to solve the problem.