A Bayesian Estimation Technique for Multilevel Damage Classification in DBHM

Abstract

Structural health monitoring (SHM) is the process of deploying a network of online sensors that are leveraged to perform real-time updating on numerical models for the purpose of detecting damage. SHM has been deployed on many long-span bridges. However, direct SHM strategies using a network of on-bridge sensors have proven difficult to implement towards small to medium span bridges due to high equipment and maintenance costs. Drive-by health monitoring (DBHM) is an indirect health monitoring strategy that addresses these limitations by allowing vehicle mounted sensors to capture data across a network of bridges. This chapter begins with an overview of existing DBHM damage detection methodologies leveraging machine learning approaches. Then, a Bayesian estimation technique is presented that can reliably analyze experimental DBHM data to detect the presence of damage and accurately identify its location and magnitude without referencing a baseline set of healthy bridge data or requiring labeled data. A spike and slab prior distribution is assumed on the crack depth which allows for damage classification and, subsequently, estimation of the crack location and depth using Metropolis-Hastings steps. The performance of the method is evaluated by conducting an analytical study that considers a variety of damage states and operating conditions.