Adaptive Approach for Sensor Placement Combining a Quantitative Strategy with Engineering Practice

Abstract

Infrastructure-capacity challenges due to growing populations, increasing urbanization and ageing existing assets are widespread. The assessment of remaining life of existing infrastructure has the potential to improve decision-making on asset management. However, this task is challenging due to the difficulties in modelling of infrastructure behavior. Error-domain model falsification (EDMF) is an easy-to-use model-based structural-identification methodology where field measurements are used to improve knowledge of the real behavior of structures. This methodology accommodates aleatory and systematic uncertainties induced by sources such as modelling assumptions, boundary conditions and numerical computation. Field-measurements, collected during load tests, lead to the identification of bridge characteristics such as geometric and material properties as well as support conditions. Benefits of structural-identification practice depend upon the methodology chosen but also upon the choice of sensor type and its location. In spite of such obvious importance, sensor types and positions are usually chosen using only qualitative rules of thumb coming from engineering experience. A more rational strategy to design optimal sensor configuration is justified and this is the aim of the study described in this paper. First, two quantitative methodologies for sensor-configuration optimization are compared with the solution designed by engineers using experience only on a full-scale case study in Singapore. The first quantitative sensor-placement methodology selects sensor locations with the largest signal-to-noise ratio in model prediction. The second strategy maximizes the joint entropy of the sensor configuration, using the hierarchical algorithm for sensor-placement. The joint entropy evaluates redundant information between possible sensor locations to select locations delivering the largest information gain when coupled. The performance of sensor configurations is evaluated using two information-gain metrics: information gain and the expected number of candidate models using simulated measurements. The hierarchical algorithm outperforms the strategy based on the maximal signal-to-noise ratio and the sensor configuration chosen empirically without calculation. However, the sensor configuration proposed by the hierarchical algorithm may be non-intuitive for practitioners. Eventually, an adaptive approach, involving engineering judgement and the hierarchical algorithm is proposed to outperform engineering judgement and to avoid non-intuitive sensor configurations proposed by the hierarchical algorithm. Results highlight that information gain metrics combined with quantitative and qualitative strategies for sensor selection and placement lead to a useful tool for asset managers.