Optimal sensor placement for bridge structural health monitoring: Integration of physics-based models with data-driven approaches

Optimization of sensor configurations for cost-efficient structural health monitoring (SHM) plays a pivotal role in ensuring the safety and longevity of critical infrastructures such as bridges while minimizing maintenance expenses. Infrastructure systems may evolve over time, and their monitoring needs may change. Hence, the best sensor configuration must be designed to be scalable and adaptable to future requirements without major overhauls. The present work explores this optimization problem as an essential step of the development and implementation of a robust digital twinning strategy, by combining the strengths of physics-based models with data-driven insights. Conventional methods for sensor placement typically rely on expert knowledge, oftentimes resulting in suboptimal configurations and excessive installation costs. Conversely, physics-based models provide a rigorous comprehension of the structural behavior and can inform sensor placement procedures by assessing the impact of sensor locations on the monitoring accuracy. However, these models have limitations, such as simplifying assumptions, uncertainties, and computational complexities. To overcome these issues and upgrade the sensor placement process, data-driven approaches can be integrated to extract patterns, correlations, and unforeseen anomalies. To demonstrate the effectiveness and benefits of this hybrid framework, a case study of an existing bridge instrumented with sensors is presented. First, a simplified physics-based model is developed and calibrated through real-world data obtained from an extensive dynamic identification campaign comprising 108 instrumented degrees of freedom. Then, a reduced number of sensors are selected through a data-driven optimization strategy as best candidates for the deployment of a long-term monitoring system. Finally, the virtual model is employed to simulate varying damage scenarios and validate whether the sensor location experimentally identified as optimal would remain the best even when structural conditions change. The ultimate goal is to foster proactive maintenance strategies by striking a balance between data quality, sensor coverage, and SHM cost.