Bridge health monitoring via self-sensing concrete sensors (SSCSs): A numerical and machine learning approach

Smart self-sensing concrete sensors (SSCSs) have been proposed as an alternative to commercial strain sensors. Laboratory calibrations are typically conducted under cyclic compressive loads, which fail to fully represent real-world bridge monitoring conditions. To address this, a series of experiments was conducted using an innovative SSCS installation method, focusing on the tensile loads generated during service and the corresponding piezoresistive signals. The SSCS responses under tensile and compression loads were compared with strain measurements. A dataset obtained from laboratory test results was used for machine learning (ML) model training and validation. The performance of SSCSs as a strain sensor on bridge was evaluated in a field validated 3D finite element (FE) model. Artificial Neural Networks (ANN) were applied to convert and predict SSCS signals to strain results, including laboratory and FE model dataset. The final SSCSs were manufactured using a hybrid combination of Carbon Nanotubes (CNT) and Nano Carbon Black (NCB) to optimize sensing performance. Results confirmed that SSCSs are significantly more sensitive in tension, with a higher gauge factor (GF) of approximately 54% and 34% than compression loading for short and long hooks embedded in SSCSs, respectively. The inverse ANN, predicting strain from the sensor fractional change in resistance (FCR), achieved R 2 values of 0.90 and 0.91 for FCR-to-strain and strain-to-FCR mappings, respectively. The successful integration of the FE model and ANN validates a self-sufficient monitoring framework. This work advances the practical implementation of SSCS technology for ensuring the safety and longevity of critical bridge infrastructure.