Fusion of Temporal Datasets for Non-Destructive Evaluation of Reinforced Concrete Elements

Structural health monitoring has always relied on nondestructive testing for detecting deterioration in structures. Individually, various sensors like Electrical Resistivity, Ground Penetrating Radar, Half-Cell Potential, Impact Echo, and Ultrasonic Surface Waves give valuable yet sometimes incomplete perspectives which show deterioration according to one sensor reading; leaving out deterioration according to other sensor perspectives. This study aims to develop a data fusion framework where information from these diverse sensor readings is integrated to improve deterioration assessment. By aligning the sensor readings, quantifying inter-sensor dependencies, and applying graph-based learning methods, the study extracts deterioration patterns that emerge only when sensors are analyzed jointly. Field data was collected at the Bridge Evaluation and Advanced Structural Testing (BEAST) lab at Rutgers University, and the data was aligned. Using direct spatial interpolation of the raw sensor readings, thereby avoiding signal transformation, sensor readings were transformed into vectors. Using Principal Component analysis, we linearly extracted shared deterioration patterns without assumptions/ labels. Fused deterioration heat maps and quantitative deterioration index trends linking deterioration growth to calendar time interpolation. The resulting fused deterioration heat maps and deterioration index trends provided a clear visualization of deterioration progression from different sensor perspectives over time and space.