ABSTRACT
The Robert-Bourassa/Charest (RBC) overpass was built in 1966 and demolished in 2010/2011 due to numerous signs of distress developed mostly due to alkali-silica reaction (ASR). Cores were extracted prior to RBC’s demolition, which enabled a thorough condition assessment of the structure, mainly using the Damage Rating Index (DRI). The DRI is a semi-quantitative and qualitative microscopic procedure performed using a stereomicroscope (15-16x magnification), where damage features are counted then multiplied by weighting factors, whose purpose is to balance their relative importance towards the computation of an overall damage degree of the affected concrete, the so-called DRI number. The DRI has shown to be a reliable tool to appraise ASR-affected concrete especially when combined with mechanical testing. However, performing the DRI is very time-consuming and highly relies on the petrographer’s experience making it an exhaustive process. This work therefore presents an automated DRI procedure using artificial intelligence (AI) to improve speed and reduce the variability when assessing ASR-affected concrete. The DRI’s automation process was previously accomplished by training a Convolutional Neural Network (CNN) using a pixel-level assessment, which can predict results (machine assessment) that are similar to the expected ones (human assessment). The algorithm presents an overall precision of 76% and accuracy of 74% when assessing laboratory made concrete specimens affected by ASR. In this work, cores extracted from block foundations of the RBC overpass were evaluated through the manual and automated DRI procedures. Results show that the automated DRI procedure is a reliable approach to appraise damage in ASR-affected concrete. The algorithm’s DRI number along with the crack pattern and propagation was able to capture that the degree of damage is very high for the block foundations.
