Automated estimation of early-age concrete compressive strength using EMI signature-driven deep learning technique

Abstract

Concrete is a vital construction material for civil infrastructure, whose early-age compressive strength estimation plays a critical role in determining the proper time for formwork removal and the readiness of the structures for service. Electromechanical impedance (EMI) technique has been proven to be a non-destructive and effective method for monitoring concrete early strength gain. Nonetheless, such a technique heavily relies on effective frequency band selection and statistical parameter-based analysis, which may prevent the technique from a real-time operation. This study proposes an EMI signature-driven deep learning (DL) technique to enable the automated estimation of early-age concrete compressive strength. A novel DL framework called multi-scale learning residual denoising network (MLRD-Net) was developed to autonomously learn sensitive features from raw EMI signatures and ultimately estimate the compressive strength. The EMI signatures were collected during the strength development process of concrete specimens produced from two different mix designs based on strength. Two independent databases were constructed from recorded EMI signatures to test the effectiveness of the MLRD-Net. The results demonstrated the excellent performance of the proposed model with coefficients of determination exceeding 0.99 for both databases. In addition, the MLRD-Net showed superior performances over other methods in terms of prediction accuracy, noise resistance and fitness capacities, indicating its great potential for real-time and accurate estimation of the strength gain of early-age concrete.