Enhancing concrete strength monitoring via deep learning fusion of non-destructive testing data

Accurate monitoring of concrete strength evolution is critical for construction safety and timeline optimization. Traditional Non-Destructive Testing (NDT) methods, such as Ultrasonic Pulse Velocity (UPV) or Rebound Hammer, often suffer from low accuracy when used in isolation due to the influence of aggregate types and moisture content. This study employs a Self-Normalizing Neural Network (SNN) to fuse multisensor NDT data for predicting compressive strength. The model utilizes a dataset of 4,420 monitoring points from concrete mixtures containing various aggregate types (including recycled and volcanic) and additives. The input variables include Curing Age, Ultrasonic Pulse Velocity (UPV), and Electrical Resistivity, while the output is Compressive Strength. Results indicate that the Deep Learning fusion model significantly outperforms traditional regression curves, achieving high accuracy (> 0.90) by effectively capturing the non-linear relationships between NDT metrics and strength development. This approach offers a non-invasive, sustainable method for verifying structural integrity in aggressive environments . Crucially the analysis identifies a specific “High Risk Zone” where concrete exhibits adequate structural strength (>30 MPa) but critically low electrical resistivity. This discrepancy highlights a matrix that is mechanically sound yet highly permeable to ionic ingress, identifying vulnerabilities to acid attack that standard strength testing would miss. These findings validate the SNN framework as a dual-objective monitoring tool for ensuring the resilience of wastewater infrastructure.