Speakers
Description
Structural integrity assessment of concrete infrastructures is increasingly reliant on non- destructive testing (NDT) techniques capable of delivering real-time, in-situ information about damage evolution. Among these, Acoustic Emission (AE) stands out as a passive monitoring method that captures stress-induced ultrasonic wave emissions generated by internal material changes, such as microcracking. AE offers a unique advantage: it allows volumetric, continuous monitoring of structural elements without invasive procedures. However, a critical challenge in AE-based monitoring is the vast volume of signal data generated during structural loading, which complicates manual interpretation and reduces the practicality of the technique in operational settings. To address this, Machine Learning (ML) methods have been introduced to support automated signal classification and pattern recognition. In this work, we present experimental results from reinforced concrete beams subjected to four-point bending until failure under controlled conditions. AE signals were continuously recorded using a sensor array, and parameters such as amplitude, energy, and duration were extracted. These signals were then analyzed through a trained Multilayer Perceptron (MLP) neural network model to classify AE events into “cracking” and “non-cracking” categories. The approach was specifically designed to enhance real-time interpretation of AE data in structural applications. The results show a strong correlation between AE signal characteristics and the evolution of damage in concrete elements. The ML approach significantly enhanced the detection accuracy, enabling automated, real-time identification of crack initiation and propagation. These findings highlight the combined power of AE and AI for effective structural health monitoring and early warning systems in aging concrete infrastructure.