Feature-informed machine learning for detecting material deformation and failure in aluminum pipes under bending load using acoustic emission sensors

Abstract

Plastic deformation or the initiation of cracks in metal materials generates elastic wave energy, which can be captured by acoustic emission (AE) sensors. This AE energy can be leveraged for early leak detection, potentially before an actual leak occurs in metal piping systems. While much of the existing research focuses on tensile testing, limited work has been done on detecting plastic deformation or cracks during bending deformation in metal pipes using AE signals. This study evaluates and compares several feature-based machine learning techniques to identify the onset of plastic deformation or early failure in aluminum pipes under bending conditions. The results show that the average accuracy for the feature-based ML models is 79.8 %, with the Support Vector Machine achieving the highest accuracy of 83.5 %. Additionally, we propose a novel Feature-Informed Convolutional Neural Network (FI-CNN), which integrates the features into the CNN framework, yielding an accuracy of 92.7 %, outperforming the traditional machine learning methods. These findings highlight the potential of combining AE sensors with FI-CNN as an effective, non-destructive approach for real-time leak detection and predictive maintenance in piping systems.