Acoustic emission sensing for assessing compressive strain in impacted thermoplastic composites

Abstract Composite materials used in aircraft during in-flight service are vulnerable to impacts, which can lead to undetected damage and deterioration. Impact incidents can cause considerable structural harm that might not be immediately apparent, resulting in a decline in the mechanical properties of material and overall performance. There are currently limited advanced models capable of assessing the state of compressive strain (or stress) in a compression after impact (CAI) test based on acoustic emission (AE) data. Developing such a model may enable real-time monitoring and evaluation during in-flight service, providing users with alerts when composite material failure is approaching. This paper presents a method for assessing the state of compressive strain for an impacted specimen using AE monitoring. The main objective of this study is to evaluate the level of straining on impacted composite specimens by investigating AE signals generated during the CAI test and how a material can resist further loads prior to failure. CAI tests were performed using various impact energy levels, with analysis concentrated on the most critical scenario (impact energy = 30 Joules), where the impact-induced damage is barely detectable and the compressive strength of the material is insufficient to withstand additional compressive loading. An enhanced algorithm is suggested to estimate the likelihood that the applied strain falls within different categories, offering an evaluation of the material condition. To develop this method, a composite panel was fabricated and subjected to a controlled impact test, followed by a post-impact compression test. AE signals recorded during the experiment provided initial validation of the proposed approach, demonstrating their capability for accurately predicting applied strain. The findings of this study suggest that integrating AE with machine learning (ML) algorithms can provide an effective solution for evaluating the structural health of composite materials and guaranteeing their dependability in crucial applications. Results indicate that among the evaluated ML models, extreme gradient boosting and random forest outperformed the artificial neural network.