Assessing Damage in Composites with Acousto-Ultrasonics and Machine Learning

Abstract

This work presents an approach to assessing the severity of damage consisting of delamination emanating from transverse cracks using features of ultrasonic waves and machine learning tools. Matrix cracking and delamination are two commonly observed damage mechanisms in composite laminates. Assessing the severity of these damage modes during service is important for structural safety. Structural health monitoring (SHM) methods are commonly concerned with detecting the location and extent of damage but do not provide quantitative measures to evaluate the ability of structures for continued safe operation. Ultrasonic waves based SHM methods, known as acousto-ultrasonics (AU), have shown great potential for in-field material inspections due to their ease of execution with simple portable sensors and their sensitivity toward damage. The transmitted waves in AU interact with damage, producing complex displacement patterns that cause alterations in the signals at the receiving sensor. The current work aims to utilize the features of those received wave signals to establish correlative relationships between them and the severity of damage for complex damage cases observed in composite laminates. The features extracted from the wave signals may not always provide direct and unambiguous measures reflecting the damage state, making it difficult to assess damage with a single signal. However, the advances in machine learning and pattern recognition make it possible to optimize the feature selection process, eliminating the need to inspect individual signal features. In this work, data from the finite element simulations of various realizations of delamination emanating from transverse cracks are used in machine learning models. It is found that these models successfully identify the signal features that correlate with the damage for the random test cases used and thereby give an excellent prediction tool to assess the damage severity.