Towards diagnostics of damage state in self-healing composites using an AI-driven acousto-ultrasonic approach

Self-healing composites are a novel class of composites which can autonomously heal matrix damages and recover their mechanical properties through external stimulus. For efficient recovery of their mechanical properties, it is essential to establish their damage state non-destructively. In this investigation, an AI-driven Acousto-Ultrasonic approach is designed to analyse the damage and heal states of self-healing composite specimens. Accordingly, artificial stress waves are generated and propagated through the self-healing composites in their different damage states and are evaluated. The stress waves in the time domain are transformed into coefficients using Mel frequency spectral analysis. The resulting Mel frequency cepstral coefficients are used to extract the underlying features in the stress waves originating from the different damage states. The features are used to train a lightweight convolutional neural network (CNN) to automatically classify the damage states. The results show that the CNN classifies the damage states in the test specimens with an exceptional accuracy of 98.66% and an F1 score of 99.18%. Therefore, this AI-driven Acousto-Ultrasonic approach has the potential to be upscaled for large structures and be used as an efficient non-destructive tool to evaluate the damage states of the self-healing composites.