Explainable artificial intelligence for deep learning-based delamination detection in Carbon Fiber Reinforced Polymer composites

Abstract

The heterogeneous and anisotropic nature of Carbon Fiber Reinforced Polymers (CFRP) introduces challenges in characterizing failure mechanisms, particularly delamination, in advanced composite engineering structures used across multiple industrial sectors. Although deep learning models, especially Convolutional Neural Networks (CNNs), have demonstrated high diagnostic performance in Structural Health Monitoring (SHM), their inherent black-box nature limits interpretability and trust, thereby restricting deployment in real-world engineering applications. To address these limitations, this work proposes a methodology that integrates deep learning with Explainable Artificial Intelligence (XAI) to enable transparent and reliable delamination detection in CFRP structures. The main contribution in Artificial Intelligence lies in the integration of convolutional neural networks with time–frequency representations of guided-wave signals and model-agnostic explainability techniques for intelligent fault detection in composite materials. From an engineering perspective, the proposed approach is applied to SHM and delamination diagnosis of CFRP specimens subjected to fatigue loading in engineering systems using Lamb waves acquired by piezoelectric transducers. The recorded Lamb wave signals were transformed into time–frequency representations and used as inputs to multiple CNN architectures, evaluated under a leakage-controlled, leave-one-cycle-out cross-validation protocol with multi-seed statistical aggregation, achieving accuracy values of up to 95.9% with statistically validated inter-architecture comparisons The explainability analysis highlighted physically meaningful indicators of delamination, such as amplitude attenuation and frequency shifts, demonstrating the potential of the proposed framework as an intelligent decision-support tool for fault detection and monitoring in engineering applications involving composite structures.

The authors' abstract, as published at the source. Engineering Applications of Artificial Intelligence, 2026 · DOI ↗