Evaluating the Reliability of Carbon-Fiber Composite Through Applying a Tree-Based-Ensemble Method to Analyze Acoustic-Emission Signals

Abstract

This paper outlines the creation of a non-destructive testing (NDT) technique that incorporates advanced machine learning (ML) methodologies to enhance damage detection and diagnosis capabilities in various industrial sectors. The selected NDT approach for identifying damage in composite structures is based on acoustic emission (AE) principles. The diagnosis process involves the analysis of acoustic signals obtained through a rigorous experimental procedure. A tree-based-ensemble technique, rooted in ML methods, was employed to process the acquired acoustic dataset. The application of this tree-based-ensemble technique in the diagnosis of composite structures consists of two critical steps. The first step entails the collection of an experimental dataset that provides a comprehensive AE-based dataset for a specific composite structure sample, particularly one composed of carbon fiber (CF) composite. The second step includes the signal processing of the collected dataset utilizing the tree-based-ensemble technique. The results generated from this ensemble-based approach demonstrated significant performance when compared to the diagnosis outcomes produced by the Vallen-AE suite software for data acquisition. Besides, it outperforms other ML-based methods in terms of many metrics evaluated. This developed methodology offers innovative and effective solutions for quality inspection across various industrial applications, achieving an accuracy rate exceeding 96%.