Ultrasonic based defect detection in steel-reinforced laminated timber structural elements using Uniform Manifold Approximation and Projection (UMAP)

Abstract

The increasing use of innovative engineered timber elements in construction necessitates a focus on damage identification using non-destructive techniques (NDT) to allow practitioners to make informed decisions regarding rehabilitation. This paper provides a unified approach for the application of a novel feature extraction algorithm based on Uniform Manifold Approximation and Projection (UMAP) in detecting weak zones in steel-reinforced timber beams and columns caused by common natural defects such as grain deviation and knots. Ultrasonic pressure waves are transmitted through discrete sections along test specimens of varying cross-section configurations, thereafter analysed with UMAP. The tests result in higher confidence data separation between healthy and naturally defective sections, relative to popular wavelet decomposition frameworks. UMAP is successfully applied, for the first time, for damage detection in steel-reinforced timber elements. Additionally, the proposed framework with the application of UMAP is demonstrated to be able to effectively identify and distinguish between differing defect types.