Stationary Wavelet Transforms with Binary Trees for Acoustic Emission

• Lucas Lisboa dos Santos Instituto SENAI de Inovação

Resumo

This work proposes a method for feature extraction from acoustic emission signals using a binary tree structure based on the Stationary Wavelet Transform (SWT). The objective is to differentiate corrosion levels in carbon steel pipelines through multiscale decomposition. Each signal is transformed into 62 components using a SWT tree of depth 5, and basic statistical features (mean and standard deviation) are computed from each node. To evaluate the discriminative capacity of these representations, dimensionality reduction techniques (PCA, UMAP, and t-SNE) and quantitative clustering metrics were applied. Results show that nonlinear approaches, particularly t-SNE and UMAP, are more effective in revealing the separation between low and moderate corrosion signals, highlighting the potential of the proposed method for structural health monitoring applications.

Palavras-chave: Wavelet transforms, Dimensionality reduction, Wavelet domain, Corrosion, Trees (botanical), Binary trees, Acoustic emission, Feature extraction, Monitoring, Standards, Stationary Wavelet Transform, Acoustic Emission, Corrosion Detection, Structural Health Monitoring, Dimensionality Reduction