Comparison between Supervised and Unsupervised Learning for Autonomous Delamination Detection Using Impact Echo

Abstract

Impact echo (IE) is a non-destructive evaluation method commonly used to detect subsurface delamination in reinforced concrete bridge decks. Existing analysis methods are based on frequency domain which can lead to inaccurate assessments of reinforced concrete bridge decks since they do not consider features of the IE signals in the time domain. The authors propose a new method for IE classification by combining features in the time and the frequency domains. The features used in this study included normalized peak values, energy, power, time of peaks, and signal lengths that were extracted from IE signals after they are preprocessed. We used a dataset containing IE data collected from four in-service bridges, annotated using chain dragging. A support vector machine (SVM) classifier was constructed using combined features to classify IE signals. A 1DCNN with unfiltered IE signals and a two-dimensional CNN using wavelet scalograms (2D representations of unfiltered IE signals) were also used to classify IE signals. The SVM model performed significantly better than the other models, with an accuracy rate, true positive rate, and true negative rate of 97%, 92%, and 98%, respectively. The SVM model also generated more accurate defect maps for all investigated bridges. IE data from the Federal Highway Administration’s InfoBridge website were used to investigate the efficacy of the developed models. The investigation yielded promising results for the proposed SVM model when used for a new set of IE data.

Enhancing non-destructive testing in concrete structures: a GADF-CNN approach for defect detection

“…The methods evaluated encompass the Naive Bayes classifier, Support Vector Machines (SVM), standard Convolutional Neural Networks, 1D Convolutional Neural Networks, and the GADF-CNN. [10] 0.92 0.91 0.92 0.92 SVM [13] 0.88 ---Convolutional neural networks [14] 0.94 ---Convolutional neural networks [16] 0.96 ---1D Convolutional neural networks [18] 0.91 0.92 0.91 0.91 GADF-CNN 0.98 0.97 0.97 0.98…”

“…However, its foundational assumption of feature independence may limit its applicability to more intricate datasets. The SVM application by Dorafshan et al [13] yields an accuracy of 0.88, but the absence of additional metrics precludes a comprehensive assessment. SVM's effectiveness is notably dependent on kernel selection and may be less suitable for large, multi-class datasets.…”

“…With the advent of machine learning, there has been a significant shift towards enhancing the interpretative capabilities of impact-echo data. Machine learning has substantially improved the accuracy and reliability of data interpretation from impact-echo tests [8][9][10][11], leading to advancements in automating NDT and defect recognition in concrete structures [12], [13]. The evolution from machine learning to deep learning, particularly Convolutional Neural Networks (CNNs), has further expanded the possibilities in defect detection and classification [14][15][16][17][18][19].…”