Deep convolutional neural network for multi-level non-invasive tunnel lining assessment

Abstract

In recent years, great attention has focused on the development of automated procedures for infrastructures control. Many efforts have aimed at greater speed and reliability compared to traditional methods of assessing structural conditions. The paper proposes a multi-level strategy, designed and implemented on the basis of periodic structural monitoring oriented to a cost- and time-efficient tunnel control plan. Such strategy leverages the high capacity of convolutional neural networks to identify and classify potential critical situations. In a supervised learning framework, Ground Penetrating Radar (GPR) profiles and the revealed structural phenomena have been used as input and output to train and test such networks. Image-based analysis and integrative investigations involving video-endoscopy, core drilling, jacking and pull-out testing have been exploited to define the structural conditions linked to GPR profiles and to create the database. The degree of detail and accuracy achieved in identifying a structural condition is high. As a result, this strategy appears of value to infrastructure managers who need to reduce the amount and invasiveness of testing, and thus also to reduce the time and costs associated with inspections made by highly specialized technicians.

Ground Penetrating Radar Fourier Pre-processing for Deep Learning Tunnel Defects’ Automated Classification

“…The results obtained by training the network with the GPR profiles, extracted during the different investigation campaigns, were compared with the ones achieved by training the same network with the FTs of the same profiles. Whereas the accuracy obtained following the first approach showed values greater than 90% and with an average of 94.5% [31], the ones obtained with the trained ResNet-50 on FTs show values greater than 80%, unless level 2b, and with an average of 86.8%. Table 3 reports the confusion matrices for each level obtained with the FT.…”

“…The outcomes obtained with this last one showed a decrease in accuracy for all levels with respect to the results obtained by training ResNet-50 with GPR profiles, keeping an accuracy higher than 90% for the levels 3,5, and 6. [31]. Comparing the results of the two different training and test valuation of the ResNet-50 proposed, level 1 (healthy and reinforced/damaged) shows an accuracy of 88% compared to 92.6% previously obtained, level 2a (healthy/reinforced) shows an accuracy of 83.1% compared to 97.3%.…”

“…The pooling one improve the performance of the algorithm and reduces the cost by compressing the convolutional features. The last one is the layer that interprets the extracted features and creates a vector containing the probability of the analysed image to belong to each class [31]. In the literature, deeper neural networks are to perform better than shallower ones especially in the training phase, as shown in several studies [32,33].…”

“…The authors in [31] have already performed a study on the multilevel tunnel linings defects classification working with raw GPR data profiles. On the other hand, the strategy proposed in this work is related to the possibility of using as input data of a CNN, not only GPR profile images, but also the FTs of the same ones.…”

A deep learning framework based on improved self‐supervised learning for ground‐penetrating radar tunnel lining inspection

“…Accurate and precise interpretations primarily rely on the experiences of an interpreter. However, once the data volume is huge, conventional manual interpretations seem impractical due to great time consumption and high labor costs (Chiaia et al., 2022).…”