Aug 2026· e-Journal of Nondestructive Testing· Vol 31· 0 citations
TL;DR
This project aims to develop an AI model that processes images to classify whether a defect is structural or non-structural, identify the type of defect, assign a severity score, suggest causes and remedies using an LLM, and estimate the remaining service life.
Abstract
Civil structures such as buildings, roads, railways, bridges, tunnels, and dams are present in every society, and the safest and most durable ones are those that are properly managed and maintained. Health monitoring plays an important role in ensuring safety, but engineers still depend on traditional inspection tools like crack gauges or comparator cards, which are slow, labour-intensive, and differ based on the inspector's judgment. There is also no proper tool that systematically analyses this data or suggests solutions based on standard codes using AI. Considering these limitations, this project aims to develop an AI model that processes images to classify whether a defect is structural or non-structural, identify the type of defect, assign a severity score where 0 is worst and 100 is best, suggest causes and remedies using an LLM, and estimate the remaining service life. Literature supports the need for such a system, as studies have focused on deep-learning-based detection of concrete damage, CNN-based comparison between traditional and AI inspection, and 3D scanning with machine learning for defect detection. Surveys on NDT tools like UPV and X-ray show improved efficiency over manual methods but still remain time-consuming and less precise.
Most research concentrates on a single defect such as cracks, and although some models can detect or localize them, there is still a major gap in systems that quantify severity or provide causes and repair suggestions. Many approaches stop at detection without offering guidance for maintenance, making it difficult to translate inspection data into decisions. To address this, datasets in this project are collected from public sources and field surveys, followed by preprocessing such as histogram equalization and median filtering. Images are then manually labelled into structural and non-structural categories and further classified by defect type. An image classification model is trained on this dataset to automatically detect and classify defects. Severity scoring is done using the formula Intensity × Extent, where intensity depends on measurable parameters like crack width or spall area, and extent depends on defect length or percentage affected. The model uses an LLM aligned with Indian Standard codes to suggest possible causes and remedies and predicts remaining service life based on severity. The novelty of this work lies in combining multiple functions that previous studies treat separately: multi-defect classification, quantitative severity scoring, LLM-based engineering interpretation, and service-life estimation. No existing system integrates all these components into a single workflow dedicated to structural inspection.
Buildings are long-term investments, and regular monitoring helps extend their lifespan and prevent failures. This project aligns with SDG 9, SDG 11, and SDG 12 by supporting safer, more resource-efficient, and sustainable infrastructure management. By bringing together defect detection, severity assessment, interpretation, and life prediction, the work addresses limitations of manual inspections and gaps in current automated tools. It supports better maintenance planning and contributes to a more comprehensive AI-based structural health-monitoring framework.
Structural Health Monitoring (SHM) has become increasingly important in civil engineering due to the aging of infrastructure such as bridges, buildings, tunnels, and dams. Traditional inspection methods rely on manual visual assessments, which are time-consuming, labor-intensive, and often unable to detect early-stage damage. Recent advances in Artificial Intelligence (AI) have enabled the development of intelligent damage detection systems that improve the accuracy and efficiency of structural assessments. AI techniques, including Machine Learning, Deep Learning, Computer Vision, and Pattern Recognition, analyze sensor data, vibration signals, and images to identify defects and predict structural failures. This study presents a comprehensive survey of AI-based damage detection methods for civil infrastructure. The proposed framework integrates sensor data acquisition, feature extraction, machine learning classification, and automated damage assessment. Various AI algorithms such as Artificial Neural Networks (ANN), Support Vector Machines (SVM), Decision Trees (DT), and Convolutional Neural Networks (CNN) are evaluated for detecting structural deterioration. Experimental results demonstrate that AI-based approaches, particularly deep learning models, achieve higher damage detection accuracy than conventional inspection techniques. Furthermore, AI-powered SHM systems enable real-time monitoring, reduce maintenance costs, and support proactive infrastructure management, highlighting their potential to transform modern civil infrastructure maintenance and safety.
Rahul Mehta· International Journal of Mod...· 0 citations
This review has organized improvements in SHM along the lines of vibration-based anomaly detection, vision-based defect recognition, and multi-modal data fusion along the lines of vibration-based anomaly detection, vision-based defect recognition, and multi-modal data fusion.
Bellal Mia, Md Umar Faruk, M. Hasan et al.· Scientia. Technology, Scienc...· 0 citations
Structural health monitoring is needed to gauge the safety and sustainability of civil infrastructure. Conventional crack detection methods are through manual inspection which is time consuming, labor intensive and can easily be compromised through human error. The article offers a powerful method of identifying cracks in concrete buildings through transfer learning with deep convolutional neural networks. Models like ResNet50 and MobileNetV2, which are pretrained, are trained on a dataset of real crack images of concrete to be able to classify between cracked and non-cracked surfaces. On model generalization and performance, data augmentation and preprocessing methods are implemented. The experimental findings indicate that the suggested approach is highly accurate, precise and recalls, despite having a small training set. It is an effective system that may be used in real-time structural health monitoring applications.
Thushar S. Shetty, G. P. Dharshini, K. Kowsalyadevi et al.· International Conference on...· 0 citations
Cracks indicate the deterioration of civil engineering structures, and early detection through regular inspection is crucial for structural safety. However, traditional manual inspection is time- and labor-intensive and depends heavily on the inspector’s expertise. Although deep learning-based crack detection has been actively studied, most prior studies focus on a single environment such as asphalt or concrete, and research integrating asphalt, concrete, and tunnel environments remains scarce. In this study, we propose a Dense U-Net++-based model trained on a unified dataset of drone-captured crack images across all three environments. The model combines dense connections with U-Net++’s nested skip pathways to mitigate the semantic gap in the encoder–decoder structure, enabling effective feature fusion and precise restoration of fine crack boundaries. Focal loss addresses the severe class imbalance between background and crack pixels, and area-based postprocessing suppresses spurious detections. The proposed method achieved a precision of 95.47%, a recall of 92.35%, an F1-score of 93.91%, and an IoU of 88.58%, outperforming both the baseline Dense U-Net++ and Mask R-CNN. Qualitative evaluation confirmed reliable detection across rough asphalt textures, striped concrete patterns, and low illumination in tunnels. These results indicate that the proposed framework achieves promising crack segmentation performance for automated structural inspection.
Dong-bum Kim, Jin-Chul Heo, Sitara Afzal et al.· Algorithms· 0 citations
As service life increases and concrete structures are exposed to variable loads and harsh environments, surface cracks inevitably develop. Cracks affect not only surface appearance but also the long-term durability of concrete components; therefore, accurate identification and real-time monitoring are essential throughout the engineering life cycle. To address the low efficiency of conventional manual inspection, the limited recognition accuracy for microcracks, and the inadequate feature-extraction and interference-suppression capabilities of the original YOLOv11 model in complex field environments, this study proposes an improved YOLOv11 model for concrete crack detection. First, the original backbone is replaced with FasterNet to reduce redundant computation and strengthen the extraction of shallow and deep crack features. Second, VoVGSCSP, VoVGSCSPC, and GSConv modules are used to reconstruct the neck network, replacing the default C3k2 and standard convolution blocks to improve multi-scale feature fusion for cracks of different widths. Third, TripletAttention is embedded in the detection pipeline to capture three-dimensional feature correlations, suppress complex background noise, and emphasise discriminative crack information. Experiments were conducted on a self-built dataset of 7,353 crack images, randomly divided into training, validation, and test subsets in an 8:1:1 ratio. Ablation and comparative experiments show that the improved model achieved higher precision and mAP@0.5 than the original YOLOv11. The method supports fine-grained detection of concrete microcracks under diverse and complex construction conditions and provides a feasible approach for automated structural-damage inspection.
Zhihui Bao· Journal of Engineering Resea...· 0 citations
The natural and accidental deterioration of civil structures, combined with the increasing complexity of engineering projects, demands continuous monitoring strategies to ensure structural safety and durability. In this context, Structural Health Monitoring (SHM) techniques based on variations in modal properties have been widely investigated due to their capability for global assessment. However, practical limitations—such as information loss during modal identification and the low sensitivity of certain modal parameters to damage—have motivated the development of alternative approaches. Among these approaches, methodologies based on statistical features extracted directly from acceleration time series have gained prominence, particularly when combined with Machine Learning (ML) classifiers. Although several studies have successfully employed statistical indicators for damage detection, the relative relevance of these indicators and the potential benefits of dimensionality reduction can still be explored through a systematic evaluation, particularly under Sim-to-Real conditions. This study investigates the performance of ten statistical indicators as input features for ML-based damage detection, applied to an experimentally tested frame structure, using numerically obtained data for training. A structured variable analysis was conducted to assess the individual and combined impact of the indicators, supported by visualization techniques and classifier performance metrics. The results showed some possible redundancies between certain indicators and demonstrated that a reduced set of attributes can achieve comparable or superior classification performance, while simultaneously improving training stability and computational efficiency. These findings highlight the importance of systematic feature selection and the effectiveness of reduced-dimensional statistical representations for SHM applications in Sim-to-Real scenarios.
Victor Carvalho, M. Marcy, G. Doz· Journal of Civil Structural...· 0 citations