Damage Classification Research Articles

Categorization of Post-Earthquake Damages in RC Structural Elements with Deep Learning Approach

ABSTRACT The aim of this study was to develop an innovative deep learning based intelligent software (DamageNet) and its mobile applications to classify seismic damage of Reinforced Concrete (RC) elements. Images of 2455 damaged elements that have been exposed to different destructive earthquakes were collected from the “datacenterhub” database. The DamageNet algorithm has been compared with the pretrained convolutional neural networks (CNN) algorithms (VGG16, ResNet-50, MobileNetV2 and EfficientNet) according to performance metrics. With the other models, a maximum test success of 89% was achieved, while with DamageNet a test success of 92% was achieved in damage classification. The mobile application developed based on the DamageNet model was tested in the field after the earthquakes (Mw:7.7 and Mw:7.6) in Kahramanmaraş/Turkey and classification success of 88% was obtained.

Machine learning and soft voting ensemble classification for earthquake induced damage to bridges

Seismic vulnerability assessment of bridges is essential for both pre-earthquake planning and post-earthquake rapid loss assessment to facilitate maintenance planning and recovery initiatives. Given the impracticality of creating numerical models for every conceivable attribute variation within bridge classes, data-driven approaches offer a versatile solution to address such variations. While data-driven techniques are widely adopted in bridge engineering, many existing models rely on numerical or experimental datasets. However, such models may not adequately capture the dynamic and complex nature of seismic response of bridges. This presents an opportunity to evaluate the effectiveness of data-driven approaches using post-earthquake damage data. This paper assesses the efficacy of machine learning models and develops a robust framework for the classification of earthquake-induced damage incurred to some bridge classes. We deployed damage data collected following the 2015 Gorkha earthquake, incorporating taxonomical, stiffness, and excitation variables as predictor variables. Considering three distinct damage classes—minor, major, and critical—in a heterogeneous bridge class environment, we evaluated multiple machine learning classifiers. To account for the influence of heterogeneous bridge classes in the dataset, we conducted separate assessments for the reinforced concrete (RCC) bridge class. The Random Forest classifier achieved a substantial accuracy improvement, reaching 83% when compared to the mixed bridge class accuracy of 79% reflected by the ANN classifier. Similarly, a minor improvement was observed when deploying soft voting classifiers for RCC only bridge data. The sum of observations suggest that tree-based classifiers consistently outperform other classifiers when considering single bridge classes. This study identifies stiffness variables as the most critical factor in bridge damage classification.

A domain adaptation approach to damage classification with an application to bridge monitoring

Data-driven machine-learning algorithms generally suffer from a lack of labelled health-state data, mainly those referring to damage conditions. To address such an issue, population-based structural health monitoring seeks to enrich the original dataset by transferring knowledge from a population of monitored structures. Within this context, this paper presents a transfer learning approach, based on domain adaptation, to leverage information from completely-labelled bridge structure data to accurately predict new instances of an unknown target domain. Since intrinsic structural differences may cause distribution shifts, domain adaptation attempts to minimise the distance between the domains and to learn a mapping within a shared feature space. Specifically, the methodology involves the long-term acquisition of natural frequencies from several structural scenarios. Such damage-sensitive features are then aligned via domain adaptation so that a machine-learning algorithm can effectively utilise the labelled source domain data and generalise well to the unlabelled target-domain data. The described procedure is applied to two case studies, including the Z24 and the S101 benchmark bridges and their finite element models, respectively. The results demonstrate the successful exchange of health-state labels to identify the damage class within a population of bridges equipped with SHM systems, showing potential to reduce computational efforts and to deal with scarce or poor data sets in application to bridge network monitoring.

Damage Classification Using CNN-Based Model for Multi-Part Strengthening System

Strengthening and retrofitting systems, including FRCM (Fiber-Reinforced Cementitious Matrix), play a vital role in improving the durability and safety of structures facing challenges such as aging, increased loads, or seismic risks. To ensure the long-term integrity and performance of these structures, it’s crucial to monitor the health of repaired structural components. Acoustic emission (AE) is a non-destructive and passive technique that involves the detection and analysis of stress waves emitted by various material components undergoing damage. In this particular research, the study focuses on the classification of damage in different material components of RC (Reinforced Concrete) beams strengthened with FRCM. This classification is achieved using a convolutional neural network (CNN) model that utilizes image-based waveform data from the AE testing. The study collected waveforms from four distinct failure modes: fabric-matrix debonding, fabric rupture, tensile cracking in the cementitious matrix, and yielding of steel. These waveforms were used to train the CNN model. Each waveform was converted into a Discrete Wavelet Transform (DWT) scalogram as an input into the model. The model demonstrated prediction accuracy of approximately 95% during training. The pre-trained model was further employed to classify failure mechanisms with a separate test dataset with an accuracy of approximately 87%, which is considered satisfactory considering added noise to the test dataset.

Structural Damage Classification in Offshore Structures Under Environmental Variations and Measured Noises Using Linear Discrimination Analysis

Changing environmental conditions and measured noises often affect the dynamic responses of structures and can obscure subtle changes in the vibration characteristics caused by damage. To address this issue, a new method for classifying damage in offshore structures under varying environmental conditions and measured noises is proposed using linear discrimination analysis (LDA). Two sets of data on dynamic characteristics, one from healthy structures and the other from unknown testing structures, are used to determine the optimal projection vector. This vector is perpendicular to the discriminant hyperplane and is used for damage classification. The damage‐sensitive features are extracted by projecting both sets of data onto this vector. These features are then used with the hypothesis test technique to determine the condition state of the testing structure. Numerical studies on offshore wind turbine structures and experimental validations of a deep‐sea mining system are being conducted to evaluate the effectiveness of the proposed approach. The study also examines the impact of mode combinations, measured noises and samples on the performance of the approach. The results indicate that the proposed approach can accurately assess the structural health state even in the presence of environmental changes and noise contamination, even with limited samples. The promising performance of the approach will facilitate the establishment of an online structural monitoring system to ensure the safety of offshore structures.

Damage Classification of a Bolted Connection using Guided Waves and Explainable Artificial Intelligence

With the improvements in computational power and advances in chip and sensor technology, the applications of machine learning (ML) technologies in structural health monitoring (SHM) are increasing rapidly. Compared with traditional methods, deep learning based SHM (Deep SHM) methods are more efficient and have a higher accuracy. However, due to the black box nature of deep learning, the trained models are usually difficult to interpret, which blocks their practical application. Therefore, it is of great importance to develop explainable artificial intelligence (XAI) methods to understand the internal decision-making mechanisms of damage classification in Deep SHM. In this paper, a novel XAI algorithm named Deep Gradient-weighted Class Activation Mapping (Deep Grad CAM) is proposed by combining the existing method Grad CAM with the convolutional neural network (CNN) deconvolution mechanism. In this paper, Deep Grad CAM is used to interpret a one-dimensional convolutional neural network trained to detect bolt loosening based on guided wave propagation. The interpretation performance of Deep Grad CAM is compared with Grad CAM, and their performances are quantified using Infidelity. The results show that the Infidelity of Deep Grad CAM is much smaller than that of Grad CAM, indicating significant improvements in explanation accuracy and reliability.Keywords

A Novel Approach to Discriminate Between Structural and Non‐Structural Post‐Earthquake Damage in RC Structures

In post‐earthquake damage assessment studies of reinforced concrete (RC) buildings, the most important consideration is to determine the damage and severity of structural elements is the determination of the damage and severity of structural elements. For earthquake damage to structural elements, such as columns and shear walls, decisions can be made about strengthening or demolition. However, due to the need for rapid assessment of damaged buildings after earthquakes, these two damage mechanisms can sometimes be confused. Non‐structural damage is classified as structural, leading to uneconomic structural decisions such as demolition or strengthening, or conversely, life safety issues arise from misidentifying critical structural damage with non‐structural damage. In this study, an artificial intelligence‐based damage assessment algorithm has been developed to accurately and quickly differentiate between structural and non‐structural damage. For the damage classification model, a deep learning algorithm was developed using the 9680 damage images obtained from field studies after the recent earthquakes of Mw ≥ 5; Istanbul‐Silivri (Mw: 5.8), Elazığ‐Sivrice (Mw: 6.8) and Izmir‐Seferihisar (Mw: 6.6) in Turkey. With an accuracy rate between 93% and 96%, the models constructed by selecting the optimal values correctly detected and categorised structural and non‐structural element damages in RC structures. The method developed in this study can help experts in damage assessment studies as a decision support mechanism.

A pre-failure narrow concrete cracks dataset for engineering structures damage classification and segmentation

Monitoring of structures’ condition plays a fundamental role in providing safety for users and extending the structures’ lifespan. The monitoring is conducted through on-site inspections by engineers thus this process is time-consuming, labor-intensive and prone to subjective engineering opinions. Detecting damage using machine learning algorithms on images can support engineers’ work, especially for early damages which are difficult to see with the human eye. This article is focused on the concrete crack detection problem in engineering structural elements. Despite the availability of several concrete crack detection datasets, no dataset allows semantic segmentation of cracks narrower than 0.3 mm (the crack width limit for typical engineering structures elements and environmental conditions according to EC 1992-1-1) and the ability for crack classification is limited. The provided open dataset represents only cracks below the crack width limit of 0.3mm, which do not yet indicate concrete elements failure. It is dedicated for early crack classification and segmentation, so that damage protection can be taken at an early stage to prevent structural element damages.

Deep Learning Model for Classifying and Evaluating Soybean Leaf Disease Damage.

Soybean (Glycine max (L.) Merr.) is a major source of oil and protein for human food and animal feed; however, soybean crops face diverse factors causing damage, including pathogen infections, environmental shifts, poor fertilization, and incorrect pesticide use, leading to reduced yields. Identifying the level of leaf damage aids yield projections, pesticide, and fertilizer decisions. Deep learning models (DLMs) and neural networks mastering tasks from abundant data have been used for binary healthy/unhealthy leaf classification. However, no DLM predicts and categorizes soybean leaf damage severity (five levels) for tailored pesticide use and yield forecasts. This paper introduces a novel DLM for accurate damage prediction and classification, trained on 2930 near-field soybean leaf images. The model quantifies damage severity, distinguishing healthy/unhealthy leaves and offering a comprehensive solution. Performance metrics include accuracy, precision, recall, and F1-score. This research presents a robust DLM for soybean damage assessment, supporting informed agricultural decisions based on specific damage levels and enhancing crop management and productivity.

Rapid mapping of volcanic eruption building damage: A model based on prior knowledge and few-shot fine-tuning

Large-scale volcanic eruptions always inflict severe damage upon facilities and cause significant environmental pollution. Building damage caused by lava flow and volcanic ash coverage can reflect the infrastructure devastation and victim scope within the affected region. The application of machine learning methods for building damage automated identification from remote sensing imagery typically relies on a large number of training samples. However, labeled data scarcity is a common issue in the field of disasters, particularly in volcanic eruptions. To address this, we propose a two-stage building damage quick mapping workflow, which combines a building localization model trained on prior knowledge and a damage classification model fine-tuned on few-shot volcanic eruption-related samples. The classification model utilizes a CNN-based Siamese network for bi-temporal image feature extraction and comparison, with the backbone initialized with pre-trained weights from ImageNet. We conducted building damage classification tasks for single-disaster scenarios and cross-disaster domain scenarios in the eruptions of Mount Semeru, Tonga, and ST. Vincent; the visual damage level of each building was used as ground truth. The results demonstrate that our model can identify building damage efficiently and accurately in different volcanic eruption scenarios, with an over 93% F1-score on the 2-way 20-shot tasks. Furthermore, though building samples from different volcanic eruption regions present cross-domain challenges, our model can adapt to different feature domains by being supplemented with a few samples of another volcanic eruption disaster. Additionally, in the case of Mount Semeru Eruption, we gain insights into the potential of building damage statistics in post-eruption environmental assessments. To further enhance the model robustness on mixed-domain samples and multi-level damage classification tasks, issues including sample bias of certain disaster sources should be addressed.

Wind turbine blade damage detection and classification based on sound feature signal using machine learning

Wind turbine blades experience variable environmental conditions and are prone to failure due to fatigue and weather-induced damage. In this study, the aerodynamic sound signal of wind turbine blades is collected while in operation. Short-Time Fourier Transform method identified and used to analyze the periodic signal of the rotation of the wind turbine blades. The spectral centroid of the pulse signal and the variation of the corresponding three blades spectral centroid are extracted as the input features for supervised machine learning. Gaussian mixture model algorithm classified features and used with blades damage trend for structural health monitoring. Through the Gaussian mixture model established by using the mean value and standard deviation of the peak value of the spectral centroid, the results after the model test show the accuracy of the damaged blade judgment can reach 98 %. In addition, from the visual classification chart, it can be known that the blade will enter the damage state, and maintenance should be carried out in advance to avoid damage to the blade and maintain the stable operation of the wind turbine.

Road Damage Detection and Classification

Abstract: This is a road damage detection desktop application. In the subject of transportation engineering, identifying road damage early on is essential because it can save maintenance costs and avoid accidents. Deep learning methods have demonstrated encouraging performance in a number of computer vision tasks recently, including the detection of road damage. In this study, we suggest a region-based convolutional neural network (R-CNN)-based method for detecting road degradation. Using a publicly accessible dataset of road photos with different kinds of damage, such as cracks, potholes, and patches, we trained our R-CNN. Our approach outperformed cutting-edge techniques with an accuracy of 85% in identifying road damage.

Impact dynamic response of large aperture space deployable antenna supporting structures based on a dual-scale model

This study investigates the joint damage patterns and changes in shape accuracy of large aperture space deployable antenna supporting structures subject to hypervelocity impacts from micro space debris. Based on the dual-scale method, a joint scale model is constructed for a hexagonal prismatic space deployable truss antenna supporting structure with an aperture of six meters and 37 modules. ANSYS/AUTODYN is used for the impact damage analysis, with an equivalent model of impact force integrating the structure's actual damage characteristics in the structure scale model. The dynamic response and deformation performance of the antenna supporting structure under impact velocities from 2.5 to 15.0 km/s, with spherical space debris of 2.5 mm and 5.0 mm diameters, impacting modular central points in both positive and negative directions, are analyzed. The results show that the primary damage to the joint scale model after impact includes cratering, perforation, punch failure, and joint failure. The debris diameter significantly influences joint damage and deformation of the bolt hole diameter. Additionally, the debris cloud post-impact significantly affects the vertical bar's inner wall, spalling the wall and reducing the rod's strength. As the impact point moves towards the structure's centroid, the antenna's overall response increases. With higher impact velocities, the antenna supporting structure's overall deformation first increases, then decreases. At an impact velocity of 12.5 km/s, the impact area of the structure gradually reduces post-impact. Our proposed joint damage forms and three-case damage classification (no effect, repairable, and joint failure) may serve as a valuable reference for designing structural protection and impact identification for space deployable antennas.

Safety Evaluation of a Prototypical Diazirine-Based Covalent Crosslinker and Molecular Adhesive.

bis-Diazirine reagents are increasingly being used as polymer crosslinkers, adhesives, and photopatterning agents in the materials sciences literature, but little effort has been made thus far to document their chemical safety profile. Here, we describe the results of a detailed toxicity assessment of a representative bis-diazirine. Safety was evaluated by a series of in vitro assays, which found the product to be non-mutagenic in bacterial tester strains TA98 and TA100, non-corrosive and non-irritating to skin, and requiring no classification for eye irritation or serious damage. While in vitro tests do not capture the integrated whole animal system, and thus cannot completely rule out the possibility of adverse responses, the results of this study suggest a desirable safety profile for bis-diazirine reagents and provide a solid foundation upon which to add in vivo assessment of safety risk and dose-response studies.

Emergency Shutdown Valve damage classification by machine learning using synthetic data

Emergency Shutdown Valves (EDSVs) are used in industrial applications to interrupt internal fluid flow in pipelines during hazardous events. During their operation, these valves can suffer cumulative damage on their reinforced polytetrafluoroethylene (RPTFE) seats that can put their operation and effectiveness at risk. One of the options for detecting the occurrence of damage is to analyse data acquired from fluid pressure and torque sensors during the closing and opening cycles of the valve. The resulting operational signatures are commonly evaluated through so-called Transition Points (TPs) that are manually marked in a subjective manner by a trained operator. In addition to being slow and laborious, this approach discards most of the acquired information. Alternatives to this method would be the use of Damage Indexes (DIs) that are extracted from the signatures, or even the evaluation of the complete pressure or torque signature. These methods, when associated with machine learning (ML) algorithms, could use the acquired information more efficiently and reliably, and would have the potential to completely automate the monitoring process. In this work, these three processing options were tested and compared using real and synthetic data that were generated with the Monte Carlo (MC) method. The results indicate that the evaluation of the complete signature with a Gradient Boosting Classifier (GBC) algorithm would be the most effective strategy, which correctly detects the damage in 99% of the cases considered.

Convolutional neural network–based structural health monitoring framework for wind turbine blade

Wind energy is a renewable energy source with a significant impact on the date for harnessing it. Different types of wind turbines (WTs) are being built to generate affordable, dependable, and eco-friendly renewable wind energy. Over time, cracks are found to be developed in the blades of the WTs, which is considered the most common type of damage that ultimately causes the catastrophic failure of the structures. This study aims to develop a structural health monitoring (SHM) framework for localizing cracks of a WT blade using a convolutional neural network (CNN)–based deep learning algorithm. With that objective, the NACA (National Advisory Committee for Aeronautics) 63-412 profile (WT blade) of length 29 m was modelled in finite element analysis (FEA) ANSYS 2022 R2. A crack is introduced to the model by making a groove at different blade locations. The mode shape and natural frequencies are obtained and validated with those reported in the literature. Further, the blade is excited by an impact load applied at the tip of the blade, and acceleration time histories are collected. The acceleration time history data are converted to scalogram images and fed into the CNN algorithm for damage classification and localization. In this study, the concept of class activation maps is also utilized for visual representations of the input images’ areas that significantly influence a class’s classification score. The classified results show that using CNN can successfully localize the region of the cracks and motivate us to examine it on a laboratory-based model.

Preliminary Results from iBeat Ancillary Study: A Novel Bari Classification of Renal Damage in Diabetic Patients with CKD

Preliminary Results from iBeat Ancillary Study: A Novel Bari Classification of Renal Damage in Diabetic Patients with CKD

Problems of power transmission accident rate (on example of JSC "Omskenergosbyt" )

A study and analysis of electric power transmission through the JSC Omskenergosbyt electric networks over the period 2014 – 2022 have been carried out. The structural and balance features of the company are analyzed, the characteristics of the key elements of the studied electrical networks are given, with a number of shortcomings in the disclosure of information noted. Based on public information on the accident rate in the company's electrical networks, an estimate has been made of the number of emergency situations that occurred during the period under study, as well as the magnitude of electrical power under-supply caused by these interruptions. The classification of emergency damages according to the amount of electrical power under-supply as a result of these failures has been made. The percentages of such failures are determined. Various methods for predictive assessment of power supply reliability level in the company's networks are analyzed, and a conclusion is made about the applicability of predictive models of statistical methods. Autoregressive equations have been obtained, the analysis of which allowed to have numerical predictive values of the number of possible emergency failures broken down by months of 2023. The influence of the seasonal component on the number of emergencies during the period under study has beenis considered.General scientific methods, numerical methods of analysis, statistical and cybernetic methods of making predictive models were used in the study. Excel spreadsheet technologies, MATLAB software environment, as well as the Oracle software were used as computational tools. The results obtained may be of interest to chief executives of power grid companies , as well as researchers and engineers engaged in research in the field of power supply reliability.

Acoustic emission study on damage mechanism of composite materials repaired with bi-adhesive and damage classification model

The load-bearing capacity of damaged composite materials can be improved using the double adhesive repair technique. This study uses acoustic emission (AE) as the primary detection method to examine the damage evolution of modified double-bonded repair specimens and discover the most effective configuration for bi-adhesive repair. The introduction of machine learning and signal feature extraction in the analysis of huge acoustic emission parameters has significantly improved the accuracy and efficiency of damage classification. Following the preliminary division of the clustering results of the unsupervised algorithm, the corresponding relationship between various clustering types and damage types is confirmed using the energy characteristics of wavelet packet decomposition. Parameter and waveform analysis are extensively utilized in studying the AE damage of composite materials and have complementary advantages. Thus, a supervised classifier is established by combining characteristic parameters with time-frequency domain analysis of the signal waveform. Furthermore, machine learning is used to process the large signal set and analyze the debonding damage behavior of the patch. In the end, it is shown that the specimens with the modified adhesive in the middle of the repair area displayed superior repair results. Hence, when combined with machine learning, the signal classification of the damage evolution behavior of repaired specimens is realized with high accuracy.

Unveiling out-of-distribution data for reliable structural damage assessment in earthquake emergency situations

To ensure accurate and effective emergency responses following an earthquake, one must promptly and accurately assess damage to structures with minimal manual effort. An advanced approach for evaluating structural damage is to use machine learning techniques for automated tagging or damage classification. However, most previous studies evaluated models using simple and unrealistic datasets limited to in-distribution (ID) datasets for training and testing. Although this approach may yield satisfactory results within the confines of the training dataset, it is not justifiable for real-time scenarios, in which the testing dataset may differ. Hence, a novel methodology is proposed herein that focuses on the detection of out-of-distribution (OOD) data. By subjecting a network to outliers, the model can effectively identify data outside the domain of the training dataset. A custom loss function is adopted, where both the cross-entropy loss from the ID training dataset and the log loss from the outlier dataset are incorporated. The effectiveness of this approach is demonstrated through its application in the post-event rapid damage assessment of bridges and shear walls subjected to seismic loading. A single network is employed to classify ID data into their respective classes, whereas the OOD data are shown to belong to the OOD class. The results highlight the significant accuracy achieved in the simultaneous prediction of ID and OOD data. By incorporating the detection of OOD data, this study enriches the methods for improving the reliability and accuracy of structural damage assessments in earthquake emergency responses, thus enabling more informed decision-making.