Natural hazards cause severe damage to the built environment and community in general. In recent years, Machine learning (ML) has become a powerful tool in natural hazards engineering, offering accurate solutions for predicting, assessing, and managing natural hazard risks. Transfer learning (TL), a sub-domain of ML, has gained attention for its ability to transfer knowledge from relevant tasks to improve the model performance with limited data. There is an increasing interest in utilizing TL techniques for mitigating natural hazard risks. This study is designed to synthesize the literature at the intersection of transfer learning and natural hazards and identify the gaps in the literature for future direction. Specifically, a scoping literature review methodology is conducted to understand how the transfer learning techniques are applied in natural hazards engineering, emphasizing the impact to the civil engineering domain. From 904 records found on the Web of Science database, 122 studies were included based on inclusion criteria. The findings categorize the reviewed studies by natural hazard type, algorithm, data type, and civil engineering discipline, revealing key trends such as the dominance of parameter-based TL algorithms and the extensive use of image data. This review highlights the significance of TL in enhancing the resilience of infrastructure against natural hazards, providing a foundation for future research and development in this crucial area.

Conventional data-driven methods for predicting the seismic response of structures often require extensive data and computational resources. To address these challenges, a novel deep learning framework that can efficiently and accurately predict the structural seismic responses is proposed. The proposed framework overcomes the limitations of the conventional data-driven methods, by utilizing transfer learning based on the most relevant knowledge determined via the unsupervised learning technique. The framework leverages the seismic information history database to identify the most similar previous earthquake, and subsequently transfers the corresponding knowledge from the Structural Seismic Response network (SSR net) to predict structural responses caused by a new earthquake. This innovative method significantly reduces the need for extensive data collection and provides efficient predictions. Case studies demonstrate the framework’s ability to predict seismic structural responses without extensive training or data collection. The framework can reliably capture the complex nonlinear dynamics of structures under seismic loads and offer significant potential for advancing seismic fragility analyses and reliability assessments. Future research will focus on expanding the framework’s applicability to various structural types and further refining its prediction capabilities.

This study uses transfer learning to propose a model for predicting power outages caused by various extreme weather events that are being exacerbated due to climate change. Transfer learning is a technique that allows a model trained on a significant source domain to be fine-tuned for a target domain with limited data, enabling practical regression or classification even in scarce data scenarios. In Texas, the model was developed using data-rich events like ‘Heat’ and ‘Thunderstorm Wind’ as the source domain to predict relatively data-scarce events such as ‘Excessive Heat’, ‘Winter Storm’, and ‘Flood’ as the target domain. The results indicated significant effectiveness of transfer learning, particularly in transitions such as from ‘Heat’ to ‘Excessive Heat’, ‘Heat’ to ‘Winter Storm’, and ‘Thunderstorm Wind’ to ‘Flood’. This suggests that identifying efficient combinations of source and target weather events is crucial. Moreover, this methodology is anticipated to be applicable in Texas and other states facing similar challenges.

This study proposes a novel framework for the prediction of structural damages caused by extreme weather and climate events. In current practice, following a weather event, inspectors manually evaluate damaged structures and assign a damage state classification according to FEMA guidelines. The application of machine learning methods to postevent damage classification has received significant attention in the past decade. Current state-of-the-art applications in automating the assigning of damage states have focused on postevent unmanned aerial system (UAS)-driven image classification. These works have achieved moderate success using damage classes with varying similarities to established FEMA guidelines. This work proposes a framework for predicting FEMA damage states at a single structure level prior to an event. Using a precurated data set of structural characteristics and predicted best track storm data, a novel approach can be used to optimize postevent response efforts. The methodology was validated using a data set of structural features and best track storm data gathered following Hurricanes Harvey, Michael, Irma, and Dorian. The trained model achieved a 48.08% single damage state classification accuracy and an 84.24%+/- 1 class damage state classification accuracy. These results show that the proposed framework can perform pre-event damage prediction with performance on par with the current postevent damage classification methods. This study provides a framework for the prediction of hurricane-induced structural damage states prior to an event. Based on a lightweight artificial intelligence model, the framework is designed to be accessible to homeowners, municipalities, and relief organizations that do not have access to sophisticated hardware. The framework utilizes structural characteristics and storm information to generate structure-by-structure damage predictions based on FEMA hurricane damage states. A data set consisting of structures impacted by Hurricanes Harvey, Michael, Irma, and Dorian was used to validate the framework and provide an estimate of its capabilities. The trained model achieved an overall single state classification accuracy of 48.08% and a +/- 1 class accuracy of 84.24%. These results show that the proposed framework can provide homeowners with prehurricane predictions of the damage state their unique home is likely to suffer with the same level of performance as current state-of-the-art image-based postevent damage classification artificial intelligence models.

A non-contact, targetless approach to determine the deflection of bridges using consumer grade video cameras is presented. A total of four bridges (two concrete bridges and two steel bridges) were selected for load testing, based on typical characteristics of load posted bridges in Texas. Each bridge was instrumented using strain gauges, string potentiometers, and accelerometers to measure the response of the bridge during various load tests. In addition to these conventional measuring devices, two cameras mounted on a tripod were used to record the bridge response during each load test. An image analysis algorithm was applied to determine the displacements from the unloaded bridge image and loaded bridge image. These tests demonstrated that computer vision has the potential to measure deflections during bridge load testing without the need for targets. This method provides an efficient alternative for field evaluation that eliminates the need to instrument the bridge, which can be a time-consuming process, especially when access is restricted.

Direct integration methods are widely used for dynamic response computation. However, the performance of their computational accuracy significantly degrades with increasing the time step. Although machine learning methods can address this shortcoming, they require training data for dynamic response computation. This paper proposes a novel computational method to overcome these shortcomings. The proposed approach is a data‐free physics‐driven estimator, which minimizes the objective function of multi‐output least squares support vector machines for regression to model parameters subject to physical constraints introduced by the multi‐degree of freedom system's dynamic equilibrium equations and initial conditions in the feature space, bypassing the need for training data (due to the coupled physics) and for satisfying the requirement of the time step due to the built‐in optimization procedure. A new efficient step‐by‐step solver is developed to solve the optimization problem, and the solution is equivalent to a hyperplane satisfying the physical constraints in the feature space. The extension of the proposed approach for nonlinear dynamic response computation is also analyzed theoretically. The numerical results demonstrate that the proposed approach provides the solution with higher accuracy and efficiency and achieves the best performance for large time steps over classical integration methods.

The authors appreciate the discusser's comments on our manuscript. We believe that careful feedback enhances the impact of our paper (Pak et al, 2023) to researchers and practitioners in the structural engineering domain. The issues raised by the discusser pertain to the scope and significance, black boxes in machine learning, and engineering applicability. This document delivers our detailed responses to the discusser's comments. As clearly stated in the manuscript, the primary objective of this study is to propose a new learning algorithm integrating ensemble learning and transfer learning that can mitigate the data scarcity problem and reduce necessary expenses for gathering additional samples or training a new machine learning model. When it comes to the use of ML-based approaches, one of the necessary conditions is a large number of training samples. Otherwise, it is difficult to obtain good performance from a trained model. Furthermore, good prediction performance from a trained model can be obtained only if training and testing data have identical distributions. The TENN model proposed in this paper is designed to address the high prediction variances associated with small datasets, while maintaining robust model performance. The hypothesis of this work is that it is possible to transfer knowledge from slender to deep beams, hence, validating this type of approach for other components or materials where even less data is available. To investigate the performance of the TENN model with a relevant structural application example, the commonly used criterion, a / d < 2.5 $a/d < 2.5$ , has been used consistently to split the slender (source domain) and deep (target domain) beams. The performance of the TENN model has been compared with that of the three baseline models introduced in the paper. As demonstrated in the paper, the TENN model shows outstanding performance, especially when the available data is extremely limited. One of the popular issues in machine learning or deep learning is the discussion of their ‘black box’ nature. In contrast to the intrinsic explanations of simple models, it is usually difficult to interpret complex machine learning models and the hidden relationships between inputs and outputs. Currently, this study has demonstrated the capability of the TENN model to predict the shear capacity of reinforced concrete beams with insufficient training data. Investigating interpretable machine learning models is beyond the scope of this paper, and, of course, further investigation should be carried out. However, it is difficult to agree with the discusser's claim that artificial neural network (ANN) results are suitable only for revealing computational efficiency but are irrelevant to engineering practice. Accurate predictions of the shear capacity of RC deep beams will aid in furthering our understanding of the shear failure mechanism. Furthermore, it is not surprising that ANN algorithms are efficient in solving well-defined problems. Rather, the advantage of ANN algorithms is emphasized when a problem is too complicated to be well-defined. The authors strongly agree that an engineer is responsible for design specifications and code requirements as well as the performance and integrity of the structures that are constructed according to such guidelines. Although the highly accurate performance of the proposed model is presented in the manuscript, we have never indicated that the design principle or code equation needs to be reconsidered, or that the proposed approach should be used instead of these guidelines. The significant contributions of this work are two-fold: (1) providing a novel methodology; and (2) showing its feasibility in estimating the structural capacity of RC beams, especially when sufficient training samples are unavailable. After the validation procedure presented in Section 5.2, the proposed method was used to estimate the shear capacity of RC beams without transverse reinforcement as a more thorough case study of the performance ability of the proposed approach. To this end, we have compared the ability to accurately (high R2) and robustly (low CV value) estimate the shear capacity of RC beams without transverse reinforcement with very limited training samples. This paper provides novel research on the endeavor towards developing machine learning models to robustly predict the load-carrying capacity of various structural components. This paper also takes an extremely cautious stance to ensure a safe design. This perspective is also mentioned in the manuscript, ‘This is understandable as sufficient amounts of safety against service or structural limit states should be ensured.’ Furthermore, this paper has never denied the use or validity of the design standards. As always, safe design approaches should be the top priority for engineers when designing a structure. The latest design standard has been used as a reference to quantify the shear capacity of RC beams.

This research proposes a physics-informed few-shot learning model to predict the wind pressures on full-scale specimens based on scaled wind tunnel experiments. Existing machine learning approaches in the wind engi-neering domain are incapable of accurately extrapolating the prediction from scaled data to full-scale data. The model presented in this research, on the other hand, is capable of extrapolating prediction from large-scale or small-scale models to full-scale measurements. The proposed ML model combines a few-shot learning model with the existing physical knowledges in the design standards related to the zonal information. This physical infor-mation helps in clustering the few-shot learning model and improves prediction performance. Using the proposed techniques, the scaling issue observed in wind tunnel tests can be partially resolved. A low mean-squared error, mean absolute error, and a high coefficient of determination were observed when predicting the mean and standard deviation wind pressure coefficients of the full-scale dataset. In addition, the benefit of incorporating physical knowledge is verified by comparing the results with a baseline few-shot learning model. This method is the first of its type as it is the first time to extrapolate in wind performance prediction by combining prior physical knowledge with a few-shot learning model in the field of wind engineering. With the benefit of the few -shot learning model, only a low-resolution of the measuring tap configuration is required, and the reliance on physical wind tunnel experiments can be reduced. The physics-informed few-shot learning model is an efficient, robust, and accurate alternate solution to predicting wind pressures on full-scale structures based on various modeled scale experiments.

This chapter introduces a method that can achieve a suitable tradeoff between computational cost and predictive performance for probabilistic seismic analysis. The proposed method integrates Monte-Carlo simulation within the framework of artificial intelligence-enhanced mechanical model recently developed by Luo and Paal to consider the uncertainty associated with mechanical properties of structures, while the uncertainty related to seismic action is conveniently included by means of implementing incremental dynamic analysis. A reinforced concrete frame building is used as a testbed for validating the proposed method. In the probabilistic analysis, concrete compressive strength and longitudinal reinforcing steel yield strength are considered as two random variables to investigate the influence of uncertainty in these two mechanical parameters on seismic performance assessment of the frame building. The results show that the uncertainty in these two mechanical parameters could be negligible for spectra acceleration levels that cause slight damage or completely serious damage (structure collapse or failure) for the frame but have a relatively moderate effect on seismic performance assessment for spectra acceleration levels that lead to serious damage but not collapse or failure for the frame.

This paper proposes a novel learning algorithm, the transfer ensemble neural network (TENN) model, to increase the performance of shear capacity predictions on small datasets, illuminating the usefulness of advanced machine learning techniques in general. By incorporating ensemble learning and transfer learning, the TENN model is designed to control the high variability inherent in machine learning models trained on small amounts of data. The novel TENN model is validated to predict the shear capacity of deep reinforced concrete (RC) beams without stirrups across varying data availability levels. Knowledge acquired through pretraining a model on slender RC beams is utilized for training a model to better predict the shear capacity of deep RC beams without stirrups. To evaluate the performance of the TENN model, three baseline models are developed and examined across multiple data availability levels. The novel TENN model outperforms the baseline models, particularly when trained on a very limited dataset. Furthermore, the proposed algorithm achieves a higher accuracy than the currently accepted design standards in accurately predicting deep RC beams' shear capacity and demonstrates the capabilities of the TENN model to extrapolate in other domains where large‐scale or physical testing is cost‐prohibitive.

Machine learning (ML)-based data-driven methods have promoted the progress of modeling in many engineering domains. These methods can achieve high prediction and generalization performance for large, high-quality datasets. However, ML methods can yield biased predictions if the observed data (i.e., response variable y) are corrupted by outliers. This paper addresses this problem with a novel, robust ML approach that is formulated as an optimization problem by coupling locally weighted least-squares support vector machines for regression (LWLS-SVMR) with one weight function. The weight is a function of residuals and allows for iteration within the proposed approach, significantly reducing the negative interference of outliers. A new efficient hybrid algorithm is developed to solve the optimization problem. The proposed approach is assessed and validated by comparison with relevant ML approaches on both one-dimensional simulated datasets corrupted by various outliers and multi-dimensional real-world engineering datasets, including datasets used for predicting the lateral strength of reinforced concrete (RC) columns, the fuel consumption of automobiles, the rising time of a servomechanism, and dielectric breakdown strength. Finally, the proposed method is applied to produce a data-driven solver for computational mechanics with a nonlinear material dataset corrupted by outliers. The results all show that the proposed method is robust against non-extreme and extreme outliers and improves the predictive performance necessary to solve various engineering problems.

Pedestrian deaths account for 23% of all road traffic fatalities worldwide. After declining for three decades, pedestrian fatalities in the United States have been increasing with 6,941 fatalities in 2020, the highest number for more than two decades, impeding progress toward a zero-deaths transportation system. The Pedestrian and Bicycle Crash Analysis Tool (PBCAT) was developed to describe the pre-crash actions of the parties involved to better define the sequence of events and precipitating actions that lead to crashes involving motor vehicles and pedestrians or cyclists. Undoubtedly, police crash data influence decision-making processes in the transportation agencies. Using crash data from three major cities in Texas (during the period from 2018 to 2020), this study assessed the data quality of police-reported crash narratives on pedestrian-involved traffic crashes. As the pedestrian crash typing involves many categories, conventional machine-learning algorithms will not be sufficient in solving the classification problem from narrative texts. This study used few-shot learning (FSL), an advanced machine learning, to solve this issue. Using the pre-knowledge obtained from five different crash types and a few labeled data points of three unseen new crash types, the proposed model achieved roughly 40% overall accuracy. Also, four different configurations of crash types were formed and tested which indicates that the model is robust.

Current wind standards assume identical external wind pressure coefficients on roof soffits and adjacent wall surfaces. However, recent experiments suggest that this correlation significantly decreases with an increase in overhang width. This research introduces a few-shot learning model addressing the need for a machine learning approach that can efficiently obtain the wind pressure coefficients on roof soffits when the overhang width is large (i.e., larger than 2 ft). This research proposed a few-shot learning model to extrapolate the wind-induced pressures on roof soffits for low-rise buildings based on the wind tunnel dataset investigating the three large overhang widths (i.e., 2.4, 4.8, and 7.2 inches in 1:10 scale models). Prior knowledge relating to zonal information and wind directions shown in the standard of minimum design loads and associated criteria for buildings and other structures (i.e., ASCE 7) is incorporated into the model. The proposed few-shot learning model was trained on scale model buildings with overhang widths of 2.4 and 4.8 inches and tested on a 7.2-inch overhang width case. When predicting the minimum wind pressure coefficient for both the southside and eastside soffit surfaces, low mean-squared errors and high coefficient of determination values were observed. This study marks the first application of few-shot learning techniques to extrapolate wind pressures across different roof overhang widths and provide reliable predictions that outperform the weak correlation between the soffit and the adjacent wall surface assumed currently. This model reduces reliance on physical wind tunnel experiments and requires only a low-resolution measurement tap configuration.

Transfer learning aims to extract knowledge from one or more source tasks and apply the knowledge to a different task for more accurate predictions. The main purposes of this study are to investigate different knowledge transfer techniques, apply them to accurately predict the lateral strength of reinforced concrete columns with only a small amount of training data, and compare the transferability of each method. According to the various source and target domains, three different experiments are designed to directly compare the performance across section type, shear reinforcement area, and concrete compressive strength. In all cases in this study, knowledge transfer techniques show better prediction performance than the models trained without any knowledge transfer techniques. Therefore, we can conclude that transferring pre-trained knowledge from the source domain enables a model to better explain the response variable in the target domain. The performance improvement is particularly emphasized when the available data for the target domain is small. Thus, transfer learning can be one way to address the data scarcity problem in structural engineering. Furthermore, transferring the pre-trained knowledge is more associated with the underlying physical relationship between the source and the target domains and less associated with the discrepancy between the source and the target domain distributions.

Load-posted bridges have potential effects on traffic, commerce, and emergency egress due to detours in routes between origins and destinations. Posted bridges can also be problematic for state departments of transportation from a management standpoint because they may call for more frequent maintenance, monitoring, and inspection. For these reasons, it is beneficial for states to minimize the number of load-posted bridges. This study investigates a continuous steel plate girder bridge superstructure and proposes a methodology for exploring the refined load rating of this bridge type in a safe manner through refined modeling and load testing. Load test results were used to examine live load distribution factors, and finite-element models were developed to refine load ratings. It was determined that the refined load-rating factors for the bridge are significantly higher than the currently posted limits, and therefore, the posting could be removed. The methodology presented in this paper can potentially be used to increase the load-rating factors for similar continuous steel girder bridges.

Lateral stiffness of structural components, such as reinforced concrete (RC) columns, plays an important role in resisting the lateral earthquake loads. The lateral stiffness relates the lateral force to the lateral deformation, having a critical effect on the accuracy of the lateral seismic response predictions. The classical methods (e.g. fiber beam–column model) to estimate the lateral stiffness require calculations from section, element, and structural levels, which is time-consuming. Moreover, the shear deformation and bond-slip effect may also need to be included to more accurately calculate the lateral stiffness, which further increases the modeling difficulties and the computational cost. To reduce the computational time and enhance the accuracy of the predictions, this article proposes a novel data-driven method to predict the laterally seismic response based on the estimated lateral stiffness. The proposed method integrates the machine learning (ML) approach with the hysteretic model, where ML is used to compute the parameters that govern the nonlinear properties of the lateral response of target structural components directly from a training set composed of experimental data (i.e. data-driven procedure) and the hysteretic model is used to directly output the lateral stiffness based on the computed parameters and then to perform the seismic analysis. We apply the proposed method to predict the lateral seismic response of various types of RC columns subjected to cyclic loading and ground motions. We present the detailed model formulation for the application, including the developments of a modified hysteretic model, a hybrid optimization algorithm, and two data-driven seismic response solvers. The results predicted by the proposed method are compared with those obtained by classical methods with the experimental data serving as the ground truth, showing that the proposed method significantly outperforms the classical methods in both generalized prediction capabilities and computational efficiency.

Existing physics-based modeling approaches do not have a good compromise between performance and computational efficiency in predicting the seismic response of reinforced concrete (RC) frames, where high-fidelity models (e.g., fiber-based modeling method) have reasonable predictive performance but are computationally demanding, while more simplified models (e.g., shear building model) are the opposite. This paper proposes a novel artificial intelligence (AI)-enhanced computational method for seismic response prediction of RC frames which can remedy these problems. The proposed AI-enhanced method incorporates an AI technique with a shear building model, where the AI technique can directly utilize the real-world experimental data of RC columns to determine the lateral stiffness of each column in the target RC frame while the structural stiffness matrix is efficiently formulated via the shear building model. Therefore, this scheme can enhance prediction accuracy due to the use of real-world data while maintaining high computational efficiency due to the incorporation of the shear building model. Two data-driven seismic response solvers are developed to implement the proposed approach based on a database including 272 RC column specimens. Numerical results demonstrate that compared to the experimental data, the proposed method outperforms the fiber-based modeling approach in both prediction capability and computational efficiency and is a promising tool for accurate and efficient seismic response prediction of structural systems.

While road navigation systems seek to determine the shortest routes between a given set of origin and destination points, there are certain situations in which the fastest route increases the risk of being involved in road crashes. This implies the necessity of integrating safe route-finding into road navigation systems. This study is designed to synthesize the literature on safe route-finding and identify the gaps in the literature for future research. Specifically, a scoping literature review methodology is applied to understand how safety is incorporated in route-finding, even beyond motor vehicle navigation systems. Three databases (Scopus, Web of Science, and IEEE Xplore) are explored, and controlling for inclusion criteria, 40 studies are included in this review. The findings of this review indicated five areas through which safety was considered in route-finding: motor vehicle navigation, public safety, public health, pedestrian and cyclist navigation, and hazardous material transportation. The measurement of safety was found challenging with inconsistencies in safety quantification approaches. The safe route-finding algorithms were investigated based on their predictive/reactive, static/dynamic, and centralized/decentralized characteristics. Based on the critical review of the safe route-finding algorithms, availability of real-time data sources, accurate real-time and disaggregated crash risk prediction models, trade-off between time and safety in road navigation tools, and centralized safe route-finding are highlighted as the requirements and challenges in considering safety in road navigation systems. This study outlines a research agenda to address the identified challenges in safe route-finding.