Journal of Computing in Civil Engineering

Neural networks (NNs) have seen an increase in popularity in the last few years. As found in several papers, they outperformed other machine learning algorithms and have improved their applicability. NNs have shown potential when working with prediction and detection algorithms by addressing a large variety of problems, such as recognition of heavy machinery, project success prediction, workers’ pose assessment, price estimations, and project productivity estimation. To understand the future potential of NNs, we completed a bibliometric analysis of the existing literature, including publications since 2010. The areas in which NNs have been used in construction applications are categorized to understand their connections and underlying architectures. This work found a wider field of applications for NNs in the construction industry than originally known. New architectures such as transformer networks have not been explored fully in construction research but could lead to higher-performing networks. As far as the authors know, this is the first review to solely focus on construction, excluding areas such as structural engineering, indoor climate, occupancy modeling, and energy analysis. The limitations of NNs are discussed, and a path forward is proposed, which includes real-time models and examination of new architectures, which would allow the construction research to fully exploit the potential of NNs.

Particle shape analysis of coarse aggregates is important to ensure the quality of cement and asphalt concrete mixtures. Conventional methods for measuring the aggregate particle size, such as manual calipers or mechanical sieving, are time consuming and labor intensive. In addition, the accuracy of image processing techniques is severely limited by shadows and heterogeneous backgrounds. Hence, we developed an automatic stereo vision-based inspection system (SVIS) for the identification and shape analysis of coarse aggregate particles. We integrated a cascaded deep learning model into the SVIS to identify the types of coarse aggregate particles under offsite working conditions. Moreover, we combined deep learning and stereo vision techniques to calculate the unit conversion factors and the thickness of each particle to facilitate particle shape analysis. The precision and recall metrics obtained from the training model were $\ge 96.0\%$ for particle detection and $\ge 95.7\%$ for particle segmentation. In the experiment, the proposed inspection system accurately determined the particle size of coarse aggregates with measurement errors of $\le 4.96\%$ compared with the ground truth. Thus, the proposed system overcomes the shortcomings of image processing technologies and considerably aids the decision-making process during onsite material inspection.

In this paper, a novel method is proposed based on a windowed one-dimensional convolutional neural network (1D CNN) for multiclass damage identification using vibration responses of a full-scale bridge. The measured data are first augmented by extracting samples of windows of raw acceleration time series to alleviate the problem of a limited training data set. 1D CNN is developed to classify the windowed time series into multiple damage classes. The damage is quantified using the predicted class probabilities, and the damage is localized if the predicted class is equal to the assigned damage class, exceeding a threshold associated with majority voting. The proposed network is optimally tuned with respect to various hyperparameters such as window size and random initialization of weights to achieve the best classification performance using a global 1D CNN model. The proposed method is validated using a benchmark bridge data for multiclass classification for two different damage scenarios, namely, pier settlement and rupture of tendons, under the various extents of damage. The damage identification is carried out on various bridge components to collectively identify the structural component with a damaged signature. The results show that the proposed windowed 1D CNN method achieves an accuracy of 97%, and performs well with different types of damage.

Overhead work involving the construction and maintenance of civil infrastructure (e.g., tunnels, overpasses, and buildings) is strenuous and fatigue-inducing for human workers and is particularly well-suited for co-robotization. Such work is typically quasi-repetitive, and on-site robots must adapt to unexpected workface conditions. Methods such as learning from demonstration can leverage human experts’ demonstration to let robots directly learn new skills to perform tasks. This paper proposes a generalized cylinders with orientation approach to teach robots how to perform quasi-repetitive overhead construction tasks from human demonstration. The demonstration trajectories are first used to construct a generalized cylinder and generate the robot trajectory. To ensure that the construction component (e.g., tunnel lining segment, building ceiling tile) being installed can satisfy the geometric constraints of the workspace, orientation constraints need to be determined, and the robot must follow such constraints. A trajectory adaptation and human-in-the-loop refinement approach are developed to refine the robot trajectory. The proposed method was evaluated in a robot simulator with variable workspace. The results showed that the proposed approach achieves an improved success rate (82.0%) compared to that demonstrated in previous work (71.3%) and enables overhead construction robots to readily adapt to new worksite conditions.

To achieve meaningful results, data-driven decision-support systems in construction require the integration of fragmented data from multiple standalone databases. In practice, a manual brute-force approach is often the only available means of integrating structured, yet semantically-ambiguous, construction data. Two common data integration challenges include the identification of (1) key strings (i.e., product identification) partially shared between two data sources; and (2) relationships (overlap, included, or outside) between two 3D object lists based on coordinates. This research has developed a framework that includes two generic solutions to the identified semantic mapping challenges. The proposed framework automatically integrates fragmented and incompatible data (exhibiting similar semantic mapping challenges) from various sources into a tidy format for input into a diverse range of industrial construction applications. Verification and functionality of the framework were confirmed using both artificial data and a real case study of a large oil-and-gas project. The ability of the proposed data integration functions and framework to automate otherwise manual data integration processes was demonstrated. Results of this study are expected to enhance real-time information flow, improve data quality, and promote the use of fragmented data for critical decision support in practice.

Seismic risk assessment of road systems involves computationally expensive traffic simulations to evaluate the performance of the system. To accelerate this process, this paper develops a neural network surrogate model that allows rapid and accurate estimation of changes in traffic performance metrics due to bridge damage. Some of the methodological aspects explored when calibrating this neural network are defining sampling protocols, selecting hyperparameters, and evaluating practical considerations of the model. In addition to the neural network, a modified version of the local interpretable model-agnostic explanation (LIME) is proposed as a retrofitting strategy that minimizes earthquakes’ impact on the system. The modified version (LIME-TI) uses traffic impacts (TI) and rates of occurrence to aggregate the importance of individual damage realizations during the computation of variable importance. This study uses the San Francisco Bay Area road network as a testbed. As a conclusion of this study, the neural network accurately predicts the system’s performance while taking five orders of magnitude less time to compute traffic metrics, allowing decision-makers to evaluate the impact of retrofitting bridges in the system quickly. Moreover, the proposed LIME-TI metric is superior to others (such as traffic volume or vulnerability) in identifying bridges whose retrofit effectively improves network performance.

Urban built environments often include many negative stimuli (e.g., unleashed dogs, dead animals, litter, graffiti, abandoned vehicles) that are linked with stress symptomatology among urban populations. Biosignals (e.g., electrodermal activity, gait patterns, and blood volume pulse) can help assess pedestrian distress levels induced by negative environmental stimuli by overcoming the measurement limitations of traditional self-reporting methods and field observations. Despite their potential, biosignals from naturalistic outdoor environments are often contaminated by uncontrollable extraneous factors (e.g., movement artifacts, physiological reactivity due to unintended stimuli, and individual variability). Thus, more quantitative evidence and novel methodological approaches are required to accurately capture pedestrian environmental distress resulting from negative environmental stimuli. In this context, we investigate the interplay between pedestrians’ biosignal data and image-based data (built environment feature information and perceptual distress levels identified from images) in a machine learning model. Results from the statistical model estimated with the biosignal data demonstrated significant physiological responses to the negative environmental stimuli. The use of the features from image-based data increased the prediction accuracy of the computational model. This method can be applied to geospatial intelligence, further advancing built environmental assessments and evidence-based approaches to promote walking and walkable communities.

Permeability of subsurface porous media is one of the primary factors that affect fluid transport in porous rock. However, accurate prediction of rock permeability is a challenging task due to its intricate pore network. Development of digital rocks provides an effective approach to reveal and characterize the pore network. In this paper, a combination of digital rock petrophysics and ensemble machine learning (ML) models is proposed to improve the permeability prediction of subsurface porous media. The permeability of the numerically generated porous samples as outputs was determined by the lattice Boltzmann method (LBM). The five most important parameters (porosity, tortuosity, fractal dimension, average pore diameter, and coordination number) were selected as inputs for the permeability prediction. To improve the accuracy, feature selection and ML methods comparisons were conducted. Three feature selection methods based on expert knowledge, correlation coefficient, and importance score were compared. Moreover, a comparison was performed on six ML methods (support vector machine, artificial neural network, decision tree, random forest, gradient-boosting machine, and Bayesian ridge regression) that were optimized by particle swarm optimization (PSO). The results indicated that (1) the feature selection based on the expert knowledge obtained a higher performance than the groups based on the correlation coefficient and importance score, implying the importance of expert knowledge on feature selection, and thus on ML performance; (2) artificial neural network with hyperparameter tuning achieved the best performance in predicting permeability; and (3) the optimized ML method outperformed the empirical equations in predicting permeability. In conclusion, this study provides a fast and reliable approach predicting permeability of subsurface porous media based on numerically generated porous images. Moreover, the proposed framework can be further extended to determine other petrophysical properties, for example, the relative permeability and thermal conductivity.

In the aftermath of a disaster, natural or man-made, successful emergency management operations vitally depend on reliable communication infrastructures. However, large-scale extreme events may dismantle telecommunication infrastructures and impair rescue operations. The current state of practice to maintain communication when infrastructures are not operable mostly focuses on using temporary infrastructures including mobile telecom towers. However, these centralized solutions heavily depend on equipment that may not be readily available. To address this issue, this paper aims to create theoretical foundations for, and empirically examine the performance of, a novel decentralized and infrastructure-less communication system that uses an ad hoc mobile network, distributed ledger and Blockchain technology, and an all-to-all broadcasting routing algorithm. When the proposed communication system is used, mobile devices in the affected area connect with each other through device-to-device Wi-Fi Direct, create a temporary mobile ad hoc network, and share spatial information through the stepwise routing algorithm and distributed ledgers. At the end of this process, each device in the network will have spatial information of all available devices in the area. Therefore, when first responders arrive in the affected area, they can locate all devices and have access to all collected information immediately through a direct connection to only one device in the area. The outcomes of this study will fundamentally transform the emergency communication solutions and can significantly shorten the duration of rescue operations by helping first responders locate citizens in the affected area faster and more efficiently.