Journal of Computing in Civil Engineering

Effective distress detection and quantitative analysis play a crucial role in road maintenance and driving safety. The Pavement distress segmentation network (PDSNet) is designed to combine the pyramid scene parsing network (PSPNet) and U-Net, providing both prior global information and local features that can overcome the common detection issues on the pavement data set faced by a single network. This paper proposes an efficient and improved architecture of PDSNet called PDSNet II for enhanced global modeling and retrieving fine details capacities. The proposed PDSNet II represents two major modifications on the original PDSNet. Firstly, a shifted window based on fully connected conditional random fields (FC-CRFs) layer is purposefully introduced to provide connections among consecutive self-attention layers that significantly enhance modeling power. Secondly, PDSNet II adopts multiple-head attention mechanisms to capture diverse interaction information across multiple projection spaces. Consequently, the output maps from the pyramid pooling module (PPM) head and the U-Net tail are fed into a neural window FC-CRFs layer. PDSNet II was trained using a data set consisting of 12,648 two-dimensional (2D) intensity and three-dimensional (3D) range images depicting various pavement conditions. The experimental results demonstrate that PDSNet II outperforms the original PDSNet in terms of F1-score and intersection over union (IoU). Compared with state-of-the-art networks, PDSNet II exhibits superior performance in detecting complex distress patterns, while effectively reducing noise and maintaining robustness. Overall, the proposed PDSNet II framework shows promising results in pavement distress segmentation, highlighting its potential for practical applications.

Amid escalating global challenges such as population growth, pollution, and climate change, access to safe and clean water has emerged as a critical issue. It is estimated that there are 4 billion cases of water-related diseases worldwide that cause 3.4 million deaths every year. This underscores the urgent need for efficient water quality monitoring and assessment. Traditional assessment techniques include laboratory-based methods that are manual, costly, time-consuming, and risky. Although some studies leveraged Internet of Things (IoT)-based systems to examine water quality, they only relied on a limited number of water quality parameters (and thus do not offer a comprehensive and accurate water quality assessment), mainly due to the technical difficulties to integrate multiple sensors to a single device. In fact, due to the issues of multimodality, heterogeneity, and complexity of data, the interoperability among sensors with various measurements, sampling rates, and technical requirements makes it very challenging to seamlessly integrate multiple sensors into one device. This study overcame these technical challenges by leveraging multisensor data fusion capabilities to develop an intelligent cloud-based IoT multimodal edge sensing device to provide an automated, real-time, and comprehensive assessment process of water quality. First, a total of nine water quality parameters were identified and considered. Second, the sensing device was designed and developed using an ESP32 embedded system, which is a low-cost, low-power system on a chip (SoC) microcontroller integrated with Wi-Fi and dual-mode Bluetooth connectivity by fusing data from six different sensors that measure the nine identified water parameters on the edge. Third, the overall water quality was evaluated using the National Sanitation Foundation Water Quality Index (NSFWQI). Fourth, a cloud-based server was created to publish the data instantly, and a graphical user interface (GUI) was developed to provide easy-to-understand real-time visualization and information of the water quality. The real-world applicability and practicality of the developed IoT-enabled sensing device was tested and verified in a pilot project (i.e., a case study) of a building located in Newark, New Jersey, for a duration of 6 months. This paper adds to the body of knowledge by being the first research developing a single IoT-enabled device that is capable of reporting NSFWQI in real-time based on 9 water quality indicators encompassing both physical [temperature, total dissolved solids (TDS), turbidity, and pH] and chemical [potassium, phosphorus, nitrogen, dissolved oxygen (DO), and 5-day biochemical oxygen demand (${\mathrm{BOD}}_5$)] parameters. Thus, this study serves as a multifaceted improvement across different dimensions, fostering healthier, more efficient, and technologically advanced environments.

Abstract

With the digital transformation of the construction industry, the need to improve construction business process collaboration and automation is increasing. However, as construction projects usually involve many stakeholders with complex relationships and insufficient mutual trust, the existing technologies cannot fulfill the requirements of the construction industry. This paper explores the integration of blockchain, smart contracts, and process automation technologies into construction business processes, shedding light on both managerial and technical challenges. The study introduces a comprehensive technical framework meticulously designed to align with the intricate nature of construction practices while focusing on the resolution of legal complexities associated with smart contract applications and the facilitation of process collaboration and automation. The framework comprises three core modules: (1) action definition and extension to process models, (2) standardized mapping for automatic smart contract generation, and (3) smart contract visualization for reliable process collaboration and automation. This multifaceted approach caters to the specific needs of construction management, standardization, interoperability, and visualization. The framework’s practicality is further evaluated through real-world testing within a construction payment case, effectively showcasing its efficacy and applicability in tangible business scenarios. While this paper represents a significant step forward in addressing construction business process collaboration and automation challenges, it acknowledges the necessity for ongoing research and development to refine and expand these innovative solutions to meet the evolving demands of construction management.

Practical Applications

Currently, there is a significant escalation in the severity of disputes within the construction industry. The adoption of blockchain technology allows for the redistribution of trust within a construction project, shifting reliance from individuals to the system itself for better collaboration and fewer disputes. Although blockchain can help foster a more secure and transparent environment, at the process level, an efficient workflow engine is needed to activate processes in this environment. Therefore, this study proposed an approach to generating smart contract codes based on construction business processes in a standardized way and further visually executing these codes for the processes with historical transactions stored in blockchain. This approach can reduce the effort of smart contract programming and improve the standardization and credibility of smart contract generation. It can also streamline the procedure from construction business process design to final execution via predefined and visual modules. Meanwhile, the generation prototype is open sourced to provide transparency and further enhance the approach. To indicate practical implications in the real world, this study provided a case of construction payment, but the solution is not limited to the payment scenario.

The construction industry has long been plagued by low productivity and high injury and fatality rates. Robots have been envisioned to automate the construction process, thereby substantially improving construction productivity and safety. Despite the enormous potential, teaching robots to perform complex construction tasks is challenging. We present a generalizable framework to harness human teleoperation data to train construction robots to perform repetitive construction tasks. First, we develop a teleoperation method and interface to control robots on construction sites, serving as an intermediate solution toward full automation. Teleoperation data from human operators, along with context information from the job site, can be collected for robot learning. Second, we propose a new method for extracting keyframes from human operation data to reduce noise and redundancy in the training data, thereby improving robot learning efficacy. We propose a hierarchical imitation learning method that incorporates the keyframes to train the robot to generate appropriate trajectories for construction tasks. Third, we model the robot’s visual observations of the working space in a compact latent space to improve learning performance and reduce computational load. To validate the proposed framework, we conduct experiments teaching a robot to generate appropriate trajectories for excavation tasks from human operators’ teleoperations. The results suggest that the proposed method outperforms state-of-the-art approaches, demonstrating its significant potential for application.

Developing deep learning network models for computer vision applications in concrete damage detection is a challenging task due to the shortage of training images. To address this issue, this study proposes a novel style-controlled image synthesis method for concrete damages based on the fusion of a convolutional encoder and an attention-enhanced conditional generative adversarial network. This makes it possible to generate effective images that can improve the damage detection performance of deep learning networks. To achieve this, a network architecture for concrete damage image synthesis, named DamageGAN-AE, was designed by fusing a convolutional encoder and an attention-enhanced conditional generative adversarial network. The DamageGAN-AE networks with different attention modules were trained, and the training results show that the well-trained DamageGAN-AE enhanced by coordinate attention is the best model for concrete damage image synthesis. The well-trained DamageGAN-AE was compared with the current competing methods to verify its performance. The DamageGAN-AE with image encoder was trained to implement the style-controlled image synthesis. Finally, the generated concrete damage images with diverse styles by the DamageGAN-AE model with image encoder were used to train deep learning networks. The results indicate that the generated style-controlled concrete damage images by the proposed method can effectively improve the concrete damage detection performance of deep learning networks.

The prediction of the time-dependent behavior of high-performance concrete (HPC) structures is essential to evaluating their service life. This prediction relies on the shrinkage and creep properties of HPC. However, unlike conventional concrete, the binary and ternary composite system of supplementary cementitious materials (SCMs) in HPC has demonstrated different shrinkage and creep properties. This difference makes it challenging to accurately predict these properties using existing material models in code and standard practices, i.e., ACI, fib, B4, and GL. These models exhibit significant deviations in prediction under standard statistical evaluation methods due to the influence of SCMs. To overcome this challenge, intelligent artificial neural network (ANN) models have been developed using a feed-forward backpropagation training algorithm. The ANN models consider a widely compiled indigenous database of shrinkage and creep and consist of the most realistic experimental relationship with the effecting extrinsic and intrinsic key parameters. These parameters include standard concrete mix design material proportions, mechanical and physical properties, environmental conditions, and aging factors to obtain shrinkage and creep properties of HPC. All concrete material properties influencing the behavior of shrinkage and creep have been related based on experimentally measured results and incorporated as input parameters in both intelligent developed ANN models. The accuracy of prediction of both ANN models has been substantiated by the experimentally measured database and existing material models as comparative appraisals using several statistical metrics. The developed ANN models to predict such complex nonlinear properties of HPC are more practical and beneficial than existing material models, which will help to fulfill sustainable development and improve the service life of HPC structures.

Asphalt pavement performance is crucial for the sustainable management of road infrastructure. However, achieving accurate predictions remains challenging due to the complex interactions among materials, environmental factors, and traffic loads. In this study, the optical microscope algorithm–least squares moment balanced machine (OMA-LSMBM), an AI-based inference engine, was developed to enhance the accuracy of asphalt performance prediction. This approach integrates machine-learning techniques with optimization algorithms. In the proposed model, LSMBM considers moments to determine the optimal hyperplane, a backpropagation neural network assigns weights to each datapoint, and an OMA optimizes the LSMBM hyperparameters and identifies the optimal feature subset combination. The proposed model was tested using three simulations, i.e., benchmark functions, pavement surface temperature, and asphalt mixture flow number. OMA-LSMBM demonstrated the best function approximation performance, improving the performance metrics and achieving a root mean square error value for pavement temperature prediction that was 6.49%–72.62% less than the comparison models. In terms of predicting flow number, the proposed model showed superior performance over the comparison models with a 11.15%–54.83% lower error rate. These results demonstrate the OMA-LSMBM significantly enhances the accuracy of asphalt performance predictions, which may be directly applied to improving road maintenance strategies and planning activities.

Geopolymer concrete (GPC) is an extraordinary material for promoting sustainable development in the construction industry and reducing environmental risk. However, material properties, such as compressive strength, are commonly determined using laboratory experiments, which are costly and time-consuming to run. Therefore, optical-inspired rain forest (ORF), a sophisticated predictive model, was developed to offer an alternative mathematical solution. The developed model uses a novel mechanism that grows an operation tree into multiple operation forests and employs an optical microscope algorithm to optimize the weight and forest topology. The experimental results indicate that the proposed model outperformed several other popular artificial intelligence approaches, achieving the highest evaluation criteria of $\mathrm{RI}=0.973$ and $\mathrm{RI}=0.979$, respectively, for training and testing data sets. Hence, ORF is recommended as a viable tool to assist material engineers to significantly increase the utilization of GPC in construction projects.

Residential buildings in the US are substantial energy consumers, accounting for 39% of the country’s electricity usage and 22% of its total energy consumption. The dynamics of this consumption are intricately linked to the presence and activities of occupants. As such, a precise analysis of occupancy patterns is vital to gaining an informed understanding of the changing trends in energy use. This study harnesses data from the American Time Use Survey (ATUS) to delve into the influence of sociodemographic features on individuals’ occupancy patterns throughout the day. Employing statistical methods and exploratory machine-learning techniques, this study aims to map American occupancy patterns and investigate the impact of various demographic features on these patterns. Six key features that have a predominant effect on occupancy patterns are identified as age, gender, employment status, family income, household type, and day of the week. A predictive model has also been developed to model occupancy patterns of individuals based on the identified features using artificial neural networks (ANN). Comprehending how these features shape residential occupancy is crucial for devising specific energy-conservation strategies for residential buildings. The contributions of this research extend the current understanding of energy-efficient architecture design, providing valuable insights for stakeholders and policymakers in the energy sector.

Engineering materials in constructed systems in service exhibit complex patterns, including structural damage, environmental artifacts, and artificial anomalies. In recent years, machine vision methods have been extensively studied, most of which train models using regular grey or color images in the visible bands and label at pixel levels with a large volume of data. The authors propose using hyperspectral imaging (HSI) for structural material condition assessment in this work. Compared with visible images, the research challenge is that HSI pixels with high-dimensional spectral profiles are beyond human perceptive capabilities with hidden discriminative power. Learning from labeled and unlabeled data is one direct approach to unlocking this power. A deep neural network-enabled spatial-spectral feature extraction and a semisupervised learning architecture were developed in this work. A human-in-the-loop (HITL) framework was comparatively studied with three incremental training-data configuration schemes. The paper concludes with the following empirical findings: (1) fully supervised learning determines the baseline of the detection performance; (2) an extensive range of ratio values exists between the unlabeled and the labeled data for incremental semisupervised learning, and a $1:1$ ratio can be taken as a conservative and operational ratio; and (3) with parametric semisupervised learning with equal labeled and unlabeled data participation, the proposed HITL operational workflow can be implemented as a practical approach for HSI-based structural material and damage detection.

Abstract

The metro, renowned as an environmentally friendly mode of transportation due to its low energy consumption and minimal pollution, plays a crucial role in achieving sustainable urban growth. Due to the scarcity of information in the early stages of metro construction projects and the subjectivity of cost estimation (which relies heavily on the estimator’s experience), it is always difficult to guarantee the accuracy of metro construction project cost estimation. Furthermore, the existing methodological models commonly used for cost estimation do not adequately consider the interpretability of the estimation results, making it difficult to promote their application in real-world scenarios. In this paper, an interpretable machine learning method is introduced into the study of cost estimation of metro construction projects, and a maximum relevance and minimum redundancy (mRMR)–light gradient boosting machine (LightGBM)–Shapley additive explanations (SHAP) interpretable assisted investment decision-making framework is proposed. The results show that the negative impact of variable multicollinearity on model prediction is avoided by quantitatively identifying the key driving variables of costing through mRMR based on the historical data of metro construction projects and macroeconomic data. LightGBM is employed to predict the cost of metro construction projects with a mean absolute percentage error of 13.00%, surpassing the accuracy of the five baseline models. The SHAP method’s introduction explains the influence of key driving variables on the model prediction response at both global and local levels, which improves the decision trust of the cost estimation of metro construction projects. The study takes into account the impact of key driving variables on the model’s prediction response in a real-world context and balances the needs for estimation accuracy and variable interpretability in real-world scenarios.

Practical Applications

In this study, we propose a new interpretable machine learning method that helps improve the decision trust of cost estimation for metro construction projects. We use the mRMR module to identify the key driving variables for cost. Based on the LightGBM module, we accurately fit the nonlinear relationship between the cost of metro construction projects and the key variables. The SHAP algorithm is introduced to analyze the effects of the key driving variables on the prediction of the model at both global and local levels. The combined mRMR-LightGBM-SHAP model is proven to have better prediction accuracy than the traditional prediction model under the same conditions, and the SHAP response results prove the reliability of the model prediction. The study also found that the inclusion of economic environment variables plays a positive role in improving the accuracy of cost estimation for metro construction projects. The research results can be mainly used by metro investment units to quickly confirm or review the cost of metro construction projects, and they can also help government departments pay close attention to the impact of the external economic environment on the cost of metro construction.

In the dynamic construction environment, workers’ safety heavily depends on their ability to effectively perceive and react to hazards. Accordingly, studies have assessed the status of workers’ risk perception using advanced technologies. However, these studies have mainly focused on whether risks are perceived rather than how they are perceived. Recognizing the need for effective safety interventions that address risk-perception failures, it becomes crucial to not only classify workers’ risk-perception states but also to delve into the underlying processes of their risk perception. To address this research gap, this study examines the critical aspect of risk perception in construction safety by employing functional near-infrared spectroscopy (fNIRS) and 360° panoramas from actual construction sites to assess workers’ cognitive processes during hazard identification. Classifiers were developed using the AutoML method, and 15 advanced machine learning algorithms were compared to identify the highest-performing model. This model would then be utilized to understand the risk-perception process by incorporating the feature-importance technique. The results indicate that CatBoost emerged as the most effective classifier, achieving an accuracy rate of 90.3%. Additionally, the results identify significant brain activations in four anatomical locations: the prefrontal cortex, frontal eye fields, primary motor cortex, and primary auditory cortex. Notably, there is a significant correlation between these areas, emphasizing the importance of both visual and auditory cue perception in shaping workers’ situational awareness. This research highlights the potential of neuroimaging fNIRS in improving construction safety, and the importance of auditory perception in hazard identification, offering insights that could enhance the effectiveness of safety training programs.

To bridge the gap in as-built Building Information Model (BIM) creation between the architectural, engineering, and construction (AEC) community and the computer vision community, this paper presents an automated Scan-to-BIM framework for modeling both structural and nonstructural building components using a low-cost scanning data. The state-of-the-art instance-level semantic segmentation algorithm, SoftGroup, is adopted to classify individual building components. Detected wall segments are projected onto a two-dimensional (2D) XY grid, and an interest point detection algorithm, SuperPoint, is used to extract wall corner points. Subsequently, a series of refinement steps is proposed to generate the wall boundary. With optimized parameters, an intersection-over-union of 82.56% was achieved when tested on the benchmark Stanford Three-Dimensional (3D) Indoor Scene Data Set. Our results demonstrated the usability of the proposed wall boundary extraction to the incomplete and complex indoor scan data compared to an existing as-built modeling method. Instance-level segments and the refined wall boundary were combined to generate as-built BIM via parametric modeling.

In the construction process of prefabricated buildings, collision detection of the tower crane’s lifting path is of paramount importance. During the actual operation of a tower crane, attention is not only given to whether components collide with obstacles but also to calculating the minimum distance from hoisted components to obstacles and identifying the closest obstacle’s ID. This information provides operators or path planners with valuable insights to prevent accidents. This study addresses this issue by proposing a minimum distance calculation method for collision issues in lifting construction scenarios. First, based on the motion characteristics of tower crane hoisting, we employ vertically aligned oriented bounding boxes (VA-OBB) to fit the structural elements in the scene. VA-OBBs are compact bounding volumes (BV) with a rapid update rate. Subsequently, a method for calculating the minimum distance between VA-OBBs is introduced. With this approach, specific calculations are proposed for different risk scenarios in lifting construction. Using real building information modeling (BIM) data from a prefabricated construction scenario as a case study, we conducted an analysis that recorded the minimum distance information throughout the lifting path. This aids in analyzing the causes of accidents and rectifying unsafe path planning. Finally, we evaluate the accuracy and efficiency of the algorithm. Empirical results demonstrate that the algorithm can perform real-time and accurate collision calculations during the hoisting process of prefabricated construction. Additionally, this paper discusses the limitations of the research method and suggests directions for future improvements, providing a viable solution for collision issues in the field of hoisting construction.

This study proposes a feature relationship algorithm (FRA) to reconstruct three-dimensional wireframes of objects with non-Manhattan shapes using segmented point clouds. Instead of relying on extracting target boundaries, the FRA systematically identifies the vertex and edge nodes of objects and uses an innovative linking strategy to reconstruct a precise wireframe based on the point cloud geometry, even in the presence of data gaps. The FRA exhibits adaptability to various shapes, including curves, cones, pyramids, cylinders, octagonal prisms, and combinations. Validations on synthetic data provide valuable insights into the FRA’s parameter tuning and exceptional shape accuracy. On the other hand, the use of Light Detection and Ranging scans and benchmark data underscores the FRA’s fidelity in representing shapes from point clouds and demonstrates its improvement over baseline methods.

The thorough investigation into the evolution of concrete performance under sulfate attack environments holds significant importance for engineering applications in specific conditions. In this paper, a prediction model for the two evaluation indexes of sulfate attack resistance of concrete (SARC), namely compressive strength corrosion resistance coefficient and mass loss rate, is established based on four machine-learning algorithms: Support Vector Regression, Random Forest Regression, Gradient Boosting, and Extreme Gradient Boosting (XGB). A comparison of the various performances showed that the model based on the XGB algorithm had the strongest generalization ability and offered the best prediction of SARC (K test set ${\mathrm{R}}^2=0.963$, MLR test set ${\mathrm{R}}^2=0.903$). Feature importance and partial correlation analyses were performed for the two XGB models separately, and a graphical user interface was designed based on the two predictive models. The results reveal that the number of cycles, water-binder ratio, and cement content significantly influence the SARC. Moderately increasing cement, fly ash, and coarse aggregate content can enhance the SARC. Increasing the number of cycles, drying time, water-binder ratio, sand, and solution concentration will reduce the SARC. Therefore, measures such as moderately increasing the amount of cement, reducing the water-binder ratio, and increasing the fly ash content can be increased to improve the SARC, but overuse has no significant effect.