Human activity recognition (HAR) using machine learning has shown tremendous promise in detecting construction workers' activities. HAR has many applications in human-robot interaction research to enable robots' understanding of human counterparts' activities. However, many existing HAR approaches lack robustness, generalizability, and adaptability. This paper proposes a transfer learning methodology for activity recognition of construction workers that requires orders of magnitude less data and compute time for comparable or better classification accuracy. The developed algorithm transfers features from a model pre-trained by the original authors and fine-tunes them for the downstream task of activity recognition in construction. The model was pre-trained on Kinetics-400, a large-scale video-based human activity recognition dataset with 400 distinct classes. The model was fine-tuned and tested using videos captured from manual material handling (MMH) activities found on YouTube. Results indicate that the fine-tuned model can recognize distinct MMH tasks in a robust and adaptive manner which is crucial for the widespread deployment of collaborative robots in construction.
Since the pandemic, most work environments have changed to home-based settings, which mainly serve for living purposes other than working purposes. There is no lack of studies in the impact of indoor environmental quality (IEQ) factors on occupant productivity in regular work environments;however, limited studies are conducted in home-based work environments, not to mention in the context of the COVID-19 pandemic. Therefore, an online survey was developed to explore the impact of IEQ factors on productivity between office-and home-based work environments among occupants with different genders and ages. A comprehensive list of key indicators was first developed. Then a survey was developed, distributed, and received 204 complete responses. The descriptive analysis and t-test are performed to evaluate the impact difference of all the IEQ factors on productivity of different occupants. The findings indicate that the visual factors' impact on productivity decreases for both genders, the impact of all factors on productivity for younger occupants increases when work from home (WFH), and the impact of acoustic quality is the highest among all five IEQ factors. Because of WFH as the future of work, this study can provide insights for future built environment design. © 2021 Computing in Civil Engineering 2021 - Selected Papers from the ASCE International Conference on Computing in Civil Engineering 2021. All rights reserved.
In the last couple of decades, various industries have taken advantage of emerging advanced technologies, such as the digital twin, to find more effective solutions in their respective areas. In the transportation infrastructure sector, the implementation of digital twin technology is slowly gaining traction but lagging behind other major industries. To better understand the limitations, opportunities, and challenges for the adoption of digital twin technology in this sector, a survey questionnaire was distributed to collect information from industry professionals involved in transportation infrastructure projects. The purpose of this study is to understand how digital twin technology is being perceived by the industry. Based on the results of this survey, the current state of digital twins in the infrastructure sector, as well as the expected benefits, and potential challenges for the deployment of this technology are discussed here. This study aims to initiate discussion for future research in digital twin technology for transportation infrastructure applications.
HVAC systems account for 50% of buildings' energy use and could play a critical role in energy management in buildings both for energy-saving, and demand-side management to reduce peak energy use. According to contextual demands of occupants, efficient control of HVAC systems could result in peak energy saving and decreased energy costs in demand response programs. Accordingly, in this paper, we have introduced an agent-based model consisting of three agents: human agent, thermostat agent, and utility agent. In this model, the thermostat agent receives the real-time electricity price from the utility agent and aggregates thermal comfort profiles of the occupants from human agents. By considering these inputs, the thermostat agent employs a predictive model of a house and calculates the next set point of the HVAC system on energy cost and occupants' comfort. Then, the thermostat agent signals the suggested set point to the thermostat of the building at each time step. For evaluating the proposed controller's performance, the electricity price profile from ERCOT, which supplies the state of Texas, is used as a signal from the utility agent. Realistic thermal comfort data were used to simulate the thermal preference of occupants represented as human agents. The evaluation was carried out in a co-simulation using a Python and EnergyPlus model of a residential unit. Our results show that the proposed controller reduces the peak energy by 5.5% to 10% and increases occupants' thermal satisfaction up to 12%. The main contribution of this paper is developing an agent-based model that humans, as the main stakeholders of the buildings, play a role in controlling HVAC systems for peak reduction and energy saving.
The in-pipe inspection is one of the high frequent and challenging operations for facilities maintenance. Because the complex inner geometry and hazardous environments of pipeline systems are less accessible to humans, the in-pipe snake robots have been developed to facilitate the inspection of defects, corrosions, and the sampling of sludges. However, snake robot control is still nontrivial due to unique movements (such as helical motion) and environmental constraints like confined areas. The unique locomotion mechanisms of snake robots make it harder for the operator to perceive the remote environment and navigate the robot. This paper introduces an upper-body haptic simulator to support the snake robot controls with the haptic assistant. To prove the concept, a virtual model is built to simulate the in-pipe environments and the locomotion of a snake robot with a helical corkscrew motion. The proposed method collects signals relative direction of the gravity to the snake robot and convert the gravity direction information as vibrations of different magnitudes via an upper-body haptic suit with 40 vibrators. To demonstrate the potential of our simulator, a pilot study was performed, and the results showed that the operator could easily control the snake robot with the desired velocity in different shapes of the pipeline (straight pipes, T joints, and L joints) both horizontally and vertically.
Construction planning is a fundamental task in the management of construction projects. The most common modeling methods for construction planning are linear scheduling (LS), the critical path method (CPM), and discrete-event simulation (DES). While LS and CPM have advantages of visual feedback and simplicity, respectively, DES as a technique for simulating the behavior and output of a real-world process, facility, or system is the most versatile of these methods. However, when the constraints and dimensionality of a planning problem increase, DES methods are cumbersome and struggle to accurately reflect decision options. As an alternative to DES, deep learning artificial intelligence (AI) methods can, through the use of reinforcement learning, more rapidly review and recommend more planning options for scheduling complex construction projects. The goal of this paper is to examine the feasibility of deep learning AI as an alternative to the DES in an existing project planning method, named Foresight. Foresight is a graphical constraint-based method of planning manufacturing and construction processes that uses DES to simulate possible outcomes. This paper presents the process changes necessary in Foresight to accommodate deep learning algorithms, composed of multiple network layers. In addition, challenges of implementing deep learning AI in construction planning are discussed. Early project planning leads to improved performance in terms of cost, schedule, and operations, balancing the competing needs of a project. A new modeling paradigm is proposed that is better suited to the needs of contemporary construction project planning.
Dynamic models of occupancy patterns have shown to be effective in optimizing building-systems operations. Previous research has relied on CO2 sensors and vision-based techniques to determine occupancy patterns. Vision-based techniques provide highly accurate information; however, they are very intrusive. Therefore, motion or CO2 sensors are more widely adopted worldwide. Volatile organic compounds (VOCs) are another pollutant originating from the occupants. However, a limited number of studies have evaluated the impact of occupants on VOC level. In this paper, continuous measurements of CO2, VOC, light, temperature, and humidity were recorded in a 17,000 sq ft open office space for around four months. Using different statistical models (e.g., SVM, K-nearest neighbors, and random forest) we evaluated which combination of environmental factors provide more accurate insights on occupant presence. Our preliminary results indicate that VOC is a good indicator of occupancy detection in some cases. It is also concluded that a proper feature selection and developing appropriate global occupancy detection models can reduce the cost and energy of data collection without a significant impact on the accuracy.
Housing resilience planning is a dynamic process that involves multisector stakeholders, including public agencies, private industries, nongovernment organizations (NGOs), academia, and community residents. Despite the importance of multisector stakeholder collaboration, there is limited understanding of stakeholder collaboration in housing resilience planning. To address this gap, this study analyzes how multisector stakeholders collaborate in producing housing resilience-focused plans, reports, and guidelines utilizing social network analysis (SNA). A two-mode, stakeholder-document, SNA model was built based on secondary data collected from 39 documents on housing resilience in three regions, including the City of Miami, the City of Miami Beach, and Miami-Dade County. The network analysis shows that there are significant differences in network measures across different stakeholder sectors. The findings from this study could offer insights on how to facilitate more effective and collaborative housing resilience planning.
While the use of deep neural networks (DNN) for computer vision is increasing in the construction domain, the shortage of training data sets prevents such models from achieving their maximum potential. To address this issue, we investigate the potential of using synthetic data for vision model development. Specifically, we synthesize construction images and train a DNN model only with the synthetic data. We then evaluate the performance of the synthetic data-trained model on a worker detection task, and the results demonstrate the great potential of synthetic images: 97.3% of mean average precision. Given the benefits of synthetic data-it is possible to automatically create an unlimited number of images without manual labeling-this finding is promising. Moreover, this approach can be readily applied to other computer vision tasks, without requiring the manual labeling. This finding will enable the creation of more accurate and scalable DNN models for construction applications.
The need to increase the number of science, technology, engineering, and mathematics (STEM) students has been stressed, yet the growing number of students still fails the national demand for the STEM workforce. This mismatch between supply and demand is also found in the construction field. Storytelling can be a potential solution to this problem, as stories are believable, rememberable, and entertaining. Storytelling can allow students to experience the career planning and decision-making processes and embrace the possibility of themselves joining the construction field. This paper introduced the creation and development process of the immersive stories and assessed the influence of immersive storytelling on students’ attitudes towards the construction field based on gender groups. The results can enhance our understanding of immersive storytelling’s potential influence in motivating students and ultimately attracting them to the construction field. It was found that immersive storytelling had a positive influence on both male and female students. Moreover, it was observed that females’ aspirations to achieve a career in the construction field before and after the immersive stories were both significantly lower than males’.
The pandemic of COVID-19 has caused severe disruptions in urban lives. Understanding and quantifying these disruptions is important to inform the development of targeted and effective measures to control the pandemic and its impact. One way of achieving this object is to measure the urban mobility perturbation caused by the pandemic. In this study, we built mobility-based networks for seven major metropolitan statistical areas (MSAs) across the United States in the years of 2019 and 2020, respectively. We quantified the disruptions of urban mobility by computing and comparing a set of network-based metrics before and during the pandemic. The proposed approach is able to uncover the impact of COVID-19 in cities and provides new insights into the resilience of cities when facing large-scale disasters.
We propose a method to improve the understanding and visualization of the actual condition of a construction site by automatically developing an as-is BIM using site-appearance information from images and the as-planned BIM model. This is achieved by generating point clouds (PCs) from site images applying structure from motion (SfM). The corresponding elements between PCs and the 3D BIM were automatically determined using geometric and position assets to register PCs into the as-planned BIM accurately. Moreover, material condition classification can be done using information from the point cloud data. This way, the as-is BIM can be enriched with additional information such as the actual material conditions. The proposed method has been demonstrated using a construction environment where the as-is BIM was developed automatically from a set of 130 site images and the as-planned BIM. The as-is model has been used to identify deviations between as-planned and as-is conditions.
Industrialized construction is gaining popularity in the industry due to its advantages of reducing pollution, shortening construction time, and increasing safety. However, the design process of industrialized construction projects is different from traditional design approaches, thus should consider not only the architectural and performance requirements but also the manufacturing and assembling requirements. Specifically, during the pre-construction design process, the designers have to consider the constraints of material, shapes, or mechanical property of the components from the assembly and manufacture technicians. In order to reduce design alteration times and improve the design efficiency, it is necessary to strengthen the communication and information interoperability between the designers, manufacturers, and contractors. The use of an ontology, an explicit formalization of a conceptualization that provides an abstract schema consisting of formal definitions of concepts and their relationships, is often applied to achieve semantic interoperability among different systems. The main objective of this paper is to present an ontology to support the manufacturability and constructability checking of prefabricated components. The methodology used in this research for developing the ontology includes four main steps: (1) purpose and scope definition, (2) taxonomy building, (3) relation modeling, and (4) ontology coding. A case study using Revit and a prototype application validates the exchange of information. It is expected that this developed ontology can provide the foundation for the establishment of a comprehensive prefabricated component manufacturability and constructability checking system.
Laser scanning and virtual reality (VR) are two advanced trending technologies for better project monitoring and visualization in modern construction projects. Their uses, however, were mostly separated, and the connection between them has not been much studied in previous literature. This paper investigated the workflow of using VR to view laser scans through two different approaches, namely by point clouds and 360 degrees photos, and then compared the benefits between the two approaches. The study utilized a Native American earth lodge as an example of the built environment for laser scanning and applied two different types of VR devices. In the point cloud approach, the scans were processed by the scanner software and converted directly into VR mode for convenient viewing with a PC-based VR headset. In the 360 degrees photo approach, colored photos taken by the scanner at various angles were first stitched together by the scanner software to create 360 degrees photos, and these photos were then imported into a VR scene in Unity and compiled into an executable app for photo-realistic viewing with a standalone VR headset. The benefits of each approach were discussed, including the software and hardware requirements, efforts needed, and viewing experience. This paper provides valuable insights on the workflow of using laser scans with VR based on a real-life example.
Simulation of indoor airflow in heating, ventilation, and air conditioning (HVAC) system in construction design is critical in order to control spread of volatile organic compounds (VOC) or germs and viruses in the case of a pandemic, as we witnessed during 2019-2021. This paper presents the findings from comparing two types of fluid simulations: grid-based and particlebased methods. The former is widely used for scientific computation due to its precision; however, it is time consuming and requires designers to do pretreatment of the building model for airflow simulations in construction design using grid-based methods. The particle-based method is used in visual effects, games, and other applications requiring real-time simulation. This paper presents a review of the literature on different methods of fluid simulation like finite volume method (FVM), smoothed-particle hydrodynamics (SPH), position-based dynamics (PBD), and provides a comparison of heat transfer between particle-based and grid-based methods in indoor airflow simulations with HVAC system.
Lighting energy consumption has a significant share of energy consumed in US commercial buildings. An accurate prediction of lighting energy consumption offers energy-saving or energy-management potentials, especially in office buildings. Although office buildings’ energy performance has been studied extensively in the literature, mainly focusing on HVAC energy consumption, studies focusing on lighting energy consumption in US office buildings are limited. This study aims to develop a data-driven prediction model to estimate the office buildings’ lighting energy performance accurately. This study used Commercial Building Energy Consumption Survey (CBECS) 2012 data. First, the predictors regarding lighting energy consumption in office buildings are analyzed, and the most relevant predictors to predict lighting energy consumption are selected using LASSO regression method. Second, an artificial neural network (ANN) model is trained, using the selected predictors, to predict lighting energy consumption. This study identified 12 main variables as the most influential predictors in lighting energy prediction, including square footage of the building, percentage of light during building operation hours, number of employees working in the building, building owner type, whether there is daylight harvesting, lighting percentage during off-hours, number of businesses in the building, whether there is lighted parking area, building America climate region, percentage of occupancy, whether the building is open during the weekend, and census division. In addition, the results showed that the developed ANN model is able to predict lighting energy consumption in office buildings with an R-square value of 76% and an RMSE value of 223.17.
Although there have been many attempts to increase student engagement and interaction in class through virtual reality and augmented reality (VR/AR), structural engineering has remained as one of the disciplines that lacks interactive learning. There was an earlier attempt by the authors to improve student learning in structural analysis and dynamics through a student-centric cyber-physical system (SCPS) that reacts to student movement and outcomes structural behavior. This system used mobile devices as sensors and sent data to be processed by a server. As a result, it suffered from the inherent limitation of Transmission Control Protocol/Internet Protocol (TCP/IP)—lags and instability. Therefore, this paper presents the development of a new SCPS tool with a new system architecture that is much more stable and provide better user experience. This SCPS captures and uses student movement (i.e., walking, turning, and jumping) as point loads on a virtual bridge. As students move around, structural behavior (i.e., deflection) is simulated and visualized in real time.
