Digital Twin technology toward more sustainable buildings

Abstract

The integration of digital technologies in the form of sensor networks and automation systems has a significant impact on the Architecture, Engineering, and Construction- Facility Management (AEC-FM) industry in terms of data monitoring and manage- ment. By combining the real and digital worlds, developments in digital technologies like Digital Twin provide a high-level depiction of buildings and their assets. This thesis covers a wide range of topics, including building information management and the interaction of building systems, where the Digital Twin technology becomes a solution to organizing data and generating new study lines on data interchange and BIM (Building Information Modeling)-FM interoperability. In order to contribute to digitalization and automation solutions for building management, the initial step in this thesis was to prepare a review of research on study patterns, gaps, and trends in the AEC-FM industry. After a complete bibliometric search of Google Scholar, Web of Science, and Scopus and following selection criteria, 77 academic publica- tions about the Digital Twin application in the AEC-FM industry were labeled and clustered accordingly. The results demonstrate that information standardization, predictive maintenance, users’ comfort, and optimizations are the marked fields where the Digital Twin in the AEC-FM industry should be implemented to reach Zero Emission Buildings (ZEB). This work suggests several novel frameworks for Digital Twin for building management as a place to start a further investigation. In order to get around the shortcomings of the facility maintenance management (FMM) systems now used in buildings, the next stage in this research was to develop a Digital Twin predictive maintenance framework for the Air Handling Unit (AHU). BIM, IoT (internet of things), and semantic technologies are used as a part of the Digital Twin technology, which is still in its infancy in the facility management sec- tor, to improve building facility maintenance strategies. Three modules are used to develop a predictive maintenance framework: maintenance planning, condition pre- diction using machine learning, and operating fault detection in AHU based on the APAR (Air Handling Unit Performance Assessment Rules) approach. Inspection data and past maintenance records may also be acquired through the FM system. In order to confirm that the strategy was workable, the suggested framework was also put to the test in a real-world case study using data from August 2019 to Octo- ber 2021 for an educational building in Norway (I4Helse). The integration of BIM with predictive maintenance resulted in a powerful solution for decision-making in facility management. A RESTful API and plug-in extension were developed for Au- todesk Revit using C sharp, seamlessly connecting the BIM model with sensor data and enhancing understanding and analysis. A RESTful API served as an additional layer, enabling the extraction of data from individual devices in the building, grant- ing access to a wide range of diagnosed devices, maintenance records, and historical alarms and faults. The fully-featured plug-in empowered facility managers to ac- cess real-time sensor data through the RESTful API, update the BIM model, and save it in the relevant condition database. The COBie extension plug-in converted BIM data to COBie spreadsheets, while the mapping of COBie data attributes with the FM database was achieved using the Brick Schema. The implementation of the APAR method led to the identification of 25 faults in AHU systems. Machine learning algorithms, including ANN, SVM, decision trees, and based on data from BIM, BMS, CMMS, and sensor data, were compared to predict the faults, with ANN outperforming others in accuracy. Therefore, ANN was used to predict the future faults within one and 4.5 months ahead which could predict most of the faults in AHU. This integrated approach enabled facility managers to effectively monitor degradation, plan resources, save energy, and improve thermal comfort for occupants, creating a comprehensive solution for optimized facility management. The heating, ventilation, and air conditioning Digital Twin (HVACDT) system was presented as the third study area in this thesis to minimize energy consump- tion while enhancing thermal comfort. The framework is designed to make it easier for facility managers to comprehend how a building operates and improve HVAC system performance. The Digital Twin framework is based on BIM. It includes a newly developed plug-in for Matlab programming that allows for real-time sen- sor data, thermal comfort, and optimization processes. Data were gathered from a Norwegian non-residential building (I4Helse) between August 2019 and October 2021 and used to test the framework to see if it is viable. The HVAC system is then enhanced using a multiobjective genetic algorithm (MOGA) and an artificial neural network (ANN) in a Simulink model. The HVAC system includes, among other things, air distributors, hydronic cooling and heating units, pressure regula- tors, valves, air gates, and fans. In this context, numerous features are considered choice variables, including temperature, pressure, ventilation, cooling, heating op- eration management, and other elements. In this study, Post-occupancy evaluation (POE) was utilized to create a user satisfaction survey focusing on thermal com- fort. The survey included both physical and non-physical comfort aspects, and participants were asked to provide feedback on various comfort parameters related to their workplace. Additionally, respondents were requested to rate their satisfac- tion with the thermal characteristics of common areas in the building. To organize the spatial data, room information from the building’s incomplete BIM model was supplemented with laser scanner data. The point cloud model was processed in Autodesk ReCap Pro and Autodesk Revit to create an accurate representation for collecting occupants’ feedback. The predicted mean vote (PPD) and HVAC energy consumption are calculated to derive objective functions. The decision factors and objective function of ANN were hence highly connected. The accuracies of wavelet neural network (WNN), random forest (RF), support vector machine (SVM), and artificial neural network (ANN) forecasting models were compared using R2 and Root Mean Square Error (RMSE) metrics. The ANN model exhibited the highest R2 value (0.93) and the lowest RMSE value (0.027), surpassing the performance of the WNN, SVM, and RF models. These results indicate that the ANN model achieved the best prediction fitting outcomes and had the lowest forecast fitting error, highlighting its superiority in accuracy compared to the other models. Addi- tionally, MOGA offers a variety of design elements that may be utilized to find the greatest thermal comfort and energy-saving options. The Pareto optimum solution for minimizing energy consumption and Predicted Percentage of Dissatisfied (PPD) was presented, with an optimization process that took approximately 7.055 hours. It revealed the trade-off between the two objective functions. Reducing energy con- sumption from 62.8 kW to 46.4 kW resulted in an increase in PPD from 6.2% to 27% during winter while reducing energy consumption from 59 kW to 42.9 kW led to an increase in PPD to 22.4% during summer. The minimum PPD values were 6.2% for winter, indicating maximum thermal comfort. However, this came with the highest energy consumption of 62.8 kW in winter and 59 kW in summer. On the other hand, the lowest energy consumption was 46.4 kW in winter and 42.9 kW in summer, but with high PPD (27% in winter and 22.4% in summer). The choice of the best so- lution depends on whether energy, thermal comfort, or both were considered the main priority. The findings indicate that the average cooling energy savings for four summer days are around 13.2%. For the three summer months (June, July, and Au- gust), they are 10.8%, maintaining the PPD below 10%. Regarding data handling, the HVACDT framework exhibits a higher level of automation than conventional methods. Another plug-in was developed to stream the optimization results back to the BIM model. In order to explore how building components affect energy use and determine the best design, this research then proposed a method that integrates BIM, machine learning, and the non-dominated sorting genetic algorithm-II (NSGA II). A plug-in is being created to receive sensor data and export the required data from BIM to MSSQL and Excel. To run an energy consumption simulation, the BIM model of Tvedestrand VGS was loaded into IDA Indoor Climate and Energy (IDA ICE). The model in IDAC ICE was validated using the sensor data. To explore the poten- tial solutions, we initially generated 1,236,912 combinations of decision variables. However, by employing pairwise testing, we effectively reduced the combinations to 8000, covering all possible solutions more efficiently. Each combination was used as input for IDA ICE, running one simulation to obtain the annual energy consump- tion. Completing all simulations required approximately 16 days. The resulting database, consisting of the 8000 combinations and their respective energy consump- tion values, was then utilized as input for machine learning algorithms through vi- sual programming (Dynamo). The optimization process took a wide range of input variables into account. The most significant characteristics in the literature guided the selection of the initial set of variables related to the building envelope, such as U-values. Additionally, variables like minimum air supply, window-wall ratio, so- lar heat gain coefficient (SHGC), load (lighting), activation of shading, reflectance, ventilation, shading factor, air infiltration, supply air temperature setpoints in AHU, supply water temperature setpoints from the central heating system, and heat exchanger efficiency in AHU were considered. The optimization of the latter variables, in combination, was absent from the literature, and no research inves- tigated the combined control of these two types of variables for the optimization process, i.e., HVAC with building envelope variables. Although there were over 40 variables that could be included based on the literature, the ANOVA-SVM method was employed to identify and prioritize the most important variables for considera- tion in the optimization process. Eleven machine learning algorithms are utilized to analyze the 8000 simulations and create a prediction model between building charac- teristics and energy use including Linear Regression (LR), ANN with one layer and 10 neurons, ANN with one layer and 100 neurons, Support Vector Machine (SVM), Gaussian Process Regression (GPR), Deep Neural Network (DNN), Random Forest (RF), Extreme Gradient Boosting (XGB), Artificial Neural Network-Support Vec- tor Machine (ANN-SVM), Least Square Support Vector Machine (LSSVM), Group Method of Data Handling (GMDH), and Group Least Square Support Vector Ma- chine (GLSSVM). The best algorithm was Group Least Square Support Vector Ma- chine (GLSSVM), which was eventually used in NSGA II as the fitness function for calculating building energy consumption. According to the findings, the results indicate that the energy consumption and PPD of the Pareto optimum solutions are frequently lower, with values below 50 kWh/m².year and 9%, respectively, compared to the original design solution’s energy consumption of 61.17 kWh/m² and PPD of 18.5%. This suggests that the NSGA-II solutions have the capability to effectively reduce the building’s energy consumption while simultaneously improving thermal comfort. The results show that building energy consumption and thermal comfort may be successfully improved by the GLSSVM-NSGA II hybrid technique, which reduces energy consumption by 37.5% and increases thermal comfort by 33.5%, re- spectively. Brick, BOT, and SSN, which are based on COBie and IFC data were used for data integration in this part of the thesis. In addition, the results show that the energy-efficient design of the building envelope should prioritize the U-value of exte- rior walls, followed by roofs, windows, and the window-to-wall ratio. By utilizing the innovative GLSSVM-NSGAII multi-objective technique, design modifications can be implemented to enhance building energy consumption and thermal comfort perfor- mance even before construction. This approach aids in selecting suitable building materials and designs. Additionally, factors such as shading, solar heat gain coef- ficient (SHGC), reflectance, and activation play a crucial role and are determined based on solar radiation and outer window air infiltration. Other important input parameters for achieving optimal solutions include envelope settings and efficient heat exchangers in the air handling unit (AHU). Subsequently, adjustments can be made to the ventilation supply air temperature and flow rate in the AHU, as well as the supply water temperature from the central heating plant to the local radiators. The next stage was to assess the application of Digital Twin for fault detection in buildings, considering various building systems and aiming to enhance occupants’ comfort across various aspects. Firstly a new review study was implemented to discover this area which led to the following research. A probabilistic model based on Bayesian networks (BNs) was used to assess the performance of two buildings in this case, I4Helse and Tvedestrand VGS. The analysis of factors contributing to occupants’ comfort in a building involved three stages. Firstly, survey forms were de- veloped for a user satisfaction survey, capturing convenience factors such as thermal comfort, acoustic comfort, indoor air quality, visual comfort, and space adequacy. Occupants provided feedback on various aspects using a Likert scale and had the opportunity to provide additional comments. Secondly, a probabilistic model based on a Bayesian Network (BN) was developed, considering survey findings and im- portant factors for discomfort in buildings. The BN model incorporated building and environmental information, complemented by parameters added to the BIM model. Lastly, a plug-in and visual programming interface were used to connect the BIM model, occupants’ feedback, and the probabilistic model, enabling the FM team to interpret data through BIM visualization and causal analysis. Then the re- search work in this thesis provided a framework that follows a systematic approach to address comfort issues in buildings. Firstly, it checks for electrical issues in the HVAC system, and if none are found, it utilizes a Bayesian Network (BN) to iden- tify HVAC design issues related to thermal comfort. If there are design issues, it examines whether the HVAC system is inadequate to meet occupants’ thermal de- mands. Proper architectural design allows for automatic computation of the thermal load and retrieval of indoor unit capacity from the equipment database. In cases where undersized HVAC components cause discomfort, options include insulating the room’s façade or using interior units with larger cooling or heating capacities. If the indoor unit capacity exceeds the thermal load, the framework applies the APAR rules to identify failures in indoor or outdoor HVAC system equipment. The frame- work also addresses issues related to visual comfort, acoustic comfort, and spatial adequacy, considering factors such as window-to-wall ratio, room lighting, shade management, acoustic insulation materials, room cleanliness, adaptability, accessi- bility, and ergonomic furnishings. This comprehensive decision-making framework assists facility managers in detecting and addressing building faults effectively. To improve the predictive maintenance framework, the forecasting procedure considers fault detection findings from a Bayesian network over three years of data, generat- ing building faults and maintenance requests. The proposed predictive maintenance system supports adaptive model training and prediction, continuously adjusting pa- rameters based on updated sensor data and service logs. The study’s results indicate that acoustic discomfort is primarily attributed to inadequate internal wall insula- tion rather than ventilation or the absence of attenuators. Using BIM visualization, facility managers can simulate different scenarios by manipulating causal elements to assess occupants’ satisfaction likelihood. Isolating internal walls can enhance acoustic comfort, but if budget constraints exist, incorporating acoustic attenuators into the ventilation systems is a more practical option. The sensitivity analysis implemented in this work shows that occupancy density and HVAC design faults are identified as crucial factors influencing indoor air quality. Visual dissatisfac- tion with the building I4Helse is linked to its low window-to-wall ratio (WWR), while occupants of Tvedestrand School express greater satisfaction with light qual ity. Window-to-wall ratio plays a vital role in light quality, with an optimal range of 10 to 40 percent. The trained ANN model is chosen for HVAC system predic- tions, demonstrating the framework’s ability to anticipate future scenarios up to two months in advance, emphasizing the dynamic nature of maintenance schedules. This study was extended to explore the space adequacy problem and advance the predictive maintenance process by testing 9 machine learning algorithms. The sensi- tivity analysis categorizes potential reasons for space inadequacy into low, medium, and high potential based on their likelihood to cause discomfort and impact em- ployee comfort. Reasons with low potential have minimal impact and are easily addressed, such as a lack of personalization. Reasons with medium potential require more significant changes, like noise pollution or poor ergonomics. Reasons with high potentials, such as poor lighting or air quality, have a significant impact and require substantial modifications. By prioritizing these reasons, resources can be allocated effectively to improve employee comfort. BIM is valuable for assessing and addressing space adequacy as part of a Digital Twin. For predictive maintenance, the Extreme Gradient Boosting (XGB) algorithm outperforms others, while Ran- dom Forest is faster and easier to implement. The study introduces a method to determine HVAC’s remaining useful life, potentially extending it by at least 10% and resulting in cost savings. Poor air quality, lack of natural light, and uncomfortable temperature emerge as the most influential factors affecting occupant comfort. By employing the Digital Twin architecture, this study utilizes operational data, testing, and assessments to ensure the integrity of the system model. Decision- makers can rely on the information generated by the digital system to support their real-world project decisions. Furthermore, the Digital Twin can forecast upcoming changes in the physical system by allowing users to evaluate and simulate various scenarios and devise effective strategies. This framework has the potential to uncover new practical opportunities that can be implemented in both the physical system and its simulated counterparts, offering significant benefits and enhancing system performance in the real world.