The research develops a digital twin of a building facility, leveraging Building Information Modeling’s (BIM) contextual data alongside two artificial intelligence techniques: machine learning and ontologies. This approach enhances detection, diagnostics, and semantic capture, providing a spatial enabled decision support system that aids building operators and boosts productivity.
HVAC systems account for 50% of building energy use, and faults can impact energy efficiency by up to 15%. Currently, pure data-driven machine learning techniques are favored for their simplicity in model development compared to physics-based white-box models for Fault detection and diagnostics. These black-box AI models, while effective in many scenarios, often overlook critical building semantics such as spatial information and are limited by the availability and location of sensors. This is because they are typically developed using sensor data stored in Building Automation Systems (BAS) or Building Management Systems (BMS), which lack the semantic information necessary for comprehensive analysis. As a result, important contextual information about the interaction between building spaces and HVAC assets is missing, potentially limiting the effectiveness of fault detection and diagnostics.
This research aims to bridge this gap by leveraging the rich building models commonly available in the construction industry through Building Information Modeling (BIM). BIM provides detailed contextual data, including spatial relationships, HVAC-related data, and building semantics, which can significantly enhance the performance of AI models. By integrating BIM with data from BMS and BAS, this research facilitates a more holistic approach to automated fault detection and diagnostics in HVAC systems. The integration allows for a deeper understanding of the building's operational context, leading to more accurate and reliable diagnostics resulting in efficient energy management and sustainable operations in building.
The developed solution addresses the challenge by creating a comprehensive digital twin of the facility, integrating both building and HVAC systems into a unified knowledge model. This semantic model enables users to detect and diagnose HVAC faults more effectively by considering the contextual information of building spaces and the interconnected nature of HVAC systems and their interaction with the building environment. The digital twin leverages Building Information Modeling (BIM) to represent the building's semantics and utilizes data from Building Automation Systems (BAS) and Building Management Systems (BMS) to reflect the current state of the building. Additionally, BIM data enriches the Automated Fault Detection and Diagnostics (AFDD) model developed using machine learning. The results are captured by the knowledge model introduced through the developed ontology, ensuring the digital twin evolves over time.
By combining these data sources, the digital twin can capture and track changes in the building's state, enabling continuous monitoring and analysis. This capability allows facility managers to make informed decisions based on up-to-date information, optimize building performance, and proactively address issues before they escalate. Additionally, the digital twin supports predictive maintenance, energy management, and other advanced applications, ultimately enhancing the efficiency and sustainability of the facility.
The proposed digital solution will be populated with Canadian building stock and HVAC assets, leading to a solution that is applicable to Canadian needs. This will allow for a reduction in carbon footprint while maintaining occupant comfort levels.
