Comparison and assessment of information-based, physics-based and hybrid digital twins: an HVAC case study

PI: Burcu Akinci

Co-PI(s): Mario Bergés

University: Carnegie Mellon University

Today, heating, ventilation, and air conditioning (HVAC) systems account for the largest portion–35%–of the total building energy consumption, and approximately 20% of that energy is wasted by poorly maintained, degraded, and faulty components. Maintenance of complex systems such as HVAC still mostly centers around manual fault detection, isolation, and recovery strategies. At the same time, digital twins have emerged as a paradigm for virtual representation of physical components and systems across its life-cycle through utilization of real-time data and other sources to support learning, advanced reasoning, and dynamic calibration for improved decision-making. Digital twin technologies coupled with advanced analytics can enable not only reliable fault detections and diagnosis, but also root cause analyses of issues observed in HVAC systems. Depending on the industry, such as mechanical and aeronautical engineering versus architecture and civil engineering, the origins of digital twins are physics-based simulations or information/data-driven. Recently, the need for integrating these two distinct but complementary schools of thoughts has arisen, and consequently hybrid approaches for digital twins are being developed. The proposed PITA projects targets comparing and assessing the strengths and limitations of information-based, physics-based, and hybrid digital twin approaches within the context of HVAC system fault detection, diagnosis, and root cause analyses. CMU is teaming up with ANSYS, which is a world-leader in physics-based digital twin modeling technology, and will jointly compare and assess three different digital twin modeling paradigms in a real-life testbed at Porter Hall at Carnegie Mellon.