Adaptive HVAC design : a CFD based machine learning model to increase thermal comfort in a large indoor office space

• Main Author: Thangaveloo, Kashveen
• Other Authors: Ng Bing Feng
• Format: Theses and Dissertations
• Language: English
• Published: 2019
• Subjects: DRNTU::Engineering::Mechanical engineering
• Online Access: http://hdl.handle.net/10356/77133
• Institution: Nanyang Technological University

Thermal comfort in a large indoor office space with open office plan are highly influenced by the HVAC(Heating, Ventilation and Cooling) design. Using Fanger’s method that is also part of ISO Standard 7730, thermal comfort scores can be computed using physiological and environmental factors tracked within the office space. Based on that method, experimental and numerical study are done to estimate the thermal comfort at different locations within the office space of an engineering company in Senai, Malaysia. The experimental results and numerical study revealed that the office plan is on average at a thermally neutral state. However, the experimental results also revealed that there are local hotspots and coldspots that might need change in diffusor settings or redesigned HVAC layout. A survey is conducted within the office space to estimate thermal comfort directly from employees in the working environment. The survey further confirmed the hotspots and coldspots determined from the experimental results. Change in local diffusor settings or HVAC design to combat this could cause unpredictable changes around the vicinity and might move the hotspots and coldspots instead of eliminating it. Hence, regression algorithms that are based on machine learning are employed to make a prediction model for varying HVAC design parameters. Many CFD simulations are done by varying HVAC parameters like air velocity and direction. These cases are used to train the prediction model which helps in the design of intelligent HVAC systems that can predict the change in thermal comfort if one of the HVAC parameters changes. The accuracy of the prediction model provided good comparison when compared with the data from the non - trained data set. Limitations of the model and further work needed to improve the robustness of the model have also been discussed in detail.