Research shows that 5% to 30% of a building's annual energy consumption is wasted due to malfunctioning systems. Automated Fault Detection and Diagnosis (AFDD) systems can help reduce this waste and improve occupant comfort. A widely used tool for AFDD is Diagnostic Bayesian Networks (DBNs), which are probabilistic graphical models linking faults with symptoms. The DBN structure, prior probabilities (the likelihood of faults), and conditional probabilities (the likelihood of faults given symptoms) often require expert empirical estimations. Accurate estimations of these parameters are crucial for effective DBNs, yet current studies often rely on limited expert opinions, typically the researchers themselves. This study interviewed seventeen experienced experts from building installation companies with diverse technical backgrounds to address this issue. They were asked to identify the most frequent faults of the Heat Recovery Wheel (HRW) in Air Handling Units (AHUs) and, afterward, estimate the occurrence of these faults over a period of five years. Based on the data, the prior probabilities of the faults were derived. The findings indicate that expert knowledge-based estimation can enhance the performance of DBNs by providing reliable prior probabilities for fault diagnosis. Future work should extend this methodology to other AHU components and incorporate structure and conditional probabilities to develop a comprehensive and robust DBN for Heating, Ventilation, and Air Conditioning (HVAC) fault diagnosis.