Brad
Brad Schultz, M.AIRAH
Hao
Hao Huang

Brad Schultz, M.AIRAH

Buildings Alive

Hao Huang

Buildings Alive

Unsupervised learning for anomaly detection and diagnostics in commercial buildings 
 
Abnormal operation of HVAC systems can result in an increase in energy usage, poor indoor air quality, thermal discomfort, and higher maintenance costs. 

Automated fault detection and diagnostics (AFDD) tools help to systematically control these faults to achieve better occupant comfort and energy savings. In recent years, machine learning-based fault detection and diagnosis methods have gained prominence because of their effectiveness in handling large data sets and independency of user-defined rules. So far, the most adopted machine learning-based methods are supervised learning techniques.

Supervised methods need to be trained with an extensive database that covers both faulty and healthy operational conditions. This is not always possible. To address the problem, this presentation introduces an unsupervised anomaly detection approach for detecting anomalous equipment performance and to conduct root cause analysis. The results show that, when using a short historical dataset, the unsupervised algorithm can accurately detect and diagnose control issues and mechanical faults of individual components. This proves that unsupervised learning can be complementary to supervised learning for HVAC systems. 
 

About Schultz:

Schultz is a mechatronic engineer with over 20 years’ experience working in technology and energy efficiency domains. He has spent the last 10 years working in energy management/building automation/analytics with several controls and mechanical companies. He now leads the services delivery team at Buildings Alive, working with portfolios of high-profile facilities around the world to reduce their energy costs and carbon emissions.

(Linkedin)


About Huang:


Huang has a master’s degree in mechatronics engineering and a PhD degree in mechanical engineering. Since 2016, he has worked as a building systems engineer as well as a data scientist with Buildings Alive. His research interests include building modelling and control, HVAC system fault detection and diagnostics, predictive optimisation, and measurement and verification.
 

(Linkedin)