It is almost certain that all systems contain faults, and Heating, Ventilation, and Air Conditioning (HVAC) systems are no exception. About 41% of total energy consumption in the U.S. of year 2014 is used for heating and air conditioning, that is about 40 quadrillion (10^15) British thermal units (BTU)! In the past, fault detection and diagnosis (FDD) has been commissioned by man, which is costly and inefficient. If FDD can be done automatically and continuously, a great amount of energy could be saved and our buildings would become more sustainable.

People have been working on HVAC FDD for years, and will continue to make progress. However, buildings are complicated, and so are the HVAC systems which comes with them. Thanks to the fast growth of computers and developments of machine learning algorithms in recent years, we have more tools to work with.

In despite of many researches done on HVAC FDD, we have noticed that very few publications have focused on scalability and low cost. In order to address this challenge, we will propose an approach which focuses on control data. In this thesis, we will be relying on simulation data because of data access issues and the lower cost for experiments. A list of machine learning algorithms are introduced for data exploration and analysis, a control data focused model free approach is presented as well, and finally, FDD is carried out by implementing anomaly algorithms.