Faults indicate degradation or sudden failure of equipment in a system. Widely existing in heating, ventilating, and air conditioning (HVAC) systems, faults always lead to inefficient energy consumption, undesirable indoor air conditions, and even damage to the mechanical components. Continuous monitoring of the system and analysis of faults and their major effects are therefore crucial to identifying the faults at the early stage and making decisions for repair. This requires the method of fault detection and diagnosis (FDD) not only to be sensitive and reliable but also to cause minimal interruption to the system's operation at low cost. However, based on additional sensors for the specific information of each component or black-box modeling, current work of fault detection and diagnosis introduces too much interruption to the system's normal operation associated with sensor installation at unacceptable cost or requires a long time of parameter training. To solve these problems, this thesis first defines and makes major innovations to a change detection algorithm, the generalized likelihood ratio (GLR), to extract useful information from the system's total power data. Then in order to improve the quality of detection and simplify the training of the power models, appropriate multi-rate sampling and filtering techniques are designed for the change detector. From the detected variations in the total power, the performance at the system's level is examined and general problems associated with unstable control and on/off cycling can be identified. With the information that are basic to common HVAC systems, power functions are established for the major components, which help to obtain more reliable detection and more accurate estimation of the systems' energy consumption. In addition, a method for the development of expert rules based on semantic analysis is set up for fault diagnosis . Power models at both system and component levels developed in this thesis have been successfully applied to tests in real buildings and provide a systematic way for FDD in HVAC systems at low cost and with minimal interruption to systems' operation.