Rogers, Austin Paul (2019-08). FAULT DETECTION AND DIAGNOSIS METHODS FOR RESIDENTIAL AIR CONDITIONING SYSTEMS USING CLOUD-BASED DATA. Doctoral Dissertation. Thesis uri icon

abstract

  • Buildings account for nearly 40% of total energy consumption and nearly 75% of electrical energy consumption in the United States, and a significant portion of this energy consumption is due to the heating and cooling systems. Both commercial and residential heating, ventilation, and air conditioning (HVAC) systems are prone to faults that degrade performance and increase energy consumption. Furthermore, these systems are robust to faults in that they will operate with faults present for an extended period of time and will often continue to maintain a comfortable indoor environment. While considerable work has been devoted to developing fault detection and diagnosis (FDD) strategies for large and small commercial systems, relatively little has been done specifically for residential systems. This research presents novel FDD methods developed specifically for residential air conditioning systems. By using a novel set of virtual sensing methods, the proposed methodology eliminates the need for installing sensors on the outdoor unit. This is a significant advantage for residential 'split' air conditioning systems because installing sensors on both the indoor and outdoor units increases the complexity and cost of the data acquisition system. In addition to the proposed set of virtual sensors, this research provides solutions to two other problems that arise when implementing FDD methods on field-operating systems. (1) While most FDD methods use static models and rely on steady state analysis, field-operating systems often will not achieve steady state operation. This research provides a method for predicting the equilibrium operating point for many air conditioning parameters while the system is still in a transient response. This enables the equilibrium point to be determined before steady state operation has been achieved, and thus a static analysis may be performed without the system reaching steady state. (2) Existing change-point detection methods that could be used for detecting faults are impractical to implement on a large scale because they may require a priori knowledge, extensive tuning, or high computational loads. This research proposes a change-point detection algorithm for the purpose of fault detection which requires minimal assumptions, tuning, and computation. This change-point detection algorithm is suitable for deployment across many different systems simultaneously. In addition to the solutions outlined above for performing FDD using installed sensors, this research also proposes methods for performing fault detection and diagnose using only thermostat data. While a full strategy for thermostat data is not presented, crucial preprocessing methods that more complete methods will be built on are presented in detail. Nearly all of the data analyzed for each method described in this study uses event-based data uploaded in real-time to a cloud-based database and then queried and analyzed to perform FDD.

publication date

  • August 2019