Fault detection and diagnosis for refrigerant leakage in a distributed refrigeration system for supermarket applications

This research is motivated by the global environmental problems that are caused by refrigerants. Supermarket stores are chosen to be the central focus because they are one of the biggest sectors of the refrigerant industry and their equipment leaks a great amount of refrigerants into the atmosphere.; This research develops a technique for automatic refrigerant leakage detection and diagnosis for a distributed refrigeration system with a liquid receiver. A model for testing these techniques in a simulation environment is developed. The first-principle simulation model is described and validation results are presented in this dissertation. The component-based model has a modular design that enables easy adaptation to different system configurations. The model has been used to study behaviors of the distributed refrigeration system and to test the leakage detection technique.; The refrigerant leakage detection and diagnosis technique developed in this research is based on probabilistic models and uses the belief network technique to compute the probability of leakage in the system. The technique can predict the distribution of the refrigerant leakage without speculating its probability density function. The basic concept of the technique is to compare the predicted liquid level or liquid volume fraction in the receiver with the measured one. This technique does not require to learn the fault mode of the system before being capable to distinguish a "leak" system from a "no leak" system. The decision is made based on the principle of minimum risks, which is to incur the least cost of action. Also, the technique includes the impacts of measurement uncertainties and system economics on the confidence of making decisions. A neural network model of system performance is proposed as a part of the technique to make the evaluations faster and more practical in the field implementation with small microcomputers.; A real supermarket store located at Longmont, Colorado is used as a prototype for developing the system model as well as testing and evaluating the detection technique. All data are retrieved from the data acquisition system of the store. Within the simulation environment, the developed refrigerant leakage detection and diagnostic technique is capable of detecting the leakage if the system loses more than 1.0% of the installation charge or about 5.2 kgs. The technique has the advantage that as the leakage becomes more severe, the classifier becomes more capable of distinguishing between a "leak" and a "no leak" system. The technique misclassified one data set from ten data sets when tested. From the analysis and testing, the developed refrigerant leakage detection and diagnostic tool for a distributed refrigeration system successfully distinguishes between a leak system and a no leak system.