| dc.description.abstract | This paper focuses on the energy-saving maintenance of the chiller unit in the HVAC system. The chiller machine is a very power-consuming device. The efficiency improvement and parameter optimization of such large-scale equipment often require the experience and knowledge of professionals to achieve. This study attempts to analyze and adjust the uncontrollable characteristic parameters of the chiller machine based on the operating data of the chiller machine, and find out the failure factors of the chiller machine if it cannot be adjusted. It is hoped that this study will contribute to the maintenance of chiller machine equipment.
This study trains a nonlinear regression model using training data, with the variables being parameter data collected by sensors. In terms of model design, instead of directly combining the physical properties of the chiller, the study uses a black-box method to develop a data-driven research mode. As not all parameters are controllable, four are set as controllable parameters. The recommendations for adjusting parameters given by the system during analysis and diagnosis are not limited to controllable parameters and may include other non-controllable parameters. Therefore, this study further investigates this issue. When the system suggests adjusting uncontrollable parameters, a non-linear regression model for suggested and controllable parameters is established based on training data and the least square algorithm to create a non-linear relationship between controllable and suggested parameters. Genetic Algorithm is then used to find the adjustments for four controllable parameters according to the suggested parameter. Before model training, the training data is normalized to improve the accuracy of the model by min-max normalization.
This study also monitors equipment operating status to detect abnormal operations or equipment malfunction as early as possible, and expert rules are established to address this issue. Fault rules are defined based on past experiences, and t-tests are used to detect whether each specified parameter is abnormal. The selected parameters are then incorporated into the expert fault rules, and the reliability of each rule is calculated. If the reliability of a rule is lower than fifty percent, the target equipment is identified as needing maintenance or repair, and the type of equipment malfunction is determined. This study developed a user interface for users to set relevant analysis parameters and connect to the database. The analysis results are visualized, including the efficiency analysis of the chiller, the efficiency curve of the cooling tower, the fault reliability, and the current status of the specified parameters. The suggested values for controllable parameter adjustments and the equipment that needs improvement for fault analysis are also displayed. | en_US |