Study And Development On Intellegent Faults Detection And Diagnosis System Applied In Centrifugal Chillers

Centrifugal chillers are being used widely in the field of HVAC&R. In the past decades, there were some promising new technologies and methodologies which have been applied in this field. Because of the complexity of a centrifugal chiller and high cost of repair, it is necessary to develop a system of intelligent faults detection and diagnosis applied in centrifugal chillers. In this thesis, the current application of these new technologies in HVAC&R filed was reviewed comprehensively in the first part.A centrifugal chiller is mainly fabricated by centrifugal compressor, condenser, evaporator, throttling device, cooling capacity control device and lubricating oil system etc. In this thesis, the structural characteristics of these specific components were described first, and then the analysis of the characteristics of the various faults of these components was given based on this. Production rules (PRs) is the most mature and widely-used methodology for the development of intelligence/expert system, production rules was employed as a main reasoning methodology in this study. Referencing other scholars'research and our own experiment study, we gave some production rules applied in faults detection and diagnosis (FDD) of centrifugal chillers.As for intelligent FDD technologies and methodologies, each single one is flawed and ineffective during actual reasoning process. So on the base of analyzing advantages and disadvantages of various methods, we integrated the production rules, fuzzy logic and neural network into our reasoning system and production rules is the most primary one of three methods. For conventional production rules, the capacity of knowledge expression is limited, resulting in lower reasoning accuracy. Therefore, the reasoning methodology based on traditional fuzzy production rules was improved: in terms of causes of fuzziness of different rules, these rules were classified into several forms, and different forms correspond to different reasoning calculation methods. Through this improvement, the reasoning capacity of fuzzy production rules was enhanced significantly.For fuzzy production rules, there are many parameters which determine the reasoning accuracy of inference system. At present, these parameters are given by filed experts manually, and this obviously would increase uncertainty and inaccuracy of reasoning. To solve this problem, we mapped the fuzzy production rules used in our intelligent FDD system into a fuzzy neural network, and according to the characteristic of the network structure a new training model were proposed. And the actual fault diagnosis cases resulting from the inference of intelligent FDD system were considered training samples to train the fuzzy network, in another word, to upgrade and optimize the parameters of fuzzy production rules.Based on theoretical study, an intelligent FDD software applied in centrifugal chillers using Labview and C as development tools was developed in the study. The software was constructed by several modules as follow: parameters set module, files reading module, graphic display module, intelligent reasoning module, parameters optimization and upgrading module and other auxiliary modules. With this software, operation data files of a chiller could be read in, and through the above-mentioned function modules the operational status of this chiller could be obtained automatically.It has been taking our long period to observe on the running status of a centrifugal chiller, and had done some typical experiments under malfunction condition, obtained a large number of relevant experimental data. We inputted the experimental data into the new-developed intelligent FDD system to verify it is validity, and the results show that this system can depict operational condition of a chiller intuitively, and find causes of a chiller's malfunction precisely.