Research On Fault Detection And Diagnosis Methods For Chillers

The building sector is widely recognized as a major consumer of both energy and resources.Heating,ventilation,air-conditioning(HVAC)systems contribute to a significant portion of energy consumption in buildings,and chiller plants are the major consumers of the energy used in HVAC systems.Chiller faults may cause significant energy waste,shorten the life of equipment and reduce indoor environment comfort.Applying the fault detection and diagnosis(FDD)techniques to chillers is beneficial to keep the energy efficiency of HVAC systems high,to achieve the energy conservation of HVAC systems,and to maintain a comfortable indoor environment.This thesis makes research around the following three main obstacles for field implementation of FDD methods for chillers: i)lack of adequate sensors installed on the field chillers;ii)inadequate fault data;iii)the impact of false alarm rate(FAR)of an FDD method on a user.It mainly aims to develop new chiller FDD methods with the following desired attributes for field implementation: i)selecting these commonly available features in the field and supplementing these features obtained through a few low-cost sensors to indicate faults;ii)high detection and diagnosis accuracies;iii)the ability to identify new types of faults and to update FDD fault library;iv)the ability to adjust its FAR;v)the ability to address non-linear variables and uncertainties,and to merge multi-source non-sensor information.From the perspective of field applications,a feature selection(FS)method is proposed for selecting the commonly available features in the field and for supplementing a few features with low cost of measurement to indicate the typical chiller faults,so as to save the initial sensor cost meanwhile making the fault diagnosis(FD)method based on Bayesian network(BN)obtain good performance.The technological paths of FS are as follows: First,a sample investigation is conducted about the field chiller onboard sensors.The existing features that can be retained are nominated through the following criteria: high existent frequency of sensors installed on the field chillers,high sensitivity to faults and small amount of calculation.Then these features are evaluated by using the FD method based on BN merged a distance rejection technique(DR-BN)to remove redundant features.Second,when the expected performance cannot be obtained by only using these specifically retained existing features(RE-features),additional features need be supplemented.Supplemental features(S-features)are nominated through the following criteria: low cost of measurement and high sensitivity to faults.Then these features are evaluated together with RE-features by using the same method to determine the specific S-features.A novel chiller FDD method based on BN merged a distance rejection technique and multi-source non-sensor information(DR-MI-BN)is proposed.In order to identify new types of faults and to update the FDD fault libraries,the chiller FDD problem is transformed into a single-class classification problem by merging a distance rejection(DR)technique into the BN.Furthermore,the DR can be tuned to obtain an adjustable FAR.Based on the open network topology,multi-source non-sensor information(MI)is merged into the BN to increase the diagnostic correct rates(CRs)of known(or existing)faults and the identification CRs of new types of faults.The grey theory applied to the chiller FDD,a novel chiller FDD method based on grey similitude relation analysis(GSRA)is proposed to make the best of its ability to perform well on the small-sample pattern recognition problems and its advantages without requiring the specific distribution about the data.This method builds the comprehensive reference fault patterns by calculating the weighted averages of reference fault patterns at multiple severity levels,and introduces the thresholds of grey similitude relation degree(GSRD)to detect fault and identify the suspected classes of the unidentified fault pattern.In order to obtain better FDD performance,optimization algorithm is adopted to optimize the weighted indexes and the GSRD thresholds.The experimental data from ASHRAE RP-1043 are used to select features and validate the proposed FS method,and also used to evaluate the performance of the proposed FDD methods.The FDD results of the proposed methods are also compared with the previous similar studies.The results show that i)the proposed FS method is effective for chiller FD;ii)the FAR of the method based on DR-MI-BN can be tuned,and merging the MI significantly improves the diagnostic CRs of known faults(from 76.4% to 100% at most)and the identification CRs of new types of faults(from 5% to 96.7% at most),and the comparison shows that the FDD performance of this method is better than that of the previous similar studies;iii)only a small number of training data used,the FAR of the method based on GSRA is only 7.5%,and the diagnostic CRs of typical chiller faults are respectively 92.5%-100%,and compared with the conventional technological approaches,this method has better FDD performance.