Research On Fault Detection And Diagnosis For Variable-Air-Volume Air-Conditioning Systems Based On Hierachical Scheme

According to some statistics, Variable-Air-Volume (VAV) air-conditioning systems can save energy by 30% to 50% when compared with conventional Constant-Air-Volume (CAV) ones. However, VAV systems tend to have more faults due to the complexity of their control systems. If a fault fails to be detected, diagnosed and removed in time, it will bring about abnormal operation which subsequently increases energy consumption of the system, decreases comfortability of human beings and makes the air-conditioning equipment wear down. Currently, the faults of air-conditioning systems are diagnosed and removed mainly depending on feeling and experience of maintenance personnel. However, it is very difficult for the maintenance personnel to discover and remove the faults just in time. Therefore, research on automatic Fault Detection and diagnosis (FDD) of VAV systems is of great significance for energy conservation and stable operation of air-conditioning systems. Hierachical-scheme-based FDD approach is presented according to the characteristics of VAV systems and their control. The approach handles a VAV FDD system as 3 hierachical levels, i.e. system level, unit level and component level. In accord with different characteristics and requirements of the 3 levels, different modeling and fault diagnosis methods are used to improve correctness and reliability of FDD. Meanwhile, it will separate fault detection from diagnosis process and decentralize modeling complexity and FDD computation which benefits FDD implement and conformity with VAV control systems.For system level, faults are detected according to the total energy consumption of VAV systems. Or in other words, a fault is detected at the system level by checking the conformity of actual energy consumption with normal one which is predicted by a Fuzzy Neural Network (FNN). Based on the analyses of neural networks and fuzzy logics, two FNN structures, Mandani and Takagi, are established and implemented with 3 input membership functions considered for each structure. 6 FNN models are thus obtained.Under summer conditions, 3 groups of VAV operation data are obtained in a laboratory. 2 groups of the data is normal which are used for the training and testing of FNN models respectively; The third group is the data with 2 faults which is used for fault detection of 6 FNN models. The results of model train, test and fault detection show that Madani FNN structure with Gaussian or Bell membership functions has better model precision and fault detection capability.For unit level, control-loop-based fault diagnosis method is proposed. A VAV system is partitioned into several fault diagnosis units according to basic control loops. Faults are diagnosed by checking controlled variable residues and controlling variable residues of a specific control loop. Controlled variable residue is defined as the error between actual controlled variable and its setpoint. Controlling variable residue is defined as the error between actual controlling variable and its normal value. The normal value of the controlling variable is predicted with ARX or ARMAX model. Based on the analysis of ARX and ARMAX model and their identification methods, a Genetic Algorithm (GA) optimization ARMAX modeling method is proposed with the ARX model tackled as a special case of ARMAX.The basic ARMAX models of 3 main fault diagnosis units, i.e. units of supply temperature control, duct static pressure control and room temperature control, are established for the VAV system in the laboratory. The diagnosis unit of room temperature control loop is demonstrated. Using operation data and GA based ARMAX modeling method, fault diagnosis is carried out with 2 faults which are introduced into the diagnosis unit deliberately. The results show that control variable residue has stronger capability of fault diagnosis and checking both residues of controlled variable and controlling variable can improve soundness of VAV unit level fault detection and can diagnose partly with fault properties.For the faults of component level, fault pattern matching method is proposed to improve correctness and reliability of component fault diagnoses. The fault qualitative pattern and quantitative pattern can be obtained from the actual responses of the characteristic process variables when a fault occurs in a VAV system. They will be compared or matched respectively with all the fault qualitative and quantitative patterns which are stored in the fault qualitative and quantitative repositories. The result of matching is that of fault diagnosis. The quantitative patterns in the quantitative repository need to be obtained from dynamic simulation of fault models. Supply temperature control loop diagnosis unit is demonstrated as an example.Physical modeling is used for the components in the unit. 6 component fault simulation models are implemented based on the software of SIMULINK which form the basic fault simulation model repository of supply temperature control loop unit. 6 component faults are introduced purposely into the laboratory VAV system. The correspondent fault qualitative patterns are obtained which make the basic fault qualitative repository of supply temperature control loop unit. To demonstrate the process of pattern matching method and to verify its validity, fault diagnosis is carried out with the fault of higher measurement values of supply temperature transducer using pattern matching method. The fault is produced deliberately in the laboratory and concerning data is gathered for pattern matching fault diagnosis. The result confirms the validity and reliability of pattern matching fault diagnosis for component levels of VAV systems.