Fuzzy Model-based Fault-tolerant Control of Air-conditioning Systems

The paper describes the development of a supervisory control scheme that adapts to the presence of degradation faults and minimises any resulting increase in energy consumption or deterioration in occupant comfort. Since there is a high degree of uncertainty associated with the results of any fault identi cation scheme in information-poor systems of this type, the supervisory control scheme uses fuzzy models to predict fuzzy measures of the overall performance and a fuzzy decision-maker to determine the most appropriate set-points. This fault-tolerant control scheme is being developed and evaluated using a detailed computer simulation of a multi-zone, variable-air-volume, air-conditioning system. The fuzzy models relate the performance of the airhandling unit and the chiller to the supply air and chilled water temperature set-points, and to fuzzy measures of the amount of air-side and water-side fouling. Results are presented that demonstrate the ability of the fuzzy models to predict the performance and show the e ect of both water-side and air-side fouling on the performance. Problems associated with training the fuzzy models are also discussed. INTRODUCTION There has been much recent interest in the development of fault detection and diagnosis techniques that are suitable for use in building environmental control systems [Hyvarinen and Sarki, 1996]. The detection of abrupt faults, such as a broken fan belt, necessitates immediate manual intervention to eliminate the cause of the fault. Other, so called degradation faults, such as air or waterside fouling, will result in a gradual deterioration in the performance of the air-conditioning system [Pape et al., 1991, Rossi and Braun, 1996]. In some cases the control strategy can be modi ed so as to adapt to the presence of a fault. Such fault tolerant control schemes [Patton, 1997] are capable of continually re-optimising the overall control performance as the size of the degradation fault increases. Estimating the size of degradation faults is particularly challenging in air-conditioning systems where sensor bias and estimation o sets are signi cant [Ngo and Dexter, 1999], and no training data are available from the actual plant that is typical of faulty operation. In practice fault diagnosis may have to be based on qualitative descriptions of the behaviour of the plant in the presence of faults [Dexter, 1995; Glass et al. 1995] and only imprecise estimates of the size of the fault will be generated. Supervisory controllers optimise the performance of air-conditioning systems by changing the values of the set-points for the local controllers. Some supervisory schemes use mathematical models to predict the performance and an optimiser to nd the best values of the set-points [Curtiss et al. 1994]. There are often signi cant uncertainties associated with the de nition of the control objective [Dexter and Trewhella, 1990], the estimation of the cooling demand [Henze et al., 1997] and the models used to predict the performance. In such cases, precise optimisation is unnecessary and suboptimal rule-based supervisory control schemes may be more appropriate [Keeney and Braun, 1996; Drees and Braun, 1996]. This paper describes the development of a fuzzy model-based fault tolerant supervisory control scheme. The basic scheme is rst described before the approach used to generate the fuzzy models is explained. Results are then presented that demonstrate the sensitivity of the performance to the presence of both degradation faults and setpoint changes. FUZZY MODEL-BASED SUPERVISORY CONTROL The basic scheme (see Figure 1) consists of a fuzzy objective function, a fuzzy decision-maker and one or more fuzzy models. The fuzzy models are used to predict the performance of the air-conditioning system for particular values of the set-points, operating conditions and sizes of the faults. Each model consists of a set of fuzzy IF-THEN rules. The number of fuzzy sets used to describe inputs and outputs of the models