Fault Detection And Performance Optimization Based Online Fault-Tolerant Control For Air Conditioning Systems

The faults generated in air conditioning equipments or sensors, willresult in increasing the energy consumption, decreasing the thermal comfort,deteriorating the indoor air quality, removing the regulating role from controlsystems, or even damaging some certain equipment. It has very important andpractical significance to detect these faults accurately in time, to diagnose thedamage severity to system performance, and thus to make the correctingdecision. In real systems, however, some problems are inevitable to face andto be solved. The operating or performance parameters should be simulated orpredicted accurately. The fault detection methods should find out thegenerated faults reliably. The effect of faults on system operating performanceshould be evaluated objectively and fairly. And the decision-making schemesshould be established appropriately.In view of the preceding discussions, this dissertation focuses on fourmain problems.Firstly, the operating performance prediction model is established bysupport vector regression (SVR) to provide the fault-free operating referenceand system performance parameters. The prediction model is validated byseveral different operating conditions. The prediction results show that theinput parameters to SVR prediction model must be selected appropriately toreflect the operating performance of the air conditioning system.Then, to balance the strengths and weakness of the statistics residualsand fractal dimension, a hybrid method has been proposed to detect fault. Themethod is validated by fixed bias and drifting bias faults of supply airtemperature. The fault detection results show that, for the lower bias faults,the statistics residual method presents smaller correct fault detection rates. Forthe supply air temperature fixed bias faults±0.2oC,+0.3oC, and drifting biasfaults±0.05oC/h,+0.1oC/h, the correct fault detection rates are lower than50%. The fractal dimension method can detect these relatively small faults, but needs a long period to collect the measured signals. In view of theirstrengths and drawbacks, two methods can be combined to a novel faultdetection tool. The statistics residual method can detect the relatively largebias faults, and the fractal dimension method can detect the small ones. Thestatistics residual method uses larger residual threshold, while the fractaldimension method must employ the suitable dimension threshold. Thecombined method can effectively detect the various degrees of faults.Furthermore, an improved ELECTRE (Elimination Et ChoiceTranslating Reality) model has been developed to evaluate air conditioningsystem performance. The absolute thresholds and dividing thresholds hasbeen introduced to construct the pseudo-criteria in the improved evaluationmodel. The minimum valid control efficiency has been defined as anevaluation criterion to represent the influence of fault on the performance ofthe control system. The evaluation model is validated by dozens of operatingstates under supply air temperature faults and fresh air flow rate faults,respectively. The ELECTRE evaluation model can be completely suitable tooutrank the system performance under different fault and different operatingconditions.Fourthly, an online fault-tolerant control strategy based on operatingperformance optimization is developed to correct the faulty measurementsonline. The strategy is validated by supply air temperature faults, fresh airflow rate faults, and fresh air temperature faults. By outranking the differentoperating alternatives, the online fault-tolerant control strategy can achievethe fault correction factor according to the optimal operating scheme. Theonline fault tolerant can thus be implemented to correct the faultymeasurements. For the supply air temperature fault+13%, the total faultcorrection factor keeps at0.84from the sixth correction. For the fresh air flowrate fault+39%, the total fault correction factor maintains at0.66from theseventh correction. For the fresh air temperature fault–34%, the faultcorrection responses are output by the first calculation of the outrankingsatisfaction scores.Generally, the proposed fault detection method combined the statisticalresiduals with the fractal dimension, can detect the various degrees of faults with more effectiveness. The improved ELECTRE evaluation model canmore outrank dozens of operating alternatives with more reasonability. Thedeveloped online fault tolerant control strategy can correct the sensor faultsgenerated in air conditioning systems, and can optimize the system operatingperformance. Further investigation should focus on how to identify severalfaults and how to complete the online fault tolerant control under the data lossor several faults conditions.