For example, the combined fault can be detected by using the ratio of heaterpower residual and supply temperature sensor residual. For instance in the case of single faultdue to ST sensor, the heater power consumption residual is 0.638 kW and supply temperatureresidual is 3.298C. In the case of combined fault in OD and ST sensor, heater powerconsumption is 0.983 kW and supply temperature residual is 3.578C.The residual ratios of heater power consumption to the output value of supply temperature inthe cases of single fault and combined fault were computed as shown below.Table VII. Normalized patterns for AHU single fault diagnosis used in ANN training.Neural network inputHeat VAV ST SF SFR Neural network output Fault diagnosis0 0 00 0100000 Normal1 0 00 0010000 Outdoor damper fault0 1 00 0001000 Indoor temperature sensor fault0 1 00 1000100 Supply flowrate sensor fault1 0 10 0000010 Supply temperature sensor fault0 0 01 0000001 Supply fan faultTable VIII. Normalized patterns for AHU multi fault diagnosis used in ANN training.Neural network inputHeat VAV ST SF SFR Neural network output Fault diagnosis0 0 0 0 0 100000000 Normal1 0 0 0 0 010000000 Outdoor damper fault0 1 0 0 0 001000000 Indoor temperature sensor fault0 1 0 0 1 000100000 Supply flowrate sensor fault1 0 1 0 0 000010000 Supply temperature sensor fault0 0 0 1 0 000001000 Supply fan fault1 0 1 0 0 000000100 ST+OD fault1 0 1 0 1 000000010 ST+SFR fault1 0 1 1 0 000000001 ST+SF fault The residual ratios are respectively, 5.75 and 3.63 in each of the two cases. When a combinedfault and a single fault have the same pattern, we propose comparison of residual ratios(Table IX) in order to conclusively identify the fault. From the residual ratios computed fromEquations (7a) and (7b) we note thatRRjSTfault¼ RSTRpower    ST> 5:0 ð8aÞRRjSTþODfault¼ RSTRpower    STþOD55:0 ð8bÞWhen fault pattern is the same, and if RRST which is the residual ratio of RST and Rpower isgreater than 5.0, it can be classified as a single fault. If RRST+OD is less than 5.0, it can beidentified as a combined fault. In other words, the magnitude of the residual ratio in multi-faultalways remains smaller than single-fault residual ratio. This pattern was noted consistently atthree different fault rates of 10, 15 and 20% as shown in Table IX.
Thus, these results show thatpattern diagnosis using residual ratios is a useful technique in uniquely identifying single andmultiple faults in HVAC systems.6. CONCLUSIONSA pattern diagnosis method for detecting single fault or combined fault occurring in HVACsystems is developed. Based on experimental results and analysis of the residual data from bothsingle and multiple fault modes the following conclusions are offered.(1) Multiple fault pattern diagnosis analysis is necessary for more accurate diagnosis offaults in HVAC systems.(2) It was found that the multiple fault patterns fall in three distinct categories. First, thereexist a group of faults, which show similar residuals in single as well as multi fault modes.In the second category of the faults the combined fault residuals were found to be sum ofthe inpidual fault residuals. In the third category of faults, the single fault outputs hadno influence on the combined faults.(3) It is shown that residual ratio is a good indicator of isolating and detecting faults, whichshow similar residual patterns in single and multi-faults in HVAC systems. NOMENCLATUREF =flow sensorHeat =heater power (kW)IT =room temperature (8C)OD =outdoor damper opening (%)OT =outdoor air temperature (8C)R =residualRR =residual ratioSF =supply fan temperature (8C)SFR =supply airflow rate (CMH)ST =supply air temperature (8C)VAVD =VAV damper opening (%)SubscriptF =faultMAX =maximumN =no faultREFERENCESAnderson D, Grave L, Reinert W, Kreider JF, Dow J, Wubbrna H. 1998. A quasi-real-time expert system for HVACSystems. ASHRAE Transactions 95:954–960.Chen B, Braun JE. 2000. Simple fault packaged air conditioners. Proceedings of the Purdue University. West Lafayette,U.S.A., July 25–28, 321–328.Chen B, Braun JE. 2001. Simple rule based methods for fault detection and diagnosis as applied to packaged airconditioners. ASHRAE Winter Meeting, Atlanta, GA.House JM, Lee WY, Shin DR. 1999. Classification techniques for fault detection and diagnosis of an air-handling unit.ASHRAE Transactions 105:1087–1097.Katipamula S, Pratt RG, Shassin DP, Taylor ZT, Gouwri K, Brambley MR. 1999. Automated fault detection anddiagnosis for outdoor-air ventilation systems and economizer: methodology and results from field tests. ASHRAETransactions 105:555–567.Lee WY, House JM, Park C, Kelly JE. 1996. Fault diagnosis of an air-handling system using artificial neural network.ASHRAE Transactions 102:540–549.Liu ST, Kelly GE. 1998. Rule-based diagnostic method for HVAC fault detection. Proceedings of Simulation Conference89, Vancouver.Norford L, Little R. 1993. Fault detection and load monitoring in ventilation systems. A 暖通空调系统故障检测英文文献和中文翻译(4):http://www.youerw.com/fanyi/lunwen_33211.html