当前位置：文档库 › Cai_2014_Applied-Energy

Cai_2014_Applied-Energy

Multi-source information fusion based fault diagnosis of ground-source heat pump using Bayesian

network

Baoping Cai,Yonghong Liu ?,Qian Fan,Yunwei Zhang,Zengkai Liu,Shilin Yu,Renjie Ji

College of Mechanical and Electronic Engineering,China University of Petroleum,Qingdao,Shandong 266580,China

h i g h l i g h t s

A multi-source information fusion based fault diagnosis methodology is proposed. The diagnosis model is obtained by combining two proposed Bayesian networks. The proposed model can increase the fault diagnostic accuracy for single fault. The model can correct the wrong results for multiple-simultaneous faults.

a r t i c l e i n f o Article history:

Received 22July 2013

Received in revised form 7September 2013Accepted 17September 2013

Keywords:

Multi-source information fusion Ground-source heat pump Bayesian network Fault diagnosis

a b s t r a c t

In order to increase the diagnostic accuracy of ground-source heat pump (GSHP)system,especially for multiple-simultaneous faults,the paper proposes a multi-source information fusion based fault diagnosis methodology by using Bayesian network,due to the fact that it is considered to be one of the most useful models in the ?led of probabilistic knowledge representation and reasoning,and can deal with the uncer-tainty problem of fault diagnosis well.The Bayesian networks based on sensor data and observed infor-mation of human being are established,respectively.Each Bayesian network consists of two layers:fault layer and fault symptom layer.The Bayesian network structure is established according to the cause and effect sequence of faults and symptoms,and the parameters are studied by using Noisy-OR and Noisy-MAX model.The entire fault diagnosis model is established by combining the two proposed Bayesian net-works.Six fault diagnosis cases of GSHP system are studied,and the results show that the fault diagnosis model using evidences from only sensor data is accurate for single fault,while it is not accurate enough for multiple-simultaneous faults.By adding the observed information as evidences,the probability of fault present for single fault of ‘‘Refrigerant overcharge’’increases to 100%from 99.69%,and the probabilities of fault present for multiple-simultaneous faults of ‘‘Non-condensable gas’’and ‘‘Expansion valve port largen’’increases to almost 100%from 61.1%and 52.3%,respectively.In addition,the observed information can correct the wrong fault diagnostic results,such as ‘‘Evaporator fouling’’.Therefore,the multi-source information fusion based fault diagnosis model using Bayesian network can increase the fault diagnostic accuracy greatly.

1.Introduction

Ground-source heat pumps (GSHP)recovering heat from ground,have been widely utilized all over the world,which result in primary energy consumption reduction up to 60%compared to conventional heating systems,are of great signi?cance in energy saving and environment protection [1–4].Failure of the heat pump will cause reduction of energy ef?ciency and increment of environ-mental pollution.The relevant faults occurred in GSHP are divided into hard faults and soft faults.Generally,hard faults are easy to be detected and estimated,and soft faults are more dif?cult to be discovered [5].The common hard faults include (a)compressor hard shutdown;(b)valve choke completely;(c)fan stop running,and so on.And the common soft faults include:(a)refrigerant overcharge;(b)Refrigerant leakage;(c)evaporator fouling,and so on.Various fault diagnosis techniques are developed and used,to locate the soft faults exactly in heat pump systems.

Using fault diagnosis techniques,the degradation performance of heat pump systems can be detected early,and the exact reasons for degradation can be diagnosed [6].Xiao et al.[7]presented a fault diagnosis strategy based on a simple regression model and a set of generic rules for centrifugal chillers.Lee et al.[8]described a scheme for on-line fault detection and diagnosis at the subsystem level in an air-handling unit using general regression neural net-works,which consisted of process estimation,residual generation,

Corresponding author.Tel.:+86053286983303;fax:+86053286983300.

E-mail address:liuyhupc@https://www.wendangku.net/doc/c04592829.html, (Y.Liu).

and fault detection and diagnosis.Wang and Cui[9]developed an online strategy to detect,diagnose and validate sensor faults in centrifugal using principal-component analysis method.Mohanraj et al.[10,11]review the applications of arti?cial neural networks for refrigeration,air conditioning and heat pumps,and presented the suitability of arti?cial neural network to predict the perfor-mance of a direct expansion solar assisted heat pump,and the experiments were performed.Li and Braun[12]extended the decoupling-based fault detection and diagnosis method to heat pumps,and developed diagnostic features for leakage within check valves and reversing valves.Sun et al.[13]developed an online sensor fault detection and diagnosis strategy based on data fusion technology to detect faults in the building cooling load direct mea-surement.Naja?et al.[14]developed diagnostic algorithms for air handling units that can address such constraints more effectively, such as modeling limitations,measurement constraints,and the complexity of concurrent faults,by systematically employing ma-chine-learning techniques.Gang and Wang[15]developed arti?-cial neural network models for predicting the temperature of the water exiting the ground heat exchanger.A numerical simulation package of a Hybrid ground source heat pump system is adopted for training and testing the model.

Bayesian network(BN)is considered to be one of the most use-ful models in the?led of probabilistic knowledge representation and reasoning,which has been widely used in reliability evalua-tion and fault diagnosis.Cai et al.[16–18]studied the reliability of subsea blowout preventer control system,subsea blowout pre-venter operations and human factors on offshore blowouts by using Bayesian network or dynamic Bayesian https://www.wendangku.net/doc/c04592829.html,ngseth and Portinale[19]and Weber et al.[20]presented a bibliograph-ical review over the last decade on the application of Bayesian network to reliability,dependability,risk analysis and mainte-nance.Recently,the application of Bayesian network on fault diagnosis has been investigated deeply.Dey and Stori[21]devel-oped and presented a process monitoring and diagnosis approach based on a Bayesian belief network for incorporating multiple process metrics from multiple sensor sources in sequential machining operations to identify the root cause of process varia-tions and provide a probabilistic con?dence level of the diagnosis. Sahin et al.[22]presented a fault diagnosis system for airplane engines using Bayesian networks and distributed particle swarm optimization.Gonzalez et al.[23]developed a methodology for the real-time detection and quanti?cation of instrument gross er-ror.Zhu et al.[24]proposed an active and dynamic method of diagnosis of crop diseases to achieve rapid and precise diagnosis of crop diseases,using Bayesian networks to represent the rela-tionships among the symptoms and crop diseases.However,there are few application of Bayesian network in the heating,ventila-tion,and air conditioning system.Zhao et al.[25]proposed a gen-eric intelligent fault detection and diagnosis strategy to simulate the actual diagnostic thinking of chiller experts,and developed a three-layer diagnostic Bayesian network to diagnose chiller faults based on the Bayesian network theory.

In order to increase the diagnostic accuracy,especially for mul-tiple-simultaneous faults,this work presented a multi-source information fusion based fault diagnosis methodology for GSHP system by using Bayesian network method.The proposed Bayesian network consists of two layers:fault layer and fault symptom layer.The fault symptom layer includes not only sensor data but also observed information,which can increase the fault diagnostic accuracy greatly.The paper is structured as follows:Section2pre-sents the faults and fault symptoms of GSHP system.In Section3, the fault diagnosis methodology is developed using Bayesian net-work.In Section4,the fault diagnosis results using evidences from sensor data and observed information is researched.Section5 summarizes the paper.2.Faults and fault symptoms

The schematic diagram of a GSHP system in the heating mode is depicted in Fig.1[26–28].The system mainly consists of three ma-jor circuits:(a)the ground heat exchanger circuit,(b)the heat pump unit circuit,and(c)the indoor fan coil circuit[29–32].The ground heat exchanger circuit composes of a ground heat exchan-ger and a water pump;the heat pump unit circuit composes of a compressor,an evaporator,a condenser,an electronic expansion valve and a4-way valve;and the indoor fan coil circuit composes of several indoor fan coils and a water pump.The Coef?cient of Performance(COP)of the GSHP system is3.5,and the noise can be controlled less than65decibels.

As mentioned above,the soft faults of GSHP system are dif?cult to detect,which are diagnosed by monitoring the system status. According to references review and practical experience,eight soft faults are imposed in this work:(a)refrigerant overcharge(ReOv);

(b)refrigerant leakage(ReLe);(c)evaporator fouling(EvFo);(d) condenser fouling(CoFo);(e)non-condensable gas(NcGa);(f) compressor suction or discharge valve leakage(CoVL);(g)expan-sion valve port largen(ExPL);and(h)high pressure pipe line block-age(HPLB)[33–36].Each fault has two states,which are present and absent.

The status of GSHP is monitored by using temperature sensors and pressure sensors.The fault symptoms therefore include:(a) evaporating pressure(EvaPr,Pe);(b)condensing pressure(ConPr, Pc);(c)evaporating temperature(EvaTe,Te);(d)condensing temperature(ConTe,Tc);(e)compressor suction temperature (ComST,Ts);(f)compressor discharge temperature(ComDT,Td);

(g)evaporator water temperature difference(EvaTD,D Te);and

(h)condenser water temperature difference(ConTD,D Tc).Each fault symptom obtained from sensor data has three states,which are higher,lower and normal.

The relationship between faults and symptoms obtained from sensor data are given in Table1.Taking ReLe for example,the refrigerant in the heat pump unit circuit decreases because of refrigerant leakage,making both of evaporating pressure and con-densing pressure decrease.The refrigerant discharge superheat temperature therefore increases,making the compressor suction temperature and discharge temperature increase.Due to the fact that the heat pump work in a state of ill health with insuf?cient refrigerant,the heat absorption capacity and heating capacity de-crease,therefore,all of the evaporating temperature,condensing temperature,evaporator water temperature difference and con-denser water temperature difference decrease.

In addition,several fault symptoms can be observed directly by human being,such as(a)compressor can not stop;(b)compressor surface frost;and(c)compressor vibration.These symptoms can help to diagnose the faults of GSHP system more accurately.The relationship between faults and symptoms obtained from observed information is given in Table2.Similarly,taking ReLe for example, four observed fault symptoms including(a)too much foam (ToMuF),(b)compressor surface frost(CoSuF),(c)pungent odor (PunOd),and(d)grease stains in wiped joint(GrSWJ)can be caused by refrigerant leakage.Each fault symptom obtained from observed information has two states,which are present and absent.

3.Fault diagnosis methodology

3.1.Fault diagnosis based on sensor data

The fault diagnosis model of GSHP system is established by using Bayesian network method.Speci?cally,each Bayesian net-

2 B.Cai et al./Applied Energy114(2014)1–9

work is constructed in two consecutive steps,which are de?ning the network structure and de?ning the network parameters.3.1.1.Bayesian network structure

The Bayesian network structure is established according to the cause and effect sequence of events.In this work,faults of GSHP system,such as refrigerant overcharge and evaporator fouling,are the causes;and fault symptoms,such as evaporating pressure is higher compressor suction temperature is lower,are the conse-quences.The relationship is denoted by an arc between them.According to the relationship between faults and fault symptoms obtained from sensor data given in Table 1,the Bayesian networks for fault diagnosis are established as shown in Fig.2.The proposed Bayesian network structure consists of two layers:fault layer and fault symptom layer.The fault layer consists of eight parent nodes,indicating eight potential faults concerned.The symptom layer consists of eight child nodes,indicating eight fault symptoms ob-tained from sensor data.Taking the node EvaPr for example,it is

connected to its eight parent nodes according to eight arcs,which indicates that the fault symptom EvaPr is related to all of the eight faults.

3.1.2.Bayesian network parameters

The prior probabilities and conditional probabilities are re-quired to specify for Bayesian networks.The prior probability of an event is the probability of the event computed before the arrival of new evidence or information.It is obtained according to experi-ences of experts and statistical analysis of historical data.The high-er the prior probability of an event,the more likely the event is to happen.For the GSHP system,the same prior probabilities of faults are assumed,in order to emphasize the posterior probabilities gi-ven new evidences.As shown in Fig.2,the probabilities of faults are all 2%.

A conditional probability is the probability that an event will occur,when another event is known to occur or to have occurred.It is also obtained according to experiences of experts and statisti-cal analysis of historical data.One of the major issues faced is the exponential growth of the number of parameters in the conditional probability tables.The speci?cation of a complete conditional probability table for a child node m with s m states and n parent

nodes requires the assessment of es m à1TQ

n i ?1s i probabilities,where s i is the number of states of parent node i [37].The most common practical solution is the application of Noisy-MAX to sim-plify the conditional probability tables.The noisy gate needs to meet following three assumptions:(a)the child node and all its parents must be variables indicating the degree of presence of an anomaly;(b)each of the parent node must represent a cause that can produce the effect (the child node)in the absence of the other causes;and (c)there may be no signi?cant synergies among the

causes [38].Therefore,only P

n 1es m à1Ts i probabilities are

required

https://www.wendangku.net/doc/c04592829.html,yout of a GSHP system in the heating mode.

Table 1

Relationship between faults and symptoms obtained from sensor data.Fault

Fault symptom EvaPr

ConPr EvaTe ConTe ComST ComDT EvaTD ConTD ReOv Higher Higher Higher N/a Lower Lower Normal Higher ReLe Lower Lower Lower Lower Higher Higher Lower Lower EvFo Lower Lower Lower Lower Lower Lower Higher Higher CoFo Lower Higher Higher Higher Higher Higher Lower Higher NcGa Lower Higher N/a Higher Higher Higher Lower Lower CoVL Lower Lower Higher Lower Higher Lower Lower Lower ExPL Higher Higher Higher N/a Lower Lower Lower Normal HPLB

Lower

Higher

Lower

Higher

Lower

Table 2

Relationship between faults and symptoms obtained from observed information.Fault Fault symptom

ReOv

Compressor can not stop (CoNoS)Compressor surface frost (CoSuF)Compressor vibration (CoVib)ReLe

Too much foam (ToMuF)

Compressor surface frost (CoSuF)Pungent odor (PunOd)

Grease stains in wiped joint (GrSWJ)

NcGa Compressor discharge pressure gauge vibration (CoGaV)CoVL Compressor can not stop (CoNoS)ExPL

Too much foam (ToMuF)

Compressor surface frost (CoSuF)

to specify the conditional probability tables,thereby simplifying knowledge acquisition greatly.

Suppose for example,there are n causes X 1,X 2,...and X n of Y ,by using Noisy-MAX,the full conditional probability relationship can be written as [39,40]

P eY 6y j X T?Y n i ?1x i –0

X y y 0?0q x

i i ;y 0

e1T

P eY ?y j X T?

P eY 60j X Tif y ?0;P eY 6y j X TàP eY 6y à1j X Tif y >0:

e2T

where X represents a certain con?guration of the parents of Y ,X =x 1,...,x n ,and P (Y =0|X 1=0,...,X n =0)=1.

It can be seen from Table 1that when a fault occurs,the corre-sponding fault symptom are occurs theoretically.For example,the fault ReOv causes the fault symptom EvaPr ‘‘Higher’’.However,in practice,the fault symptom is uncertain,for example,the fault ReOv causes the fault symptom EvaPr ‘‘Higher’’,‘‘Lower’’or ‘‘Nor-mal’’.The existed uncertainty problem is caused by various rea-sons,and sensor accuracy and measure uncertainty are the important causes.In the current work,one designer and two repairmen of GSHP systems were invited to determine the rela-tionship between parent nodes and child nodes for sensor data,as given in Table 3.By using the relationship and Eqs.(1)and (2),the conditional probability table can be

computed.

Fig.2.Bayesian networks for fault diagnosis using sensor data.

Table 3

Relationship between parent nodes and child nodes for sensor data.Child node

State

Parent node (present)ReOv

ReLe EvFo CoFo NcGa CoVL ExPL HPLB EvaPr

Higher 0.800.000.110.040.120.110.780.10Lower 0.050.950.850.680.590.820.010.89Normal 0.150.050.040.280.290.070.210.01ConPr

Higher 0.750.000.080.780.990.010.840.73Lower 0.050.900.690.050.000.830.050.10Normal 0.200.100.230.170.010.160.110.17EvaTe

Higher 0.650.000.200.840.010.820.900.01Lower 0.100.920.720.120.010.020.010.80Normal 0.250.080.080.040.980.160.090.19ConTe

Higher 0.970.050.120.780.890.000.010.05Lower 0.000.810.870.100.000.880.020.89Normal 0.030.140.010.120.110.120.970.06ComST

Higher 0.020.690.050.860.590.850.080.84Lower 0.860.000.650.140.100.010.690.02Normal 0.120.310.300.000.310.140.230.14Fig.3.Bayesian networks for fault diagnosis using observed information.

Fig.5.Flow chart for the development of fault diagnosis methodology.

3.3.Multi-source information fusion based fault diagnosis

In order to increase the fault diagnostic accuracy of GSHP sys-

tem,the data and information obtained from sensor and human

being are fused by suing Bayesian network method,and the entire

fault diagnosis model shown in Fig.4is established by combining

the two sub-model in Figs.2and3.Taking the fault‘‘Refrigerant

overcharge’’for example,it can cause not only the change of eight Fig.4.The entire Bayesian networks for fault diagnosis.

4.1.Fault diagnosis using evidences from only sensor data

Table 5gives three fault diagnosis cases using evidences from only sensor data,and the fault diagnosis results is shown in Fig.6.For the case No.1,when the eight sensor data are set as the evidences in the Bayesian networks shown in Fig.2,the poster-ior probabilities of all of the faults are calculated,as shown in Fig.6(a).It is can be seen that the probability of fault present for ‘‘Refrigerant overcharge’’is 99.69%,and the probabilities for other seven fault are almost 0.The diagnostic result is in accordance with the fault found in the practical operation.It indicates that the fault diagnosis model using evidences from only sensor data is accurate Table 5

Three fault diagnosis cases using evidences from only sensor data.Evidence

Case No.1

No.2No.3EvaPr Higher Lower Lower ConPr Higher Higher Lower EvaTe Higher Higher Lower ConTe Higher Higher Lower ComST Lower Lower Higher ComDT Lower Lower Lower EvaTD Normal Lower Lower ConTD

Higher

Lower

Fig.6.Three fault diagnosis results using evidences from only sensor data (a)Case No.1,(b)Case No.2and (c)Case No.3.

6 B.Cai et al./Applied Energy 114(2014)1–9

Fig.7.Three fault diagnosis results using evidences from only sensor data(a)Case No.1+,(b)Case No.2+and(c)Case No.3+.

increases to almost 100%from 61.1%and 52.3%,respectively.The two cases show that the observed information can increase the fault diagnostic accuracy greatly.

As shown in Fig.8(c),the probability of fault present for ‘‘Evap-orator fouling’’decreases to 1.8%from 51.5%,while the probability of ‘‘Refrigerant leakage’’increase to 98.3%from 28.2%,and the probability of ‘‘Compressor suction or discharge valve leakage’’in-crease to 96.2%from 43.8%.The diagnostic result is in accordance

with the faults found in the practical operation.Therefore,the ob-served information can correct the wrong fault diagnostic results.Above all,the multi-source information fusion based fault diag-nosis model can increase the fault diagnostic accuracy greatly.According to the research above,it can be seen that the proposed Bayesian network based fault diagnosis methodology is different from other arti?cial intelligence method based fault diagnosis,such as arti?cial neural networks [10]and data fusion technology [13]based fault diagnosis.The proposed methodology can deal with the uncertainty of fault and fault symptoms well.For example,the present of ReOv can cause EvaPr higher,lower and normal with three probabilities of 80%,5%and 15%,which can be de?ned in the conditional probability table of Bayesian net-works.In addition,several new observed information can be added into the fault diagnosis model easily to increase the diagnosis accuracy.5.Conclusions

In order to increase the diagnostic accuracy,especially for mul-tiple-simultaneous faults,the work proposed a multi-source infor-mation fusion based fault diagnosis methodology for GSHP system.(1)The entire fault diagnosis model of GSHP system is estab-lished by combing two proposed Bayesian networks,which are established according to the cases and effect sequence of faults and fault symptoms,including sensor data and observed information of human being.

(2)The fault diagnosis model using evidences from only sensor

data is accurate for single fault,for example,the probability of fault present for single fault of ‘‘Refrigerant overcharge’’is 99.69%.

(3)The fault diagnosis model using evidences from only sensor

data is not accurate enough for multiple-simultaneous faults,for example,the faults ‘‘Evaporator fouling’’and ‘‘Compressor suction or discharge valve leakage’’have the maximum posterior probabilities of 51.5%and 43.8%,which are not in accordance with the faults found in the practical operation.

(4)The observed information can increase the fault diagnostic

accuracy greatly for single fault,for example,the probability of fault present for ‘‘Refrigerant overcharge’’increases to 100%from 99.69%,while the probabilities of other faults decreases slightly.

(5)The observed information can increase the fault diagnostic

accuracy greatly as well as correct the wrong fault diagnostic results for multiple-simultaneous faults.For example,the probabilities of fault present for ‘‘Non-condensable gas’’and ‘‘Expansion valve port largen’’increases to almost 100%from 61.1%and 52.3%,respectively.

(6)The cases show that the multi-source information fusion

based fault diagnosis model using Bayesian network is effec-tual for GSHP system.

The work focuses on the Bayesian network based fault diagnosis methodology,and a future scope of work can be directed toward the development and validation of Bayesian network based GSHP system automatic fault diagnosis software.Acknowledgements

The authors wish to acknowledge the ?nancial support of the National High-Technology Research and Development Program of China (No.2013AA09A220),National Natural Science Foundation of China (No.51309240),Program for Changjiang Scholars

and

https://www.wendangku.net/doc/c04592829.html,parison of fault diagnosis results (a)Case No.1and No.1+,(b)Case No.and No.2+and (c)Case No.3and No.3+.

Energy 114(2014)1–9

Innovative Research Team in University(IRT1086),Taishan Scholar Project of Shandong Province(TS20110823),Science and Technol-ogy Development Project of Shandong Province(2011GHY11520) and the Fundamental Research Funds for the Central Universities (No.13CX02077A).

References

[1]Michopoulos A,Papakostas KT,Kyriakis N.Potential of autonomous ground-

coupled heat pump system installations in Greece.Appl Energy 2011;88:2122–9.

[2]Bi YH,Wang XH,Liu Y,Zhang H,Chen https://www.wendangku.net/doc/c04592829.html,prehensive energy analysis of a

ground-source heat pump system for both building heating and cooling modes.Appl Energy2009;86:2560–5.

[3]Chua KJ,Chou SK,Yang WM.Advances in heat pump systems:a review.Appl

Energy2010;87:3611–24.

[4]Chua KJ,Chou SK,Yang WM,Yan J.Achieving better energy-ef?cient air

conditioning–a review of technologies and strategies.Appl Energy 2013;104:87–104.

[5]Chen YM,Lan LL.A fault detection technique for air-source heat pump water

chiller heaters.Energy Build2009;41:881–7.

[6]Zogg D,Shafai E,Geering HP.Fault diagnosis for heat pumps with parameter

identi?cation and clustering.Control Eng Pract2006;14:1435–44.

[7]Xiao F,Zheng CY,Wang SW.A fault detection and diagnosis strategy with

enhanced sensitivity for centrifugal chillers.Appl Therm Eng 2011;31:3963–70.

[8]Lee WY,House JM,Kyong NH.Subsystem level fault diagnosis of a building’s

air-handling unit using general regression neural networks.Appl Energy 2004;77:153–70.

[9]Wang SW,Cui JT.Sensor-fault detection,diagnosis and estimation for

centrifugal chiller systems using principal-component analysis method.Appl Energy2005;82:197–213.

[10]Mohanraj M,Jayaraj S,Muraleedharan C.Performance prediction of a direct

expansion solar assisted heat pump using arti?cial neural networks.Appl Energy2009;86:1442–9.

[11]Mohanraj M,Jayaraj S,Muraleedharan C.Applications of arti?cial neural

networks for refrigeration,air conditioning and heat pumps–a review.Renew Sust Energ Rev2012;16:1340–58.

[12]Li H,Braun JE.Decoupling features for diagnosis of reversing and check valve

faults in heat pumps.Int J Refrig2009;32:316–26.

[13]Sun YJ,Wang SW,Huang GS.Online sensor fault diagnosis for robust chiller

sequencing control.Int J Therm Sci2010;49:589–602.

[14]Naja?M,Auslander DM,Bartlett PL,Haves P,Sohn MD.Application of machine

learning in the fault diagnostics of air handling units.Appl Energy 2012;96:347–58.

[15]Gang WJ,Wang JB.Predictive ANN models of ground heat exchanger for the

control of hybrid ground source heat pump systems.Appl Energy2013.http:// https://www.wendangku.net/doc/c04592829.html,/10.1016/j.rser.2011.10.015.

[16]Cai BP,Liu YH,Liu ZK,Tian XJ,Dong X,Yu https://www.wendangku.net/doc/c04592829.html,ing Bayesian networks in

reliability evaluation for subsea blowout preventer control system.Reliab Eng Syst Saf2012;108:32–41.

[17]Cai BP,Liu YH,Zhang YW,Fan Q,Liu ZK,Tian XJ.A dynamic Bayesian networks

modelling of human factors on offshore blowouts.J Loss Prev Process Ind 2013;26:639–49.

[18]Cai BP,Liu YH,Liu ZK,Tian XJ,Zhang YZ,Ji RJ.Application of Bayesian networks

in quantitative risk assessment of subsea blowout preventer operations.Risk Anal2013;33:1293–311.[19]Langseth H,Portinale L.Bayesian networks in reliability.Reliab Eng Syst Saf

2007;92:92–108.

[20]Weber P,Medina-Oliva G,Simon C,Iung B.Overview on Bayesian networks

applications for dependability,risk analysis and maintenance areas.Eng Appl Artif Intell2012;25:671–82.

[21]Dey S,Stori JA.A Bayesian network approach to root cause diagnosis of process

variations.Int J Mach Tools Manuf2005;45:75–91.

[22]Sahin F,Yavuz M?,Arnavut Z,Uluyol O.Fault diagnosis for airplane engines

using Bayesian networks and distributed particle swarm optimization.Parallel Comput2007;33:124–43.

[23]Gonzalez R,Huang B,Xu FW,Espejo A.Dynamic Bayesian approach to gross

error detection and compensation with application toward an oil sands process.Chem Eng Sci2012;67:44–56.

[24]Zhu YG,Liu DY,Chen GF,Jia HY,Yu HL.Mathematical modeling for active and

dynamic diagnosis of crop diseases based on Bayesian networks and incremental learning.Math Comput Model2013;58:514–23.

[25]Zhao Y,Xiao F,Wang SW.An intelligent chiller fault detection and diagnosis

methodology using Bayesian belief network.Energy Build2013;57:278–88.

[26]Capozza A,Carli MD,Zarrella A.Investigations on the in?uence of aquifers on

the ground temperature in ground-source heat pump operation.Appl Energy 2013;107:350–63.

[27]Sagia Z,Rakopoulos C,Kakaras E.Cooling dominated Hybrid Ground Source

Heat Pump System application.Appl Energy2012;94:41–7.

[28]Self SJ,Reddy BV,Rosen MA.Geothermal heat pump systems:status review

and comparison with other heating options.Appl Energy2013;101:341–8. [29]Man Y,Yang HX,Wang JG,Fang ZH.In situ operation performance test of

ground coupled heat pump system for cooling and heating provision in temperate zone.Appl Energy2012;97:913–20.

[30]Montagud C,Corberán JM,Ruiz-Calvo F.Experimental and modeling analysis

of a ground source heat pump system.Appl Energy2013;109:328–36. [31]Ozyurt O,Ekinci DA.Experimental study of vertical ground-source heat pump

performance evaluation for cold climate in Turkey.Appl Energy 2011;88:1257–65.

[32]Park H,Lee JS,Kim W,Kim Y.The cooling seasonal performance factor of a

hybrid ground-source heat pump with parallel and serial con?gurations.Appl Energy2013;102:877–84.

[33]Zhao Y,Wang SW,Xiao F.A statistical fault detection and diagnosis method for

centrifugal chillers based on exponentially-weighted moving average control charts and support vector regression.Appl Therm Eng2013;51:560–72. [34]Kim M,Payne WV,Domanski PA,Yoon SH,Hermes CJL.Performance of a

residential heat pump operating in the cooling mode with single faults imposed.Appl Therm Eng2009;29:770–8.

[35]Yoon SH,Payne WV,Domanski PA.Residential heat pump heating

performance with single faults imposed.Appl Therm Eng2011;31:765–71. [36]Gao DC,Wang SW,Sun YJ,Xiao F.Diagnosis of the low temperature difference

syndrome in the chilled water system of a super high-rise building:a case study.Appl Energy2012;98:597–606.

[37]Pradhan M,Provan G,Middleton B,Henrion M.Knowledge engineering for

large belief networks.In:Proceedings of the tenth international conference on uncertainty in arti?cial intelligence.San Francisco;1994.

[38]Zagorecki A,Druzdzel MJ.Knowledge engineering for bayesian networks:how

common are Noisy-MAX distribution in practice.IEEE Trans Syst Man Cybern Pt A Syst Humans2013;43:186–95.

[39]Díez FJ,Galán SF.Ef?cient computation for the noisy max.Int J Intell Syst

2003;18:165–77.

[40]Li W,Poupart P,Van Beek P.Exploiting structure in weighted model counting

approaches to probabilistic inference.J Artif Intell Res2011;40:729–65. [41]Neapolitan RE.Learning Bayesian networks.London:Prentice Hall;2003.

B.Cai et al./Applied Energy114(2014)1–99