- Email: [email protected]
Review of long-term fault detection approaches in solar thermal systems
Available online at www.sciencedirect.com
Solar Energy 85 (2011) 1430–1439 www.elsevier.com/locate/solener
Review of long-term fault detection approaches in solar thermal systems A.C. de Keizer ⇑,1, K. Vajen 1, U. Jordan 1 Institute of Thermal Engineering, Kassel University, 34109 Kassel, Germany Received 15 September 2009; received in revised form 4 January 2011; accepted 28 March 2011 Available online 2 May 2011 Communicated by: Associate Editor C. Estrada-Gasca
Abstract This paper presents an overview, assessment and comparison of automated fault detection methods that check if solar heating systems are functioning correctly. Fault detection in solar thermal systems is important to minimize the time when the system is not functioning well, thereby ensuring an optimal energy (and economic) yield. During the past decades many systems have been monitored, mainly for scientific or demonstration projects by logging measurement data which was subsequently analysed by an expert. Automation of fault detection is necessary to reduce costs and minimize experts’ time needed for analysis of a system. An overview of existing fault detection approaches is given; these are evaluated and compared with a multi-criteria analysis. The only commercially available automated method, the Input–Output Controller, detects faults causing more than 20% energy loss in the solar loop. The function control approach is cheap without a heat meter, and only relies on few sensors to check how several components in the solar loop are functioning with algorithms. The approach developed at Kassel University checks how well a solar plant is functioning both with plausibility checks and with energy balances based on simulations. This method includes a larger part of the solar heating system and therefore requires more measurement equipment. Further research and application of several fault detection methods should improve the effectiveness and costs of these methods. Ó 2011 Elsevier Ltd. All rights reserved. Keywords: Solar heating systems; Monitoring; Fault detection
1. Introduction Solar thermal energy is expected to make a major contribution to the sustainable supply of low temperature heat in the future. The reliability of solar thermal systems (STS) and components is important in order to obtain an optimum energy production. During the approximately 25 year life span of STS, a fault detection system could result in a quick response and prompt reparation of an occurring malfunction and can offer the following benefits:
⇑ Corresponding author. Tel.: +49 5618043890, fax: +49 5618043993. 1
E-mail address: [email protected] (A.C. de Keizer). ISES member.
0038-092X/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.solener.2011.03.025
Minimisation of energy and economic losses, due to quicker repair of the fault. Normally faults are not noticed directly, since hot water demand and irradiation are variable and since the auxiliary heating continues to supply warm water. Remote supervision, solar thermal systems are decentralised and the owners are usually not experts, therefore remote supervision is useful. Reduction of repair costs, due to more accurate knowledge of the fault and when it occurred. Improved reputation, monitoring can prove whether a system is functioning properly and will on the long term lead to better functioning systems. A threat for the solar thermal market is ‘discredit due to bad previous examples’ (Tsoutsos, 2002).
A.C. de Keizer et al. / Solar Energy 85 (2011) 1430–1439
1431
Nomenclature F0 HDay Hinsuf Kh,eff T
collector efficiency factor daily irradiation, kWh part of HDay that is insufficient to generate an energy yield, kWh effective angle of incidence correction, averaged over the operation time temperature, °C
Support for further development of solar thermal technology, components and systems can be improved based on detailed information on the functioning of certain components or system hydraulics. During the last decades several monitoring and fault detection approaches have been developed. To date, however, fault detection systems are more frequently applied to demonstration or scientific oriented projects than to commercially operated systems. Furthermore, in many cases, a classical approach is used in which the measured data from temperature, heat and flow sensors are automatically logged, but these data are not automatically analysed for faults. A detailed analysis by an expert is costly and not generally applicable for all commercial systems, because of a lack of experts. Therefore, several research groups have started developing automated monitoring approaches. Challenges during the method development are, for example, the large diversity in hydraulics of STS, preventing false alarms and a low end cost for fault detection. This paper intends to bridge the knowledge gap about existing fault detection approaches. An overview of these approaches is given and the methods are compared and evaluated using a multi-criteria analysis. In Section 2, an overview of weaknesses and faults that can occur during the operation of solar thermal systems is given. The fault detection methods are described in Section 3. Both, automated and manual fault detection methods are included as well as methods for all sizes of STS. The multi-criteria analysis and the application for fault detection methods are described in Section 4. The results of the evaluation are presented in Section 5. 2. Malfunctions during the operation of solar thermal systems The type of faults and their frequency of occurrence in STS were studied by several German and Austrian groups. Because of the large diversity in system hydraulics, locations and installation years, these experiences are not generally valid for all system types or situations. The German ‘Center for Solar Technology’ (ZfS) analysed the functioning of 98 systems built between 1978
Qcc Qv,Kap Qv,th U g0 r
daily heat gain of the collector loop, kWh capacitive losses of the collector loop, kWh thermal losses of the whole collector loop, kWh overall heat loss coefficient, W/K zero-loss collector efficiency standard deviation
and 1983 in the Future Invest Programme (ZIP) in early 2000 and of 60 systems built between 1995 and 2005 within the Solarthermie 2000 (ST-2000) Program in 2008 (Peuser et al., 2002, 2008). The results are presented in Fig. 1. The ZIP systems were analysed based on the outcome of detailed questionnaires. Many defects were related to the infancy of the technology and almost or completely disappeared for the systems built in ST-2000, especially malfunctions related to the collector. Nevertheless, in the ST-2000 Program still many system faults were found, even though the operation time of the systems was much lower. Notable is the high defect rate of the pumps and heat exchangers. Also problems with control, ranging from false settings via falsely placed temperature sensors to unsuitable controllers are noteworthy. 20% of the controllers had to be replaced. In the Austrian Optisol project the performance of 10 large solar thermal systems during start up and a two month optimization phase were analysed between 2004 and 2006 (Fink et al., 2006). Many weaknesses were identified, for example, the integration or operation of the auxiliary heating system, an unnecessary large auxiliary heating volume in storage, too high return temperatures of the net and a suboptimal speed control of pumps. On the one hand, the quality of components and systems can be enhanced by introducing a quality label like the SOLAR KEYMARK for collectors and small systems (SOLAR KEYMARK, 2008). On the other hand, a fault detection system is necessary to ensure a quick detection and reparation of an occurring malfunction. 3. Overview of fault detection methods 3.1. Fault detection with manual analysis (MM) The standard approach for fault detection in solar thermal systems consists of automatic data logging of measurements of several temperature and flow sensors and sometimes irradiance measurements, followed by a detailed manual fault analysis by an expert. There is a large variation in the amount and quality of measurement equipment installed and in how the gathered data is analysed. If the quality of the measurement data is good, temperature
1432
A.C. de Keizer et al. / Solar Energy 85 (2011) 1430–1439 0%
10%
20%
30%
40%
50%
60%
Split inner plastic film collector Condensate in collector Split/broken collector cover Corrosion of absorber surface Leaking absorber Discoloration of collector cover Discoloration of absorber plate Efficiency loss collector
Not included in ZIP
Leakage collector loop Damage of pipe insulation Damage of collector loop pump Blow-off in collector loop Insufficient deaerated collector loop Defect expansion vessel Leaking solar storage tank Damage of storage discharging pump Defective control Leakage of roof Reduced power transfer heat exchanger Fouled heat exchanger Corrosion or leakage of HX
ZIP ST 2000
% ZIP: Percentage of defects for 98 systems, based on questionaires % ST-2000: Percentage of defects for reference components,e.g.collector fields, nr of HX (60 systems)
Fig. 1. Faults found within the ZIP project (15 years in operation) and Solarthermie 2000 project (1–10 years in operation) (derived from Peuser et al. (2002, 2008)).
profiles and energy gains can be checked. Thus, manual monitoring can be very effective and accurate in detection and location of faults; however for commercial application it is too time-intensive and costly. 3.2. Optisol (OPT) The Austrian project Optisol (Fink et al., 2006) shows a state of the art example of a combination of manual monitoring together with a few automated function control algorithms. The project is connected to another Austrian project that aims to increase the quality of large solar thermal systems (Qualita¨t, 2009; Austria Solar – Verein zur Fo¨rderung der thermischen Solarenergie, 2009). In the extensive Optisol project 10 solar-supported heating networks for multi-storey residential buildings were built and scientifically monitored. An integral approach for monitoring and design was used. The monitoring part included an improvement phase of ca. 2 months and a routine supervision phase afterwards for a period of one year. In the extensive improvement phase, weaknesses of the solar-supported heating system were recognized by analysing the temperature profiles. Problems were localised, based on the measurement data or by checking the system. 35 faults in installation, design or operation were found, partly caused by the auxiliary heating system. This was unexpected since the design and construction of the systems were extensively done by the support team of Optisol.
In the routine supervision phase fundamental system functions are monitored and monthly energy balances were prepared. Furthermore, key figures are defined like the solar fraction, typical yearly solar gain, and yearly system degree of utilization. These values are compared to the values determined in the planning phase, which are based on the irradiation and the temperature profile of a typical reference year. The comparison is therefore not very useful without further analysis, since irradiation, climate and heat demand differ from a typical year. The key figures give a certain long term trend, but no quick feedback on whether the system is functioning correctly. Large faults can be detected. On the contrary, the optimisation phase is very effective in finding faults, but very time intensive and costly and therefore not applicable for general monitoring. In the Austrian document on minimal standards for operational monitoring (Austria Solar – Verein zur Fo¨rderung der thermischen Solarenergie, 2009), several function control routines are described, these are shown in Table 1. 3.3. Guaranteed Solar Results (GRS) In the Guaranteed Solar Results approach, solar thermal systems are provided with a performance contract. If the performance guarantee for the amount of solar energy is not met, compensation or reparation should take place. The concept was used in several projects. In the European THERMIE-project with the German organisations ASEW
A.C. de Keizer et al. / Solar Energy 85 (2011) 1430–1439
1433
Table 1 Fault algorithms (derived from Austria Solar – Verein zur Fo¨rderung der thermischen Solarenergie (2009)). Fault description
Algorithm
Leakage and defect of expansion tank Operational fault in solar loop
Pressure in solar loop too high or too low If Tcol > Tstor,low + xK and Tcol < 100 °C and Tcol e.g. 20 K > Tsolar,flow,prim If Tcol > Tstor,low + e.g. 45 K, when storage is not full Tsolar_return,primary > Tsolar_flow_primary + e.g. 5 K
System breakdown Unwanted nightly cooling of storage Check charging from buffer to DHW storage (DHW system) Calcifying heat exchanger buffer-DHW storage (DHW system) Too low storage temperature (combi system) Calcifying heat exchanger storage-DHW (combi system + 4 pipe) Too high or low heat distribution net flow temperature or too high net return temperature (combi + 2 pipe) Too high or low system pressure in heat distribution net pressure
and IST Energietechniek and the French TecSol, 20 solar thermal systems between 30 and 687 m2 collector area were realised and monitored (Luboschik et al., 1997). In the German Solarthermie 2000 project the GSR approach was applied to 52 large systems (Peuser et al., 2008). For clarity purposes, two variants are described here, the ‘German’ one as applied in the Solarthermie 2000 project and the ‘French’ one as applied by Tecsol (Tecsol, 1999). In the French approach (Tecsol, 1999), guaranteed results are calculated with the program Solo 2000, which uses monthly weather and hot water consumption data to calculate a yearly solar energy yield. After a year the solar energy gain is recalculated with the measured irradiation and hot water consumption data and then compared to the measured energy yield. If the measured energy yield is a certain percentage less than the calculated energy yield, compensation should be provided. The second comparison is carried out after 4 years, in this calculation only the measured hot water consumption is introduced. In the ‘German’ approach as applied in the Solarthermie 2000 project (Peuser et al., 2008), the tender for the solar thermal system should include the costs of the solar heat as well as the guaranteed solar results, to ensure quality and relatively low costs. The costs of measurement and data logging equipment, installation and one year of monitoring are estimated in the range of 10 k€. This includes a large amount of sensors that can be used for additional manual fault finding. In the ST-2000 project the calculation of the reference energy yield is carried out in three steps (Peuser et al., 2008): – Definition of a correction factor between the guaranteed solar yield calculated by the bidder and the system supervisor, that can arise due to different simulation approaches. – Calculation of solar energy yield with real weather and water consumption data for one year.
T difference between buffer discharging flow and return larger than 15 K for a minimum of 5 min e.g. Tstorage,upper <50 °C T difference between buffer discharging flow and DHW flow larger than 15 K for a minimum of 5 min Via T-sensors Via p-sensors
– Division of measured solar yield by corrected simulation result, if larger than 0.9 the guarantee has been fulfilled. The GRS approach improves the quality of a system and, if the monitoring continues after the first year, larger faults can be found. Nevertheless, a real automatic analysis is not included. Furthermore the reaction time to find a fault after it occurs can be long, since the analysis is not carried out on a short term basis. Also the uncertainties of the simulation can be large due to an inaccurate image of the system and the simulation software chosen. 3.4. Function control for small solar thermal systems without heat measurements (FUS) The ‘function control for small solar thermal systems without heat measurements’ was developed to provide a continuous function control for small solar thermal systems without heat measurements and with hardware costs below €100 (1999 prices) (Altgeld and Mahler, 1999). The method was included in two prototypes of controllers that should provide an error message if necessary. 57 possible malfunctions in the operation of solar thermal systems were identified. These were ranked according to 3 criteria: frequency of occurrence, effect on efficiency and system yield and likeliness of discovery without the fault routine. Problems with auxiliary heating have been excluded, while faults that cause problems with system security or comfort or which are easily recognized by the routine were included. In Table 2, the resulting list of faults and the algorithms to detect and identify the faults are shown. The method was, for test purposes, partially implemented in controllers of the companies Esaa and Wagner. Several faults were recognized, but there were also false positives, detection of a fault while the system was functioning correctly. The approach stays cheap by using mainly sensors that are used for the control, however, with analysis and pressure measurements, the price may be
1434
A.C. de Keizer et al. / Solar Energy 85 (2011) 1430–1439
Table 2 Fault algorithms in the FUS approach (derived from Altgeld and Mahler (1999) and Altgeld (1999)). Fault description
Result of fault
Algorithm (examples for studied system)
1. Collector connections interchanged 2. Collector T-sensor falsely positioned 3. Leakage of heat exchanger 4. Power outage 5. Incorrect controller software 6. False volume flow setting 7. Air in hydraulic circuit 8. DT setting is inappropriate 9. Valve in solar circuit closed 10. Cable breach between controller and pump 11. Defect input or output to the controller 12. Gravity brake is open 13. Fouling of gravity brake 14. Time switch is programmed wrongly 15. Impressed-current anode defect 16. Temperature sensor of collector defect or high 17. Temperature sensor in storage defect or high 18. Primary pressure in the expansion tank too high 19. Primary pressure expansion tank too low
Pump clocks Pump clocks System pressure too high System pressure too high DT too high DT too high DT too high DT too high DT too high DT too high DT too high + Pump on at night Pump on at night Pump on at night Pump on at night Storage tank corrodes Wrong DT => pump runs at wrong Wrong DT => pump runs at wrong System pressure at times too low System pressure at times too high
Pump running time < 10 s. Pump running time < 10 s. (Tcol = 20 °C) AND (psys = psys,nom + 2 bar) (Tcol = 20 °C) AND (psys = psys,nom + 2 bar) Pump on AND DT = DTnom + 15 K Pump on AND DT = DTnom + 15 K Pump on AND DT = DTnom + 15 K Pump on AND DT = DTnom + 15 K Pump on AND DT = DTnom + 15 K Pump on AND DT = DTnom + 15 K Pump on AND DT = DTnom + 15 K Pump on AND time between 22:00 and 6:00 Pump on AND time between 22:00 and 6:00 Pump on AND time between 22:00 and 6:00 Current measurement
slightly higher than anticipated. Although the method succeeds at detecting several faults in the lab, location of faults is limited, as can be seen in Table 2. Furthermore, there is no yield measurement, which could mean that large energy losses are not detected. 3.5. Spectral method (SPM) The spectral method is based on analyzing the transient temperature changes in the collector circuit after the pump is starting (Grossenbacher, 2003; Synetrum, 1998). The only extra temperature sensor is installed a few metres after the collector outlet. This temperature is measured every second, transformed to a dT/dt plot and subsequently to a spectral range via a discrete Fourier transformation. A fault free training phase results in a characteristic vector and an uncertainty boundary. A measured vector out of range shows a fault. The method was tested for two systems in Switzerland; with a temperature difference for starting the pump of 8 K and 30 K respectively (Grossenbacher, 2003). Dirty glass, or a reduction of collector performance can be detected if the loss of power is large enough, e.g. 40% reduction of collector performance. Furthermore, air in the collector or in the heat exchanger could be easily detected. A reduction of at least 20% in volume flow was recognized. Also a change of DT-settings with about 20% can be recognized for the low flow system. The method has the advantage of being cheap; only one extra temperature sensor is needed. In addition no parameterization is necessary and faults occurring over a longer time frame can be recognized. A disadvantage is that a fault free training phase is needed; a minimum time frame of 300 days and 240 start ups of the pump are suggested (Grossenbacher, 2003). Furthermore not all faults can be recognized and logging data on a secondly time basis can be complex. A
limited identification of grouped faults is theoretically possible. 3.6. Input–Output Controller (IOC) In the Input–Output Controller device a fault detection method is integrated, that compares calculated with measured energy yields. It has been commercially available since 2007 (www.resol.de, 2008; Pa¨risch and Vanoli, 2007a). The method monitors the energy yields in the solar circuit; a second approach in which also the buffer storage discharging is included was developed, but is not yet commercially available. The IOC compares the daily measured and expected energy yields of the solar loop. The measured energy input is based on the volume flow of solar or secondary loop and the flow and return temperatures of the corresponding loop. The energy gains are calculated with an approach developed by Pa¨risch and Vanoli (2007b), based on irradiation, ambient temperature and temperature of the heat sink. The daily heat gain of the collector circuit (Qcc) is calculated with Eq. (1) (Pa¨risch and Vanoli, 2007b). This is based on the daily irradiation HDay (kWh) and the part of the irradiation occurring at a time when the collector is not at a sufficient temperature to charge the storage (Hinsuf). All of these are calculated based on the expected collector circuit temperature, which is calculated with differential equations for different forms of operation (no pump running, operation, pause due to irradiance reduction, or stagnation) based on planning data, characteristics of collectors, pipe lines and pumps and the measured values of irradiance, solar last temperature and ambient temperature. Qcc ¼ g0 K h;eff ðH Day H insuf Þ QV;th QV;Kap
ð1Þ
The standard uncertainty (r) of the IOC-procedure, including measurements and yield calculation, is about 7% of the
A.C. de Keizer et al. / Solar Energy 85 (2011) 1430–1439
calculated energy gain (Pa¨risch and Vanoli, 2007b). If the difference between measured and simulated yield is larger than 20% (3 r) a fault is reported. This leads to 99% reliability for a correct fault prediction. The uncertainty margins are larger below a calculated yield of 1.5 kWh/m2d. There is a fault tree to distinguish if the fault occurred inside or outside the solar loop, and if it is for example the control or the solar station which causes the problems. The IOC is sold for 1190 € inclusive temperature and irradiance sensors, but without flow measurements. To be able to check the performance from home an extra data logger is necessary (Pa¨risch and Vanoli, 2007b). 3.7. Fault diagnostic system based on artificial neural networks (ANN) Kalogirou et al. (2008) presents a fault detection method for solar thermal systems based on simulations with artificial neural networks. Artificial neural networks are a mathematical method based on biological neural networks that can learn to predict results from examples. They are supposedly faster than simulations with e.g. TRNSYS and are able to deal with incomplete data and non-linear problems. ANN’s behave as a black box (Kalogirou et al., 1999, 2008). The method consists of three steps. In the prediction module, artificial neural networks are trained with data of a TRNSYS model that is assumed to represent a faultfree system. The model is trained so that four temperature values (collector in- and output and storage in- and output) can be predicted for different environmental conditions. The input consists of weather data (global and beam radiation, ambient temperature, incidence angle, wind speed, relative humidity, flow (yes/no)), together with one of the other measured temperatures. In the second step residual values, which characterize e.g. the actual temperature increase against the predicted temperature increase in the collector are calculated. In the last step a diagnosis module is run. Five consecutive residual values outside the boundaries are classified as a fault. Furthermore, if the average of 10 consecutive residual values is outside the boundaries this is reported as a fault. Several faults were simulated with TRNSYS and detected by the procedure, e.g. for a 5% decrease of F0 or an increase of 10% of the U-value of the pipes (Kalogirou et al., 2008). It needs to be studied how this method functions for real systems, which do have measurement uncertainties and may have different faulty behaviour than the ones modelled in TRNSYS. Furthermore the ANN was trained with TRNSYS and not yet with real data. 3.8. Method developed at Kassel University (KU) At Kassel University, a fault detection method was developed, that combines a static function control with dynamic simulation based fault detection (Wiese, 2006, Wiese et al., 2007). The method consists of three steps:
1435
1. Check for missing data 2. Plausibility check 3. Comparison of simulations and measurements In the first step the data integrity is checked. A minimum of 95% of the data points should be available to continue the fault detection. In a second step a plausibility check is carried out, which checks whether individual components function correctly. Table 3 shows the inclusion of automatic detection of faults in the second step. The approach was tested for several large German domestic hot water systems. The third step consists of a daily comparison between measured and simulated energy yields at the first and second heat exchanger. The simulation environment TRNSYS is used. These analyses include uncertainty estimates for both, the measured as well as the simulated energy yields. Several faults could be detected for large enough energy losses. However, so far faults cannot be identified in the simulation based step. 3.9. Concluding remarks After several methods have been introduced in the previous paragraphs, some more method properties are investigated and listed in Table 4. The time scale of the data logging is important, to know what type of faults can be derived from the data. Although a shorter time step is preferable for fault detection, it also leads to higher costs. The optimal time depends on a trade off between those two. The time scale of the analysis is important since it indicates how
Table 3 Fault detection with plausibility checks (Wiese, 2006). Verification of functioning of controller Breakdown of controller Breakdown of sensor Inaccurate sensors False control criteria Inappropriate control scheme Position collector T sensor Inappropriate T/DT settings Incorrect sensor position Breakdown of gravity brake
Yes ± ± ±/yes ± ±/yes ± ± Yes
Verification of volume flow Volume flow too small Too small pump, wrong pump step Air in hydraulic circuit Fouling of hydraulic circuit Primary pressure in hydraulic circuit too low Volume flow too high
Yes Yes Yes Yes Yes
Verification of heat exchanger performance Dimensioned too small Fouling Hydraulics wrongly connected
Noa ± ±
a Fault detection approaches are usually not designed to identify planning faults.
1436
A.C. de Keizer et al. / Solar Energy 85 (2011) 1430–1439
Table 4 Characteristics of fault detection methods. Characteristics a
Time scale of data logging Time scale of analysis Calculation of energy yield Scale of the tested systems (collector area in m2) Stage of developmentc Level of automationd a
MM
OPT
GRS
FUS
SPM
IOC
ANN
KU
Var Var No Var
15 min No 30–250
Var Mon or yr Yes Very large
<1 min ? No 5
1s Sec No 7–16
Min Day Yes 2–455
Hour? Hour Yes 40b
1 min 10 min & day Yes 88–400
++
++ +
++ +
++ ++
++
++ ++
++
+ ++
Time scale: Var = variable; Sec = second; Min = minute; Hour; Day = day; Mon = month; Yr = year. TRNSYS simulation. ++ well developed/can be used directly, + advanced stage of development, early R&D. not automated, + partly automated, ++ fully automated.
b c d
long it takes before a fault gets noticed. With a calculation of energy yield, a comparison with measurement data can show if the total energy gain of the solar loop or the solar heating system is sufficient. The stage of development shows if the method is commercially available, in an advanced stage of development or still in an early research stage. Finally the level of automation indicates if the method is automated, or if it is mostly manual work. The required measurement equipment of each approach is listed in Table 5.
4. Methodology for comparison 4.1. Introduction A partial multi-criteria analysis (MCA) will be used to evaluate and compare methods that can be used for detection and identification of defects or malfunctions during the operation of solar thermal systems. An MCA is often used for supporting policy decisions and evaluating different alternatives. One of the advantages of an MCA is that
Table 5 Measurement equipment required for fault detection methods. *
Sensor
Fus
Opt
SP M
T T T T T T T T T T T T T P VF VF VF VF VF
Collector output Collector input Ca. 1 metre behind collector output Flow solar loop – primary Return solar loop – primary Flow and return solar loop – secondary Flow and return auxiliary heating loop Solar tank output Auxiliary tank output Flow and return buffer storage discharge loop Distribution net flow and return Storage Cold water Pressure primary solar loop Primary solar loop Secondary solar loop Cold water Discharge loop HX – DHW storage Circulation Flow (yes/no) solar primary loop Ambient Tilted Global on a horizontal surface Relative humidity Wind speed Solar loop (secondary) Auxiliary heating Net Dwelling
x
x
x
x x
x x x x
T G G
HF HF HF HF * a
IOC 1
AN N
GR S Fr
GR S Dea
x x
KU x
x x
x x
x x
x x x x
x x x
x n.a. x x
x n.a. x x
x x x x x
x x x
x x
x x
x x
Low
n.a. x x n.a.
Max TSL x
x
x x
x
n.a.
x
x
x x
x x x x x x
x
x x x x
T = Temperature, P = Pressure, VF = volume flow, G = irradiation, HF = heat flow (VF and flow and return temperature included in HF metre). Refers to extensive measurement equipment in the Solarthermie 2000 project, only part of that is shown here.
A.C. de Keizer et al. / Solar Energy 85 (2011) 1430–1439 Table 6 Criteria used for evaluation of the methods. Criteria
Evaluation
Explanation of evaluation
Automated fault detection included? Automated fault identification included? Accuracy of fault detection
y/n
Yes/no
-/0/+/++
No/somehow/yes/very good
?
Not enough information available for analysis Detects some faults Good Very good
Accuracy of fault identification
Costs (operational and hardware)
Monitored part of solar heating system
Time until detection of fault
0 + ++ n.a.
not applicable
0 ++ -
Good Very good Expensive
0 + ++ var
Middle cost range Cheaper Cheap Variable
sl bs w-aux
Solar loop Including buffer storage loop Whole system minus auxiliary heating loop In the range of days, weeks, month(s), year(s)
d/w/m/y
the criteria on which basis the comparison is made are explicit. There are several MCA methods that weigh the different scores and combine these to a result score. In this paper a relatively simple approach will be used. The four steps of the MCA that are conducted are the following (Dodgson et al., 2000): 1. Definition of the aims of the MCA, and the aims of decision makers and other key players 2. Identification of the methods/options for achieving the objectives 3. Identification of the criteria to be used to compare options 4. Description of the expected performance of each option against the criteria (performance matrix) The objective of the multi-criteria analysis is to evaluate different fault detection methods based on the performance of the method regarding e.g. effectiveness, flexibility and costs. The fault detection methods have been identified in an extensive literature review and are described in Section 3. The methods are valid for monitoring solar thermal systems for the entire life time of the system. The following eight approaches have been discussed in this paper: Manual monitoring (MM) Optisol (OPT) Guaranteed Solar Results (GRS)
1437
Function control for small solar thermal systems without heat measurements (FUS) Spectral method (SPM) Input–Output Controller (IOC) Fault detection based on simulations with Artificial Neural Networks (ANN) Method developed at Kassel University (KU) 4.2. Criteria to analyse performance The terms of monitoring, fault detection and identification are often intertwined and go with different definitions in different papers, e.g. with fault detection sometimes only monitoring without analysis is meant. The terms will be used here in the following context. Monitoring is defined as data logging of a variable amount of measurement data, however this data is not automatically analysed, so it does not automatically lead to a fault declaration. In a fault detection method, measurement data is automatically analysed and in case of a malfunction a fault indication follows. Fault identification or localisation requires the identification of the type of fault. This will make reparation much easier. Relevant criteria to evaluate the performance of the fault detection and monitoring methods have been identified and are shown in Table 6. The possible evaluations for these criteria are shown in the second column. The first criterion indicates if automated fault detection is included (yes/no). The following criterion shows if also automated fault identification is included in the method. The first criterion can clearly be analysed, for the second criterion there is a gliding scale between no () somehow (0), yes (+) or a very good (++) and detailed identification approach. Nevertheless the quality and accuracy of detection and identification can be very different. This is what the third and fourth criteria address. These roughly refer to how many fault types can be found and how large an energy loss or other fault result needs to be before it is detected. This criterion includes the uncertainty in the analysis that results from measurement and simulation uncertainties. However, this can only be qualitatively analysed, since methods have been developed and applied to very different systems, and publications often do not provide very accurate descriptions of effectiveness and accuracies of the methods. Therefore a very detailed analysis cannot be carried out. Furthermore, it is important to address which part of the solar heating system is included, since faults also occur outside the solar loop. Costs are an important factor as well. The last criterion reflects on how long it takes before a fault is identified. This one gives only a rough indication, since it depends on the effect of the found fault, however also on e.g. the time of analysis. 5. Results The performance matrix in Table 7 shows the performance of the fault detection methods based on the different criteria.
1438
A.C. de Keizer et al. / Solar Energy 85 (2011) 1430–1439
Table 7 Performance matrix. Criteria
MM
OPT
GRS
FUS
SPM
IOC
ANN
KU
Automatic fault detection included? Accuracy of fault detection Automatic fault identification included? Accuracy of fault identification Costs Monitored part of solar heating system Time until detection of fault
No ++e ++e var Var d/w
n/y ++e 0 ++e var Whole d/w
No 0 n.a. 10 k€b aux y
Yes 0 + 0 ++100 € sl d
Yes ? n.a. +? sl d/w
Yes + 0 0 +1190 €a sl, (bsd) d/w
Yes ? n.a. +? sl d/w
Yes + + 0 0 240–960 €c aux d/w
a
IOC: hardware and temperature sensors. Costs for extensive measurement equipment, including one year monitoring in ‘German’ GRS. c Costs per year for 20 year monitoring, including hardware and at least 30 monitoring systems sold. The main costs are expected for maintenance and improvement of software (between € 15 and 50 per month) (Wiese, 2006). d Only in IOC 2 (not commercially available yet). e Includes manual monitoring. b
Manual monitoring is very effective, but since not automated, very expensive and not applicable for a large number of systems. It is, however, easier to adapt to the extensive variation in hydraulics and systems. The FUS method is very low cost and, therefore, can be applied well to small systems. FUS detects several faults at reasonable costs, but does not make an energy balance. The spectral method and the method with simulations based on artificial neural networks are not developed far enough to know how the results will be. Both methods also need a long training phase in which a system assumedly runs fault free. The IOC is the first method which could result in the commercial implementation of a monitoring and fault detection method for larger solar thermal systems. It has been tested and is commercially available at a reasonable cost. However, the commercial available version only applies to the solar loop and not to the whole solar heating system. The method developed at Kassel University is still under development, but could also provide an automatic monitoring solution for large systems. It includes more sensors and a larger part of the system than the IOC approach, and can therefore also analyse individual components. So far the auxiliary heating system is not included, but this is planned. At the moment, none of the above described approaches takes the auxiliary heating system into account, which is also an important source of errors. 6. Discussion, conclusion and recommendations In this paper an overview and assessment of fault detection methods for solar thermal systems has been presented. A multi-criteria analysis has been used to compare the different approaches. Most methods are still in development and very detailed information was not available. The only commercially available automated method, the Input–Output Controller, should detect faults causing more than 20% energy loss in the solar loop. This method is recommended for middle to large sized systems. The FUS approach without a heat flow meter is cheap and only relies on few sensors to check the functioning of several
components in the solar loop with algorithms. The approach developed at Kassel University checks the functioning of a solar plant both with plausibility checks and with energy balances based on simulations. This method includes a larger part of the solar system, requires more measurement equipment and is, due to cost reasons, more suitable for large systems. Currently methods are being further developed. That is useful, since practical experience on effectiveness has to be gained and also costs should be reasonable. Attention should be paid to what part of the system is monitored; since the auxiliary heating also causes problems, it would be wise to include this in monitoring methodology. Further research and practical application of these fault detection methods should improve the knowledge on their costs and effectiveness and their applicability for certain requirements and system hydraulics. Acknowledgement The authors gratefully acknowledge the financial support provided by the Marie Curie early stage Research Training Network ‘Advanced solar heating and cooling for buildings – SOLNET’ that is funded by the European Commission under Contract MEST-CT-2005-020498 of the Sixth Framework Programme. References Altgeld, H., 1999. Funktionskontrollen bei kleinen thermischen Solaranlagen ohne Wa¨rmemengenmessung. Hochschule fu¨r Technik und Wirtschaft des Saarlandes, Saarbruecken, Germany.. Altgeld, H., Mahler, M., 1999. Funktionskontrolle bei kleinen thermischen Solaranlagen ohne Wa¨rmemengenmessung. Testzentrum Saarbru¨cken, Saarbru¨cken, Germany. Austria Solar - Verein zur Fo¨rderung der thermischen Solarenergie, 2009. Detailbeschreibung zu Musterdokument D5, Mindeststandard fu¨r Kontrolleinrichtungen zur Anlagenoptimierung und Betriebsu¨berwachung. Austria Solar. . Dodgson, J., Spackman, M., Pearman, A.D., Phillips, L.D., 2000. Multicriteria Analysis: a Manual. Department of the Environment, Transport and Regions, London.
A.C. de Keizer et al. / Solar Energy 85 (2011) 1430–1439 Fink, C., Riva, R., Pertl, M., Wagner, W., 2006. OPTISOL - Messtechnisch begleitete Demonstrationsprojekte fu¨r optimierte und standardisierte Solarsysteme im Mehrfamilienwohnbau. AEE - Institut fu¨r Nachhaltige Technologien, Gleisdorf, Austria.. Grossenbacher, U., 2003. Qualita¨tssicherungssystem fu¨r Solaranlagen; Methode zur permanenten Funktionskontrolle thermischer Solaranlagen. EnergieBu¨ro Grossenbacher, Murten, Switzerland. Kalogirou, S.A., Panteliou, S., Dentsoras, A., 1999. Modeling of solar domestic water heating systems using artificial neural networks. Solar Energy 65, 335–342. Kalogirou, S., Lalot, S., Florides, G., Desmet, B., 2008. Development of a neural network-based fault diagnostic system for solar thermal applications. Solar Energy 82, 164–172. Luboschik, U., Schalajda, P., Halagic, N., Heinzelmann, P.J., Backes, J., 1997. Garantierte Resultate von thermischen Solaranlagen; Ein Projekt zur Markteinfu¨hrung solarthermischer Anlagen. ASEW, Schlußbericht Projekt SE/475/93/DE/FR, EU project. Pa¨risch, P., Vanoli, K., 2007a. Quality assurance with the ISFH-Input/ Output-Procedure 6-year-experience with 14 solar thermal systems. Proceedings of ESTEC 2007. Freiburg, Germany, pp. 315–320. Pa¨risch, P., Vanoli, K., 2007b. Wissenschaftlicher Schlussbericht; Forschungsvorhaben: Wissenschaftlich-technische Untersuchung des ISFH-Input/Output-Verfahrens zur Ertragskontrolle solarthermischer Systeme sowie Entwicklung und Erprobung von Input/Output-Controllern. Institut fu¨r Solarenergieforschung GmbH, Hameln/Emmerthal, Germany.
1439
Peuser, F., Remmers, K., Schnauss, M., 2002. Solar Thermal Systems. Solar Praxis AG, Germany. Peuser, F.A., Croy, R., Mies, M., Rehmann, U., Wirth, H.P., 2008. Solarthermie 2000, Teilprogramm 2 und Solarthermie 2000plus, Wissenschaftlich-technische Programmbegleitung und Messprogramm (Phase 3), Abschlussbericht zum Projekt 032 9601 L. ZfS - Rationelle Energietechnik GmbH, Hilden, Germany. Qualita¨tssteigerung und Qualita¨tssicherung großer thermischer Solaranlagen in geba¨udeu¨bergreifenden Mikronetzen und im großvolumigen Wohnbau, (accessed 24.4.2009). SOLAR KEYMARK - The Quality Label for Solar Thermal Products in Europe,
Solar Energy 85 (2011) 1430–1439 www.elsevier.com/locate/solener
Review of long-term fault detection approaches in solar thermal systems A.C. de Keizer ⇑,1, K. Vajen 1, U. Jordan 1 Institute of Thermal Engineering, Kassel University, 34109 Kassel, Germany Received 15 September 2009; received in revised form 4 January 2011; accepted 28 March 2011 Available online 2 May 2011 Communicated by: Associate Editor C. Estrada-Gasca
Abstract This paper presents an overview, assessment and comparison of automated fault detection methods that check if solar heating systems are functioning correctly. Fault detection in solar thermal systems is important to minimize the time when the system is not functioning well, thereby ensuring an optimal energy (and economic) yield. During the past decades many systems have been monitored, mainly for scientific or demonstration projects by logging measurement data which was subsequently analysed by an expert. Automation of fault detection is necessary to reduce costs and minimize experts’ time needed for analysis of a system. An overview of existing fault detection approaches is given; these are evaluated and compared with a multi-criteria analysis. The only commercially available automated method, the Input–Output Controller, detects faults causing more than 20% energy loss in the solar loop. The function control approach is cheap without a heat meter, and only relies on few sensors to check how several components in the solar loop are functioning with algorithms. The approach developed at Kassel University checks how well a solar plant is functioning both with plausibility checks and with energy balances based on simulations. This method includes a larger part of the solar heating system and therefore requires more measurement equipment. Further research and application of several fault detection methods should improve the effectiveness and costs of these methods. Ó 2011 Elsevier Ltd. All rights reserved. Keywords: Solar heating systems; Monitoring; Fault detection
1. Introduction Solar thermal energy is expected to make a major contribution to the sustainable supply of low temperature heat in the future. The reliability of solar thermal systems (STS) and components is important in order to obtain an optimum energy production. During the approximately 25 year life span of STS, a fault detection system could result in a quick response and prompt reparation of an occurring malfunction and can offer the following benefits:
⇑ Corresponding author. Tel.: +49 5618043890, fax: +49 5618043993. 1
E-mail address: [email protected] (A.C. de Keizer). ISES member.
0038-092X/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.solener.2011.03.025
Minimisation of energy and economic losses, due to quicker repair of the fault. Normally faults are not noticed directly, since hot water demand and irradiation are variable and since the auxiliary heating continues to supply warm water. Remote supervision, solar thermal systems are decentralised and the owners are usually not experts, therefore remote supervision is useful. Reduction of repair costs, due to more accurate knowledge of the fault and when it occurred. Improved reputation, monitoring can prove whether a system is functioning properly and will on the long term lead to better functioning systems. A threat for the solar thermal market is ‘discredit due to bad previous examples’ (Tsoutsos, 2002).
A.C. de Keizer et al. / Solar Energy 85 (2011) 1430–1439
1431
Nomenclature F0 HDay Hinsuf Kh,eff T
collector efficiency factor daily irradiation, kWh part of HDay that is insufficient to generate an energy yield, kWh effective angle of incidence correction, averaged over the operation time temperature, °C
Support for further development of solar thermal technology, components and systems can be improved based on detailed information on the functioning of certain components or system hydraulics. During the last decades several monitoring and fault detection approaches have been developed. To date, however, fault detection systems are more frequently applied to demonstration or scientific oriented projects than to commercially operated systems. Furthermore, in many cases, a classical approach is used in which the measured data from temperature, heat and flow sensors are automatically logged, but these data are not automatically analysed for faults. A detailed analysis by an expert is costly and not generally applicable for all commercial systems, because of a lack of experts. Therefore, several research groups have started developing automated monitoring approaches. Challenges during the method development are, for example, the large diversity in hydraulics of STS, preventing false alarms and a low end cost for fault detection. This paper intends to bridge the knowledge gap about existing fault detection approaches. An overview of these approaches is given and the methods are compared and evaluated using a multi-criteria analysis. In Section 2, an overview of weaknesses and faults that can occur during the operation of solar thermal systems is given. The fault detection methods are described in Section 3. Both, automated and manual fault detection methods are included as well as methods for all sizes of STS. The multi-criteria analysis and the application for fault detection methods are described in Section 4. The results of the evaluation are presented in Section 5. 2. Malfunctions during the operation of solar thermal systems The type of faults and their frequency of occurrence in STS were studied by several German and Austrian groups. Because of the large diversity in system hydraulics, locations and installation years, these experiences are not generally valid for all system types or situations. The German ‘Center for Solar Technology’ (ZfS) analysed the functioning of 98 systems built between 1978
Qcc Qv,Kap Qv,th U g0 r
daily heat gain of the collector loop, kWh capacitive losses of the collector loop, kWh thermal losses of the whole collector loop, kWh overall heat loss coefficient, W/K zero-loss collector efficiency standard deviation
and 1983 in the Future Invest Programme (ZIP) in early 2000 and of 60 systems built between 1995 and 2005 within the Solarthermie 2000 (ST-2000) Program in 2008 (Peuser et al., 2002, 2008). The results are presented in Fig. 1. The ZIP systems were analysed based on the outcome of detailed questionnaires. Many defects were related to the infancy of the technology and almost or completely disappeared for the systems built in ST-2000, especially malfunctions related to the collector. Nevertheless, in the ST-2000 Program still many system faults were found, even though the operation time of the systems was much lower. Notable is the high defect rate of the pumps and heat exchangers. Also problems with control, ranging from false settings via falsely placed temperature sensors to unsuitable controllers are noteworthy. 20% of the controllers had to be replaced. In the Austrian Optisol project the performance of 10 large solar thermal systems during start up and a two month optimization phase were analysed between 2004 and 2006 (Fink et al., 2006). Many weaknesses were identified, for example, the integration or operation of the auxiliary heating system, an unnecessary large auxiliary heating volume in storage, too high return temperatures of the net and a suboptimal speed control of pumps. On the one hand, the quality of components and systems can be enhanced by introducing a quality label like the SOLAR KEYMARK for collectors and small systems (SOLAR KEYMARK, 2008). On the other hand, a fault detection system is necessary to ensure a quick detection and reparation of an occurring malfunction. 3. Overview of fault detection methods 3.1. Fault detection with manual analysis (MM) The standard approach for fault detection in solar thermal systems consists of automatic data logging of measurements of several temperature and flow sensors and sometimes irradiance measurements, followed by a detailed manual fault analysis by an expert. There is a large variation in the amount and quality of measurement equipment installed and in how the gathered data is analysed. If the quality of the measurement data is good, temperature
1432
A.C. de Keizer et al. / Solar Energy 85 (2011) 1430–1439 0%
10%
20%
30%
40%
50%
60%
Split inner plastic film collector Condensate in collector Split/broken collector cover Corrosion of absorber surface Leaking absorber Discoloration of collector cover Discoloration of absorber plate Efficiency loss collector
Not included in ZIP
Leakage collector loop Damage of pipe insulation Damage of collector loop pump Blow-off in collector loop Insufficient deaerated collector loop Defect expansion vessel Leaking solar storage tank Damage of storage discharging pump Defective control Leakage of roof Reduced power transfer heat exchanger Fouled heat exchanger Corrosion or leakage of HX
ZIP ST 2000
% ZIP: Percentage of defects for 98 systems, based on questionaires % ST-2000: Percentage of defects for reference components,e.g.collector fields, nr of HX (60 systems)
Fig. 1. Faults found within the ZIP project (15 years in operation) and Solarthermie 2000 project (1–10 years in operation) (derived from Peuser et al. (2002, 2008)).
profiles and energy gains can be checked. Thus, manual monitoring can be very effective and accurate in detection and location of faults; however for commercial application it is too time-intensive and costly. 3.2. Optisol (OPT) The Austrian project Optisol (Fink et al., 2006) shows a state of the art example of a combination of manual monitoring together with a few automated function control algorithms. The project is connected to another Austrian project that aims to increase the quality of large solar thermal systems (Qualita¨t, 2009; Austria Solar – Verein zur Fo¨rderung der thermischen Solarenergie, 2009). In the extensive Optisol project 10 solar-supported heating networks for multi-storey residential buildings were built and scientifically monitored. An integral approach for monitoring and design was used. The monitoring part included an improvement phase of ca. 2 months and a routine supervision phase afterwards for a period of one year. In the extensive improvement phase, weaknesses of the solar-supported heating system were recognized by analysing the temperature profiles. Problems were localised, based on the measurement data or by checking the system. 35 faults in installation, design or operation were found, partly caused by the auxiliary heating system. This was unexpected since the design and construction of the systems were extensively done by the support team of Optisol.
In the routine supervision phase fundamental system functions are monitored and monthly energy balances were prepared. Furthermore, key figures are defined like the solar fraction, typical yearly solar gain, and yearly system degree of utilization. These values are compared to the values determined in the planning phase, which are based on the irradiation and the temperature profile of a typical reference year. The comparison is therefore not very useful without further analysis, since irradiation, climate and heat demand differ from a typical year. The key figures give a certain long term trend, but no quick feedback on whether the system is functioning correctly. Large faults can be detected. On the contrary, the optimisation phase is very effective in finding faults, but very time intensive and costly and therefore not applicable for general monitoring. In the Austrian document on minimal standards for operational monitoring (Austria Solar – Verein zur Fo¨rderung der thermischen Solarenergie, 2009), several function control routines are described, these are shown in Table 1. 3.3. Guaranteed Solar Results (GRS) In the Guaranteed Solar Results approach, solar thermal systems are provided with a performance contract. If the performance guarantee for the amount of solar energy is not met, compensation or reparation should take place. The concept was used in several projects. In the European THERMIE-project with the German organisations ASEW
A.C. de Keizer et al. / Solar Energy 85 (2011) 1430–1439
1433
Table 1 Fault algorithms (derived from Austria Solar – Verein zur Fo¨rderung der thermischen Solarenergie (2009)). Fault description
Algorithm
Leakage and defect of expansion tank Operational fault in solar loop
Pressure in solar loop too high or too low If Tcol > Tstor,low + xK and Tcol < 100 °C and Tcol e.g. 20 K > Tsolar,flow,prim If Tcol > Tstor,low + e.g. 45 K, when storage is not full Tsolar_return,primary > Tsolar_flow_primary + e.g. 5 K
System breakdown Unwanted nightly cooling of storage Check charging from buffer to DHW storage (DHW system) Calcifying heat exchanger buffer-DHW storage (DHW system) Too low storage temperature (combi system) Calcifying heat exchanger storage-DHW (combi system + 4 pipe) Too high or low heat distribution net flow temperature or too high net return temperature (combi + 2 pipe) Too high or low system pressure in heat distribution net pressure
and IST Energietechniek and the French TecSol, 20 solar thermal systems between 30 and 687 m2 collector area were realised and monitored (Luboschik et al., 1997). In the German Solarthermie 2000 project the GSR approach was applied to 52 large systems (Peuser et al., 2008). For clarity purposes, two variants are described here, the ‘German’ one as applied in the Solarthermie 2000 project and the ‘French’ one as applied by Tecsol (Tecsol, 1999). In the French approach (Tecsol, 1999), guaranteed results are calculated with the program Solo 2000, which uses monthly weather and hot water consumption data to calculate a yearly solar energy yield. After a year the solar energy gain is recalculated with the measured irradiation and hot water consumption data and then compared to the measured energy yield. If the measured energy yield is a certain percentage less than the calculated energy yield, compensation should be provided. The second comparison is carried out after 4 years, in this calculation only the measured hot water consumption is introduced. In the ‘German’ approach as applied in the Solarthermie 2000 project (Peuser et al., 2008), the tender for the solar thermal system should include the costs of the solar heat as well as the guaranteed solar results, to ensure quality and relatively low costs. The costs of measurement and data logging equipment, installation and one year of monitoring are estimated in the range of 10 k€. This includes a large amount of sensors that can be used for additional manual fault finding. In the ST-2000 project the calculation of the reference energy yield is carried out in three steps (Peuser et al., 2008): – Definition of a correction factor between the guaranteed solar yield calculated by the bidder and the system supervisor, that can arise due to different simulation approaches. – Calculation of solar energy yield with real weather and water consumption data for one year.
T difference between buffer discharging flow and return larger than 15 K for a minimum of 5 min e.g. Tstorage,upper <50 °C T difference between buffer discharging flow and DHW flow larger than 15 K for a minimum of 5 min Via T-sensors Via p-sensors
– Division of measured solar yield by corrected simulation result, if larger than 0.9 the guarantee has been fulfilled. The GRS approach improves the quality of a system and, if the monitoring continues after the first year, larger faults can be found. Nevertheless, a real automatic analysis is not included. Furthermore the reaction time to find a fault after it occurs can be long, since the analysis is not carried out on a short term basis. Also the uncertainties of the simulation can be large due to an inaccurate image of the system and the simulation software chosen. 3.4. Function control for small solar thermal systems without heat measurements (FUS) The ‘function control for small solar thermal systems without heat measurements’ was developed to provide a continuous function control for small solar thermal systems without heat measurements and with hardware costs below €100 (1999 prices) (Altgeld and Mahler, 1999). The method was included in two prototypes of controllers that should provide an error message if necessary. 57 possible malfunctions in the operation of solar thermal systems were identified. These were ranked according to 3 criteria: frequency of occurrence, effect on efficiency and system yield and likeliness of discovery without the fault routine. Problems with auxiliary heating have been excluded, while faults that cause problems with system security or comfort or which are easily recognized by the routine were included. In Table 2, the resulting list of faults and the algorithms to detect and identify the faults are shown. The method was, for test purposes, partially implemented in controllers of the companies Esaa and Wagner. Several faults were recognized, but there were also false positives, detection of a fault while the system was functioning correctly. The approach stays cheap by using mainly sensors that are used for the control, however, with analysis and pressure measurements, the price may be
1434
A.C. de Keizer et al. / Solar Energy 85 (2011) 1430–1439
Table 2 Fault algorithms in the FUS approach (derived from Altgeld and Mahler (1999) and Altgeld (1999)). Fault description
Result of fault
Algorithm (examples for studied system)
1. Collector connections interchanged 2. Collector T-sensor falsely positioned 3. Leakage of heat exchanger 4. Power outage 5. Incorrect controller software 6. False volume flow setting 7. Air in hydraulic circuit 8. DT setting is inappropriate 9. Valve in solar circuit closed 10. Cable breach between controller and pump 11. Defect input or output to the controller 12. Gravity brake is open 13. Fouling of gravity brake 14. Time switch is programmed wrongly 15. Impressed-current anode defect 16. Temperature sensor of collector defect or high 17. Temperature sensor in storage defect or high 18. Primary pressure in the expansion tank too high 19. Primary pressure expansion tank too low
Pump clocks Pump clocks System pressure too high System pressure too high DT too high DT too high DT too high DT too high DT too high DT too high DT too high + Pump on at night Pump on at night Pump on at night Pump on at night Storage tank corrodes Wrong DT => pump runs at wrong Wrong DT => pump runs at wrong System pressure at times too low System pressure at times too high
Pump running time < 10 s. Pump running time < 10 s. (Tcol = 20 °C) AND (psys = psys,nom + 2 bar) (Tcol = 20 °C) AND (psys = psys,nom + 2 bar) Pump on AND DT = DTnom + 15 K Pump on AND DT = DTnom + 15 K Pump on AND DT = DTnom + 15 K Pump on AND DT = DTnom + 15 K Pump on AND DT = DTnom + 15 K Pump on AND DT = DTnom + 15 K Pump on AND DT = DTnom + 15 K Pump on AND time between 22:00 and 6:00 Pump on AND time between 22:00 and 6:00 Pump on AND time between 22:00 and 6:00 Current measurement
slightly higher than anticipated. Although the method succeeds at detecting several faults in the lab, location of faults is limited, as can be seen in Table 2. Furthermore, there is no yield measurement, which could mean that large energy losses are not detected. 3.5. Spectral method (SPM) The spectral method is based on analyzing the transient temperature changes in the collector circuit after the pump is starting (Grossenbacher, 2003; Synetrum, 1998). The only extra temperature sensor is installed a few metres after the collector outlet. This temperature is measured every second, transformed to a dT/dt plot and subsequently to a spectral range via a discrete Fourier transformation. A fault free training phase results in a characteristic vector and an uncertainty boundary. A measured vector out of range shows a fault. The method was tested for two systems in Switzerland; with a temperature difference for starting the pump of 8 K and 30 K respectively (Grossenbacher, 2003). Dirty glass, or a reduction of collector performance can be detected if the loss of power is large enough, e.g. 40% reduction of collector performance. Furthermore, air in the collector or in the heat exchanger could be easily detected. A reduction of at least 20% in volume flow was recognized. Also a change of DT-settings with about 20% can be recognized for the low flow system. The method has the advantage of being cheap; only one extra temperature sensor is needed. In addition no parameterization is necessary and faults occurring over a longer time frame can be recognized. A disadvantage is that a fault free training phase is needed; a minimum time frame of 300 days and 240 start ups of the pump are suggested (Grossenbacher, 2003). Furthermore not all faults can be recognized and logging data on a secondly time basis can be complex. A
limited identification of grouped faults is theoretically possible. 3.6. Input–Output Controller (IOC) In the Input–Output Controller device a fault detection method is integrated, that compares calculated with measured energy yields. It has been commercially available since 2007 (www.resol.de, 2008; Pa¨risch and Vanoli, 2007a). The method monitors the energy yields in the solar circuit; a second approach in which also the buffer storage discharging is included was developed, but is not yet commercially available. The IOC compares the daily measured and expected energy yields of the solar loop. The measured energy input is based on the volume flow of solar or secondary loop and the flow and return temperatures of the corresponding loop. The energy gains are calculated with an approach developed by Pa¨risch and Vanoli (2007b), based on irradiation, ambient temperature and temperature of the heat sink. The daily heat gain of the collector circuit (Qcc) is calculated with Eq. (1) (Pa¨risch and Vanoli, 2007b). This is based on the daily irradiation HDay (kWh) and the part of the irradiation occurring at a time when the collector is not at a sufficient temperature to charge the storage (Hinsuf). All of these are calculated based on the expected collector circuit temperature, which is calculated with differential equations for different forms of operation (no pump running, operation, pause due to irradiance reduction, or stagnation) based on planning data, characteristics of collectors, pipe lines and pumps and the measured values of irradiance, solar last temperature and ambient temperature. Qcc ¼ g0 K h;eff ðH Day H insuf Þ QV;th QV;Kap
ð1Þ
The standard uncertainty (r) of the IOC-procedure, including measurements and yield calculation, is about 7% of the
A.C. de Keizer et al. / Solar Energy 85 (2011) 1430–1439
calculated energy gain (Pa¨risch and Vanoli, 2007b). If the difference between measured and simulated yield is larger than 20% (3 r) a fault is reported. This leads to 99% reliability for a correct fault prediction. The uncertainty margins are larger below a calculated yield of 1.5 kWh/m2d. There is a fault tree to distinguish if the fault occurred inside or outside the solar loop, and if it is for example the control or the solar station which causes the problems. The IOC is sold for 1190 € inclusive temperature and irradiance sensors, but without flow measurements. To be able to check the performance from home an extra data logger is necessary (Pa¨risch and Vanoli, 2007b). 3.7. Fault diagnostic system based on artificial neural networks (ANN) Kalogirou et al. (2008) presents a fault detection method for solar thermal systems based on simulations with artificial neural networks. Artificial neural networks are a mathematical method based on biological neural networks that can learn to predict results from examples. They are supposedly faster than simulations with e.g. TRNSYS and are able to deal with incomplete data and non-linear problems. ANN’s behave as a black box (Kalogirou et al., 1999, 2008). The method consists of three steps. In the prediction module, artificial neural networks are trained with data of a TRNSYS model that is assumed to represent a faultfree system. The model is trained so that four temperature values (collector in- and output and storage in- and output) can be predicted for different environmental conditions. The input consists of weather data (global and beam radiation, ambient temperature, incidence angle, wind speed, relative humidity, flow (yes/no)), together with one of the other measured temperatures. In the second step residual values, which characterize e.g. the actual temperature increase against the predicted temperature increase in the collector are calculated. In the last step a diagnosis module is run. Five consecutive residual values outside the boundaries are classified as a fault. Furthermore, if the average of 10 consecutive residual values is outside the boundaries this is reported as a fault. Several faults were simulated with TRNSYS and detected by the procedure, e.g. for a 5% decrease of F0 or an increase of 10% of the U-value of the pipes (Kalogirou et al., 2008). It needs to be studied how this method functions for real systems, which do have measurement uncertainties and may have different faulty behaviour than the ones modelled in TRNSYS. Furthermore the ANN was trained with TRNSYS and not yet with real data. 3.8. Method developed at Kassel University (KU) At Kassel University, a fault detection method was developed, that combines a static function control with dynamic simulation based fault detection (Wiese, 2006, Wiese et al., 2007). The method consists of three steps:
1435
1. Check for missing data 2. Plausibility check 3. Comparison of simulations and measurements In the first step the data integrity is checked. A minimum of 95% of the data points should be available to continue the fault detection. In a second step a plausibility check is carried out, which checks whether individual components function correctly. Table 3 shows the inclusion of automatic detection of faults in the second step. The approach was tested for several large German domestic hot water systems. The third step consists of a daily comparison between measured and simulated energy yields at the first and second heat exchanger. The simulation environment TRNSYS is used. These analyses include uncertainty estimates for both, the measured as well as the simulated energy yields. Several faults could be detected for large enough energy losses. However, so far faults cannot be identified in the simulation based step. 3.9. Concluding remarks After several methods have been introduced in the previous paragraphs, some more method properties are investigated and listed in Table 4. The time scale of the data logging is important, to know what type of faults can be derived from the data. Although a shorter time step is preferable for fault detection, it also leads to higher costs. The optimal time depends on a trade off between those two. The time scale of the analysis is important since it indicates how
Table 3 Fault detection with plausibility checks (Wiese, 2006). Verification of functioning of controller Breakdown of controller Breakdown of sensor Inaccurate sensors False control criteria Inappropriate control scheme Position collector T sensor Inappropriate T/DT settings Incorrect sensor position Breakdown of gravity brake
Yes ± ± ±/yes ± ±/yes ± ± Yes
Verification of volume flow Volume flow too small Too small pump, wrong pump step Air in hydraulic circuit Fouling of hydraulic circuit Primary pressure in hydraulic circuit too low Volume flow too high
Yes Yes Yes Yes Yes
Verification of heat exchanger performance Dimensioned too small Fouling Hydraulics wrongly connected
Noa ± ±
a Fault detection approaches are usually not designed to identify planning faults.
1436
A.C. de Keizer et al. / Solar Energy 85 (2011) 1430–1439
Table 4 Characteristics of fault detection methods. Characteristics a
Time scale of data logging Time scale of analysis Calculation of energy yield Scale of the tested systems (collector area in m2) Stage of developmentc Level of automationd a
MM
OPT
GRS
FUS
SPM
IOC
ANN
KU
Var Var No Var
15 min No 30–250
Var Mon or yr Yes Very large
<1 min ? No 5
1s Sec No 7–16
Min Day Yes 2–455
Hour? Hour Yes 40b
1 min 10 min & day Yes 88–400
++
++ +
++ +
++ ++
++
++ ++
++
+ ++
Time scale: Var = variable; Sec = second; Min = minute; Hour; Day = day; Mon = month; Yr = year. TRNSYS simulation. ++ well developed/can be used directly, + advanced stage of development, early R&D. not automated, + partly automated, ++ fully automated.
b c d
long it takes before a fault gets noticed. With a calculation of energy yield, a comparison with measurement data can show if the total energy gain of the solar loop or the solar heating system is sufficient. The stage of development shows if the method is commercially available, in an advanced stage of development or still in an early research stage. Finally the level of automation indicates if the method is automated, or if it is mostly manual work. The required measurement equipment of each approach is listed in Table 5.
4. Methodology for comparison 4.1. Introduction A partial multi-criteria analysis (MCA) will be used to evaluate and compare methods that can be used for detection and identification of defects or malfunctions during the operation of solar thermal systems. An MCA is often used for supporting policy decisions and evaluating different alternatives. One of the advantages of an MCA is that
Table 5 Measurement equipment required for fault detection methods. *
Sensor
Fus
Opt
SP M
T T T T T T T T T T T T T P VF VF VF VF VF
Collector output Collector input Ca. 1 metre behind collector output Flow solar loop – primary Return solar loop – primary Flow and return solar loop – secondary Flow and return auxiliary heating loop Solar tank output Auxiliary tank output Flow and return buffer storage discharge loop Distribution net flow and return Storage Cold water Pressure primary solar loop Primary solar loop Secondary solar loop Cold water Discharge loop HX – DHW storage Circulation Flow (yes/no) solar primary loop Ambient Tilted Global on a horizontal surface Relative humidity Wind speed Solar loop (secondary) Auxiliary heating Net Dwelling
x
x
x
x x
x x x x
T G G
HF HF HF HF * a
IOC 1
AN N
GR S Fr
GR S Dea
x x
KU x
x x
x x
x x
x x x x
x x x
x n.a. x x
x n.a. x x
x x x x x
x x x
x x
x x
x x
Low
n.a. x x n.a.
Max TSL x
x
x x
x
n.a.
x
x
x x
x x x x x x
x
x x x x
T = Temperature, P = Pressure, VF = volume flow, G = irradiation, HF = heat flow (VF and flow and return temperature included in HF metre). Refers to extensive measurement equipment in the Solarthermie 2000 project, only part of that is shown here.
A.C. de Keizer et al. / Solar Energy 85 (2011) 1430–1439 Table 6 Criteria used for evaluation of the methods. Criteria
Evaluation
Explanation of evaluation
Automated fault detection included? Automated fault identification included? Accuracy of fault detection
y/n
Yes/no
-/0/+/++
No/somehow/yes/very good
?
Not enough information available for analysis Detects some faults Good Very good
Accuracy of fault identification
Costs (operational and hardware)
Monitored part of solar heating system
Time until detection of fault
0 + ++ n.a.
not applicable
0 ++ -
Good Very good Expensive
0 + ++ var
Middle cost range Cheaper Cheap Variable
sl bs w-aux
Solar loop Including buffer storage loop Whole system minus auxiliary heating loop In the range of days, weeks, month(s), year(s)
d/w/m/y
the criteria on which basis the comparison is made are explicit. There are several MCA methods that weigh the different scores and combine these to a result score. In this paper a relatively simple approach will be used. The four steps of the MCA that are conducted are the following (Dodgson et al., 2000): 1. Definition of the aims of the MCA, and the aims of decision makers and other key players 2. Identification of the methods/options for achieving the objectives 3. Identification of the criteria to be used to compare options 4. Description of the expected performance of each option against the criteria (performance matrix) The objective of the multi-criteria analysis is to evaluate different fault detection methods based on the performance of the method regarding e.g. effectiveness, flexibility and costs. The fault detection methods have been identified in an extensive literature review and are described in Section 3. The methods are valid for monitoring solar thermal systems for the entire life time of the system. The following eight approaches have been discussed in this paper: Manual monitoring (MM) Optisol (OPT) Guaranteed Solar Results (GRS)
1437
Function control for small solar thermal systems without heat measurements (FUS) Spectral method (SPM) Input–Output Controller (IOC) Fault detection based on simulations with Artificial Neural Networks (ANN) Method developed at Kassel University (KU) 4.2. Criteria to analyse performance The terms of monitoring, fault detection and identification are often intertwined and go with different definitions in different papers, e.g. with fault detection sometimes only monitoring without analysis is meant. The terms will be used here in the following context. Monitoring is defined as data logging of a variable amount of measurement data, however this data is not automatically analysed, so it does not automatically lead to a fault declaration. In a fault detection method, measurement data is automatically analysed and in case of a malfunction a fault indication follows. Fault identification or localisation requires the identification of the type of fault. This will make reparation much easier. Relevant criteria to evaluate the performance of the fault detection and monitoring methods have been identified and are shown in Table 6. The possible evaluations for these criteria are shown in the second column. The first criterion indicates if automated fault detection is included (yes/no). The following criterion shows if also automated fault identification is included in the method. The first criterion can clearly be analysed, for the second criterion there is a gliding scale between no () somehow (0), yes (+) or a very good (++) and detailed identification approach. Nevertheless the quality and accuracy of detection and identification can be very different. This is what the third and fourth criteria address. These roughly refer to how many fault types can be found and how large an energy loss or other fault result needs to be before it is detected. This criterion includes the uncertainty in the analysis that results from measurement and simulation uncertainties. However, this can only be qualitatively analysed, since methods have been developed and applied to very different systems, and publications often do not provide very accurate descriptions of effectiveness and accuracies of the methods. Therefore a very detailed analysis cannot be carried out. Furthermore, it is important to address which part of the solar heating system is included, since faults also occur outside the solar loop. Costs are an important factor as well. The last criterion reflects on how long it takes before a fault is identified. This one gives only a rough indication, since it depends on the effect of the found fault, however also on e.g. the time of analysis. 5. Results The performance matrix in Table 7 shows the performance of the fault detection methods based on the different criteria.
1438
A.C. de Keizer et al. / Solar Energy 85 (2011) 1430–1439
Table 7 Performance matrix. Criteria
MM
OPT
GRS
FUS
SPM
IOC
ANN
KU
Automatic fault detection included? Accuracy of fault detection Automatic fault identification included? Accuracy of fault identification Costs Monitored part of solar heating system Time until detection of fault
No ++e ++e var Var d/w
n/y ++e 0 ++e var Whole d/w
No 0 n.a. 10 k€b aux y
Yes 0 + 0 ++100 € sl d
Yes ? n.a. +? sl d/w
Yes + 0 0 +1190 €a sl, (bsd) d/w
Yes ? n.a. +? sl d/w
Yes + + 0 0 240–960 €c aux d/w
a
IOC: hardware and temperature sensors. Costs for extensive measurement equipment, including one year monitoring in ‘German’ GRS. c Costs per year for 20 year monitoring, including hardware and at least 30 monitoring systems sold. The main costs are expected for maintenance and improvement of software (between € 15 and 50 per month) (Wiese, 2006). d Only in IOC 2 (not commercially available yet). e Includes manual monitoring. b
Manual monitoring is very effective, but since not automated, very expensive and not applicable for a large number of systems. It is, however, easier to adapt to the extensive variation in hydraulics and systems. The FUS method is very low cost and, therefore, can be applied well to small systems. FUS detects several faults at reasonable costs, but does not make an energy balance. The spectral method and the method with simulations based on artificial neural networks are not developed far enough to know how the results will be. Both methods also need a long training phase in which a system assumedly runs fault free. The IOC is the first method which could result in the commercial implementation of a monitoring and fault detection method for larger solar thermal systems. It has been tested and is commercially available at a reasonable cost. However, the commercial available version only applies to the solar loop and not to the whole solar heating system. The method developed at Kassel University is still under development, but could also provide an automatic monitoring solution for large systems. It includes more sensors and a larger part of the system than the IOC approach, and can therefore also analyse individual components. So far the auxiliary heating system is not included, but this is planned. At the moment, none of the above described approaches takes the auxiliary heating system into account, which is also an important source of errors. 6. Discussion, conclusion and recommendations In this paper an overview and assessment of fault detection methods for solar thermal systems has been presented. A multi-criteria analysis has been used to compare the different approaches. Most methods are still in development and very detailed information was not available. The only commercially available automated method, the Input–Output Controller, should detect faults causing more than 20% energy loss in the solar loop. This method is recommended for middle to large sized systems. The FUS approach without a heat flow meter is cheap and only relies on few sensors to check the functioning of several
components in the solar loop with algorithms. The approach developed at Kassel University checks the functioning of a solar plant both with plausibility checks and with energy balances based on simulations. This method includes a larger part of the solar system, requires more measurement equipment and is, due to cost reasons, more suitable for large systems. Currently methods are being further developed. That is useful, since practical experience on effectiveness has to be gained and also costs should be reasonable. Attention should be paid to what part of the system is monitored; since the auxiliary heating also causes problems, it would be wise to include this in monitoring methodology. Further research and practical application of these fault detection methods should improve the knowledge on their costs and effectiveness and their applicability for certain requirements and system hydraulics. Acknowledgement The authors gratefully acknowledge the financial support provided by the Marie Curie early stage Research Training Network ‘Advanced solar heating and cooling for buildings – SOLNET’ that is funded by the European Commission under Contract MEST-CT-2005-020498 of the Sixth Framework Programme. References Altgeld, H., 1999. Funktionskontrollen bei kleinen thermischen Solaranlagen ohne Wa¨rmemengenmessung. Hochschule fu¨r Technik und Wirtschaft des Saarlandes, Saarbruecken, Germany.
A.C. de Keizer et al. / Solar Energy 85 (2011) 1430–1439 Fink, C., Riva, R., Pertl, M., Wagner, W., 2006. OPTISOL - Messtechnisch begleitete Demonstrationsprojekte fu¨r optimierte und standardisierte Solarsysteme im Mehrfamilienwohnbau. AEE - Institut fu¨r Nachhaltige Technologien, Gleisdorf, Austria.
1439
Peuser, F., Remmers, K., Schnauss, M., 2002. Solar Thermal Systems. Solar Praxis AG, Germany. Peuser, F.A., Croy, R., Mies, M., Rehmann, U., Wirth, H.P., 2008. Solarthermie 2000, Teilprogramm 2 und Solarthermie 2000plus, Wissenschaftlich-technische Programmbegleitung und Messprogramm (Phase 3), Abschlussbericht zum Projekt 032 9601 L. ZfS - Rationelle Energietechnik GmbH, Hilden, Germany. Qualita¨tssteigerung und Qualita¨tssicherung großer thermischer Solaranlagen in geba¨udeu¨bergreifenden Mikronetzen und im großvolumigen Wohnbau,