Fault detection method for heat source system using comparison between simulation and measured data Y.Tai 1, D.Sumiyoshi 2, Y.Akashi 3, Y.Kuwahara 4,

Transcription

1 Fault detection method for heat source system using comparison between simulation and measured data Y.Tai, D.Sumiyoshi, Y.Akashi 3, Y.Kuwahara 4, K.Ueda 5, S.Nikaido 5, K.Tateishi 5, M. Matsuo 5 Graduate School of Human-Environment Studies, Kyushu Univ. Associate Professor, Faculty of Human-Environment Studies, Kyushu Univ. 3 Professor, Department of Architecture, Graduate School of Engineering, The University of Tokyo 4 MTD Co., Ltd. 5 Mitsubishi Heavy Industries Ltd. Abstract If it is performed on optimal design when designing a heat source system, it is unable to continue to maintain the performance without appropriate operation assessment. In this study, we propose performance management methods of the heat source system using a simulation model, and the development of a technology capable of maintaining many buildings remotely as the purpose of this study. In this paper, the simulation model of a whole heat source system and equipment unit with different input values as intended for industrial facilities is developed. Therefore, a fault detection method using a simulation model is studied. In addition, fault detection method by comparison of the measured value is studied. As a result, it was found that there is a possibility to find an equipment faults and sensor errors by comparing the measured data and the results of unit model calculation and system model calculation. Also, Fault detection method by comparing the measured values can be very useful by combining with fault detection method using simulation. Keywords Heat source system, Simulation, Fault detection, Commissioning. Introduction If it is performed on optimal design when designing a heat source system, it is unable to continue to maintain the performance without appropriate operation assessment. It is important that an administrator with the analysis ability manage the heat source system properly in order to maintain its performance. However, it is difficult in terms of human resources and cost. In this study, we propose performance management methods of the heat source system using a simulation model, and the development of a technology capable of maintaining many buildings remotely as the purpose of this study. In this paper, the simulation model of a whole heat source system and equipment unit with different input values as intended for industrial facilities is developed. Therefore, a fault detection method using a simulation model is studied. To be Corresponding author yt..yt@gmail.com 364

2 specific, the case in which trouble occurs in the sensor and the flow rate sensor is anticipated, the fault detection methods while considering the effects of the two types of simulation model results and measured results is considered. In addition, Fault detection method by comparison of the measured value is studied. Accuracy of the sensor is verified by checking the magnitude relation of the sensor, and the accuracy of the other sensor is verified by checking the heat balance of the heat source equipment.. Abstract of Targeted System The target system was at the Sony Sendai Technology Center, an industrial building facility used for the production of recording media in Sendai City, Miyagi prefecture. The specifications of each device and an overview of the system are shown in Figure and Table. The target system has been running for 4 hours 365 days a year and administrators collect actual measurement data using the Building Energy Management System (BEMS). The collection interval is minute. This study s research was performed using actual data for a year from December to November 3. The number of units in operation of the refrigerator was controlled by the flow rate and the heat amount in the target system. In addition, this system is incorporated with control logics such as primary pump-type chilled water variable flow rate control, cooling water variable flow control, cooling water bypass valve control and cooling tower variable air volume control. Symbol Name Number Specification TR- TR- CP- CP- CP- CP- CP-3 Figure. System diagram of the target system Table. Equipment specifications of the target system Inverter turbo refrigerator primary pump secondarypump 3 Refrigeration capacity:758kw(5usrt) (Input/Output):5 /7 flow rate:88 m3 /h (Input/Output):3 /37 flow rate:355m3/h INV control Rated COP: 5.9 Flow rate:88m 3 /h INV control Power consumption[kw] Flow rate:88m 3 /h 8.5 INV control CDP- Flow rate:355m 3 /h CDP- pump INV control 45 CT- Cooling capacity:66kw Flow rate:355m3 /h Cooling tower CT- INV control 7.5 [fan] 3. Development of a simulation model A simulation model of the whole target system was developed by combining a control logic model (such as a refrigerator volume control) with the equipment unit model (inverter turbo refrigerator, cooling tower) of the target system. A flow diagram

3 [ ] Power consumption[kw] [ ] of the system model is shown in Figure, and the input and output values of each model in Table. START Input the initial condition Read the input data Check the input data Control the number of the refrigerator Control the pump variable flow Calculate the flow rate of chilled water Time loop Calculate the of chilled water Determine of the cooling water bypass valve opening angle Calculate the rafrigerator COP Calculate the flow rate of Control the Cooling tower variable air volume Calculate the of cooling water Compiled & Output Figure. Flow diagram of the system model End Refrigerator unit model Cooling tower unit model System model Input value inlet Flow rate of chilled water inlet Flow rate of cooling water Cooling tower inlet Flow rate of cooling water (Air flow of cooling tower) Outside air conditions Amount of heat load Flow rate of heat load Output value outlet Amount of heat processing Refrigerator COP Power consumption etc. Cooling tower outlet Power consumption etc. Temperature of each point Flow rate of each point refrigerator COP Power consumption of each equipment etc. 3. Development of a simulation model of the equipment unit The inverter turbo refrigerator simulation model calculated the COP based on the performance curve of the refrigerator, and calculated the cooling water outlet from the heat balance of the cooling water and the chilled water. A comparison of the unit simulation results and measured data of the refrigerator power consumption is shown in Figure 3, and a comparison of the cooling water outlet in Figure 4. The refrigerator unit model captured the measured value with sufficient accuracy. The cooling tower unit model exemplifies the open type cooling tower, and calculates the amount of heat processing as well as the cooling water outlet of the cooling tower. A comparison of the simulation results and the measured data of the cooling water outlet are shown in Figure 5. The two values are almost identical, and the cooling tower unit model contains sufficient accuracy. TR- Power consumption(measured data) TR- Power consumption(measured data) TR- Power consumption(unit model) TR- Power consumption(unit model) Table. Input value and output value of each model 4 TR- outlet (Measured data) TR- outlet (Measured data) TR- outlet (Unit model) TR- outlet (Unit model) Figure3. Verification result of the refrigerator model (refrigerator power consumption) Figure4. Verification result of the refrigerator model ( outlet ) 4 Cooling tower outlet (Measured data) Cooling tower outlet (Unit model) Figure5. Verification result of the cooling tower model (Cooling tower outlet ) 366

4 Power consumption[kw] Temperature[ ] Floe rate[ m3 /h] 3. Development of a simulation model of whole system The input value of the system model is the amount of heat load and outside air conditions, as well as the flow rate of the heat load. This model was created by combining the control logic and unit model of each device; for this reason, it was possible to calculate the and flow rate of each point in the target system. A comparison result of the integrated value and the average value of the measured values and the system model is shown in Table 3. Furthermore, the comparison between the system model and measured data of the cooling water is shown in Figure 6 and the comparison of the flow rate of chilled water and cooling water in Figure 7. The comparison between the refrigerator power consumption is shown in Figure 8, and the comparison of each heat load amount of the refrigerator in Figure 9. The accuracy of the cooling water and the flow rate of chilled water are very good. The results of the flow rate of cooling water showed a different value than the measured value because the control logic of cooling water could not function due to the pollution of cooling water. The power consumption and each heat load amount of the refrigerator were calculated with high accuracy. Averaged TR- Table3. Comparison of the results the system model and measured data (average values) inlet outlet inlet outlet Flow rate of Flow rate of Amount of heat load Power Ccnsumption [-] [ ] [ ] [ ] [ ] [ m3 /h] [ m3 /h] [kw] [kw] Measured data System model Averaged TR- inlet outlet inlet outlet Flow rate of Flow rate of Amount of heat load Power Ccnsumption [-] [ ] [ ] [ ] [ ] [ m3 /h] [ m3 /h] [kw] [kw] Measured data System model inlet (Measured data) Flow rate of chilled water(measured data) outlet (Measured data) Flow rate of cooling water(measured data) inlet (System model) Flow rate of chilled water(system model) 4 outlet (System model) 4 Flow rate of cooling water(system model) Figure6. Verification result of the system model ( ) TR- Power consumption(measured data) TR- Power consumption(measured data) TR- Power consumption(system model) TR- Power consumption(system model) Figure7. Verification result of the system model (Flow rate of chilled water and cooling water) Figure8. Verification result of the system model (Power consumption) Figure9. Verification result of the system model (Amount of heat processing) 367

5 Temperature[ ] 4. Discussion on the Fault Detection Method With the comparison of the results of the kinds of simulation models developed in the previous sections and measured data, the cases where a fault occurred in the sensor were discussed to explore the possibility of fault detection. Because the models were developed based on the data when all instruments are normally operating, if any deviation between the result of the simulation and measured data is observed, some faults are considered. Certain relationships between some of the measured data are sometimes found. Discussion was made on the fault detection method by these relationships being checked. 4. Study of fault detection method using Simulation The sensor and flow rate sensor incorporated in the targeted system are shown in Table 4. The sensors incorporated in the targeted system can be classified into four groups in accordance with whether the sensors are for input or output and whether they are of the unit model or system model. In this paper, a discussion was made on the case where a measurement error was presumed to occur in the cooling water inlet for a refrigerator. The measured data when the cooling water inlet was constantly maintained at + C was prepared, with which the simulations for the system model and unit model for the refrigerator were performed. Although the cooling water inlet is the input value of the unit model for refrigerators, but it is not the input value of the system model. Therefore, it is considered that it affects the result of the unit model result only. The comparison of the measured data of the cooling water outlet and the results of two kinds of simulations are shown in Fig.. As expected, only the value of the unit model elevated. It was found that, in this case, only the system model shows different input values and the unit model only shows different output values. A summary of the results of the discussion above is shown in Fig.. It is possible to identify the spot where a sensor error occurred by confirming the spot where measured data and the simulation results deviate from each other. Whole model input value Whole model output value Table4. Temperature and flow rate sensor in the target system Single model input model Single model output model Not related single model - - Flow rate of heat load entrance entrance Cooling tower entrance Flow rate of chilled water Flow rate of cooling water exit exit Cooling tower exit Return header Supply header Measured data System model Unit model 4 Averaged values Measured data 3. System model.5 Unit model Figure. Study results of assuming the sensor error of cooling water inlet ( outlet ) 368

6 Figure. Study results of the fault detection method 4. Fault Detection Method by the Comparison of Measured data It can be considered about how to detect a fault by the comparison of measured data without using a simulation model. Specifically, it is to understand the relationships between sensors by each sensor of and flow rate and confirm the relationships every hour, and if there is a spot where the measured data do not satisfy the relationship, it can be detected as fault. The contents for discussion targeting sensors are shown as examples. For example, there are four sensors relating to chilled water in the targeted system. The following natures of the relationships can be considered:. inlet and outlet difference [K] refrigerator amount of heat load [Wh] / (flow rate of chilled water [m³ / h] * specific gravity of water [kg / m³] * water specific heat [Wh / kg k]).((tr- chilled water outlet ) * (TR- flow rate of chilled water) + (TR- chilled water outlet ) * (TR- flow rate of chilled water) / ((TR- flow rate of chilled water) + (TR- flow rate of chilled water))) [ ] supply header [ ] 3. Supply header and return header difference [ ] amount of heat load [kw] / (flow rate of heat load[m³ / h] * specific gravity of water [kg / m³] * water specific heat [Wh / kg k]) 4.((Return header [ ]) * (flow rate of heat load [m³ / h]) + (Supply header [ ]) * (bypass flow [m³ / h]) / ((flow rate of heat load[m³ / h]) + (bypass flow [m³ / h]))) chilled water inlet [ ] These natures of relationships are confirmed every hour, and if any measured data does not satisfy the nature of the relationship, it is detected as a fault. However, in the case where the number of sensors is small or the change caused by the heat loss of the plumbing line is large, it is difficult to detect a fault by this method only, therefore, it is necessary to combine this method with the fault detection method using the simulation of the previous section. 369

7 Temperature difference[k] Measured data/calculated data 4.3 Discussion on the Deviation between Measured Data and the Model Result When the method that was proposed in this chapter is put into practical use, it is necessary to understand the deviation amount between the measured data and the model result, and when a certain deviation amount is recognized, it is detected as a fault. The procedure is discussed targeting cooling water. First, the deviation between the measured data of TR- cooling water and the model result, which is evaluated by absolute value, is shown in Fig. and a result evaluated by relative value is shown in Fig. 3. There are characteristics that the deviation amount changes significantly by the evaluation by absolute value depending on the accuracy of the model and that the measure of deviation amount changes depending on the spots evaluated. On the other hand, for the relative amount, although the deviation amount of different evaluation spots can be evaluated with the same measure, there is a characteristic that increases and decreases in the deviation amount caused by faults and others are visually absorbed. The detection by absolute value is effective for instantaneous errors such as sensor errors and the detection by relative value is effective for long-term faults such as the deterioration of instruments. When a long-term evaluation is made by relative value, the moving average is useful for the quantitative evaluation of the deviation amount and for removing the effects of seasons. However, the result of the moving average is greatly different depending upon the averaging period. The relative value of Fig. 3 evaluated by a one-day moving average is shown in Fig. 4, and that evaluated by a one-month moving average is shown in Fig. 5. The one-month moving average can make evaluation so that the errors caused by the degree of accuracy of the model and faults can be more easily recognized visually than in the one-day moving average. However, it is difficult to judge instantaneous errors. An effective evaluation can be made by changing the averaging period depending upon the targets of detection and the periods of data analysis. Not only for evaluations in time series, but also it is effective to take up a specific time and show diagrammatically the relationship between the neighboring sensors (hourly evaluation). The analytical result of the measured data of the chilled water and cooling water of TR- and hourly evaluation by the model are shown n Fig. 6. The result when a fault in the cooling water inlet was presumed is shown in Fig. 7. By showing the s in the order of points in the plumbing line, it is possible to understand the deviation amount of each point and the magnitude relationship between the neighboring points. inlet outlet inlet outlet Figure. Evaluation results of divergence amount by Absolute value Figure3. Evaluation results of divergence amount by Relative value 37

8 Temperature[ ] Temperarture[ ] Measured data/calculated data Measured data/caluculated data cooling water inlet cooling water outlet inlet outlet Figure4. The evaluation result by the one-day moving average Figure5. The evaluation result by the one-month moving average Measured data Unit model Return header inlet outlet Surpply header System model Comparison of measured data Figure6. Analysis results of the hourly evaluation (No fault) Cooling tower outlet inlet outlet Cooling tower inlet Measured data Unit model Return header inlet outlet Surpply header System model Comparison of measured data Figure7. Analysis results of the hourly evaluation (Fault occurring) Cooling tower outlet inlet outlet Cooling tower inlet 5. Conclusion In this report, two different simulation models with different input values were constructed targeting a system in an industrial facility in Sendai City, Miyagi Prefecture and how the measurement errors of the sensors affect each simulation result was discussed. In this discussion, it was found out that it is possible to find out sensor errors and determine error spots by comparing measured data, unit model results and system model results. At that time, the sensor errors of the targeted system can be classified into four cases from the perspective of input and output values of simulation. Furthermore, it was also discussed how to consider the deviation between the measured data and model value when the proposed fault detection method is practically used. It is important to use different evaluation methods for the evaluation of the deviation amount depending upon targets. Hourly evaluation is needed in practical use. The proposed method will be further developed and the development of a tool to automatically detect faults from long-term measured data also will be implemented in the future. References Nikaido S., Ueda K. et al., Development of performance evaluation method for optimally controlled heat source system -Part5, Performance evaluation by using simulation data-, Technical papers of annual meeting, the Society of Heating, Air-Conditioning and Sanitary Engineers of Japan,4,B-48, (submitted to) 37