By using the raw data in the balancing equations (auxiliary conditions), broad errors of the measurement values or of the modeling can be identified. Contradictions can be discovered. For this, it is necessary that the strict adherence to the balancing equations is removed. A deviation for the balancing equations is permitted with a very broadly selected confidence interval.
By selecting the validation mode "Raw data analysis" in the menu Extras à Model Options, tab Validation, Validation method, the prerequisites are fulfilled automatically.
This will be illustrated by taking a very simple example.
In most practical cases, all the measurement values around a component are not completely occupied, in order that the balancing equations can be checked in isolation. This is illustrated by the example from VDI 2048 sheet 1 appendix A. To discover the contradictions, all measurement values are used as default values. This concerns here the measurement values MSP2, MHK and MHDNK. The simulation then shows the components, for which the balancing equations are violated.
Ebsilon indicates a mixing and a branching, whereby no measurement values are present around a branching. To register numerically the deviations including the confidence intervals, value fields are introduced at the corresponding components. The following figure shows the results of the calculation in the raw data analysis mode.
The broad confidence intervals for the calculated deviations are conspicuous. This is explained by the weak confidence of the balancing equations (broad confidence interval of the deviations). The node components represent only a weak coupling.
One gets the following result, when the confidence is improved by a factor of 10:
The violations of the balancing equations including the confidence intervals become less. However, differences between the measurement values and the result values can still be detected. Only the previous setting is available to the user.
Another example deals with the balancing violations in heat exchangers and steam turbines. A simulation with all measurement values as default shows balancing violations for the pre-heaters HD-VW-1 and ND-VW-3 as well as for the turbine disc ND-DT-1. The corresponding deviation value fields are created, and in the validation mode raw data analysis the balancing violations including the confidence intervals are calculated on the connection pipes. The following figure shows the result. In case of heat-exchangers, the violation of the energy balance is mainly indicated between the primary and the secondary side, while the balancing violation for the turbine follows from the difference in the outlet enthalpy on the lines 2 and 3. The specific enthalpy on line 2 is defined by ETAI , the specific enthalpy on line 3 by the measurement value default of pressure and temperature 41 and 42.
The specified ETAI of 0.8 must be corrected to 0.88 for removing the deviation. This can be determined easily in the identification mode for the turbine disc.
The calculation of the deviation value fields for the pre-heater has the following syntax:
{(HDVW52_1.qt-HDVW52_1.q21)/1000;%6.1f} // deviation in MW (divisor 1000)
± {sqrt(L3.m*L3.m*L3.dh*L3.dh+L3.h*L3.h*L3.dm*L3.dm+ // Confidence interval from the confidence intervals for the variables
L5.m*L5.m*L3.dh*L5.dh+L5.h*L5.h*L5.dm*L5.dm+ // Flow rate and specific enthalpy on the connection lines
L4.m*L4.m*L4.dh*L4.dh+L4.h*L4.h*L4.dm*L4.dm+ // of the component
L2.m*L2.m*L2.dh*L2.dh+L2.h*L2.h*L2.dm*L2.dm+
L1.m*L1.m*L1.dh*L1.dh+L1.h*L1.h*L1.dm*L1.dm)/1000;%6.1f} MW
This calculation syntax can be applied to other heat exchangers by changing the pipe identifications. Starting from release 5.03, the names in general can be specified through $._1, $._2 etc. The compiler then replaces these with the actual names.
If a validation calculation results in a Chi^2-test ratio (relative mean square error divided by the value of the Fisher-distribution) of greater than 1, which is a violation of the 95%-quantile of the F-distribution as per the condition 140 of the guideline VDI 2048 sheet 1, or if the measurement values show too high relative deviation as per equation 141 VDI 2048 sheet 1, then either the confidence intervals are not set suitably, or else modeling errors are present. If the latter can be excluded, then the analysis is done as per equation 140 and 141 VDI 2048 sheet 1 under consideration of the correlations as per equation 129 VDI 2048 sheet 1 and corresponding to the instructions for error analysis given in sheet 2 of the guideline section 7.7.2.
The measurement values that show a higher deviation as per the equation 141, and the measurement values with a high correlation to them must be given an improved estimation in view of their confidence intervals. This must be done by the user himself by using his knowledge about the plant and must be repeated in iterations, till the Chi^2-test ratio reaches a value of 1 or better. Here lies also the connection point for a state-oriented maintenance of the measuring process. By using an EbsScript program, a list of the measurement values with a too high relative deviation between the validated value and the measurement value can be generated. Similarly, the Excel validation list (see the retrofit-example) can also be used.One can simplify the work by using an EbsScript by applying the propagation first to individual measurement values (especially high exceptions) and then checking, whether the others are not already pushed in a valid range because of that. If a sufficiently small Chi^2-test ratio (relative mean error square divided by the value of the Fisher-distribution), one can investigate, whether certain confidence intervals of measurement values, which affect the auxiliary conditions, have been set as too broad. After a proper estimation, they can be set as more narrow, as far as the Chi^2-test ratio does not change radically and remains within the limiting conditions. This investigation is conducted systematically with an EbsScript program and results in a recommendation list for narrower confidence intervals. Changes can be made after checking, which confidence intervals were set too broadly.