For the purpose of this research, it is essential to have the following two components: (1) a test-case simulation, as shown in
Section 10, and (2) a scaled case simulation based on the initial and boundary conditions determined in
Section 11. In essence, the scaling process is recursive, to converge to the optimal set of scaling equations. The scaling-ratio determination shown in
Section 11 is, by definition, a preliminary scaling analysis solely based on feedback from the test-case data. A scaled case simulation using the preliminary scaling analysis acts as a reference to evaluate scaling performance. To investigate the scaling conclusions determined in
Section 8 and
Section 11,
Section 12.1 and
Section 12.2 provide visual evidence for TEDS and HTSE. For the measurement of scaling distortions, the general DSS local separations are calculated using Equation (
107):
where
is a sign adjuster to ensure the sign within the square-root is always positive. The subscript
denotes the transient temporal location of interest (recall, the subscripts
M and
P are for model and prototype). It should be noted that the corresponding model time location for the given
k is not necessarily equivalent to the prototype. If the scaling case’s time ratio is not 1, then it is not equivalent. The coupled model and prototype time is determined by the time ratio (e.g., if
for prototype time at
(s), Equation (
107) should be evaluated at
(s) for the model). Each temporal location separation provides insights on the distribution of scaling distortions and acts as the qualitative criteria to evaluate scaling performance. Due to the preliminary nature of the scaling analysis, the quantitative criteria is absent and will be addressed in future work. The strategy for defining the quantitative criteria will be comparisons of DSS standard error among different applied scaling decisions. The DSS standard error is the global representation of the local scaling distortions and is shown in Equation (
108):
Essentially, the decreased DSS standard error will indicate improvements based on newly applied scaling decisions and vice versa.
12.1. TEDS Scaled Case
As stated in
Section 11, the scaling objective is to preserve the physical amount of heat stored at the end of the TTSS charging mode. Considering the increase in maximum heat generation to 260.719 kW from the Chromalox heater, decreased mass-flow rate, and loop cold-side maintained at 498 K (same as the test case in
Section 10), the charge- and discharge-mode simulations were conducted and are shown in
Figure 15a,b and
Figure 16a,b. Starting from the charge-mode comparison, the anticipated maximum temperature increase from 598 K to 668 K is reflected. Note that the starting temperature is 498 K, which is common between test and scaled cases. The increased temperature difference preserves the amount of TTSS heat storage as the scaled TTSS injected hot-fluid mass is reduced to approximately 80% of the test case. This explains the reduction in charge-outlet temperature increase as the TTSS is only a relative fraction of the test case, and not enough time has elapsed to transport hot-line Therminol-66 fluid to the TTSS charge-outlet line. Because the discharge inlet pulls Therminol-66 from the cold line, the temperature remains at 498 K.
The steps to calculate the DSS projected data are to (1) determine the time ratio, (2) determine the parameter-of-interest scaling ratio (
), and (3) use Equation (
8) to post-process test-case data. Reorganizing Equation (
8) and plugging in Equation (
1) gives the relation to calculate the projected data, given the prototypical system, for the test case:
Following the defined steps and using the preliminary scaling analysis, the projected case data were calculated for the TSS inlet, outlet, centerline, and wall-centerline temperatures; these are shown in
Figure 17a,b. From the first look, the charge-outlet, discharge-inlet, discharge-outlet, charge-centerline, discharge-centerline, wall-charge, and wall-discharge temperatures show different features. For the charge-centerline, charge-wall, and all discharge temperatures, the timing and magnitude are offset. For the charge-outlet temperature, the data geometry indicates a large temperature increase after 5500 s, starting from 403 K. This behavior can be explained by considering that all variables were normalized by the corresponding temperature values at the end of the charge mode (for the projected case around 9596.167 s and for the scaled case around 8477.034 s). The definition to be normalized is to adjust magnitudes that range around 1. The issue is the low temperature value relative to the reference value.
For example, consider the normalization at the initial charge time for both cases. For the scaled case, the reference temperatures are 668, 500, and 650 K for the charge inlet, charge outlet, and centerline temperatures, respectively, which correspond to normalized values of 1.000, 0.991, and 0.842. For the test case (recall Equation (
109)), the reference temperatures are 668, 582, and 581 K, which correspond to normalized values of 1.000, 0.841, and 0.755, respectively. The difference in normalized values is mainly due to the relative magnitude of cold-side temperatures. As hot-side temperatures were upscaled, the cold-side temperatures remained the same, causing lower-temperature normalized values to differ significantly. Although from the scaling standpoint, such decisions lead to scaling distortions, for operations, it is standard. If the cold-side temperatures were upscaled, this would suggest that the TTSS was precharged before the event and would not represent a case where the system is charged with the specified heat-storage amount. This adds complications because the parameter-of-interest scaling ratio for thermal terms is approximately 1 and, when applying Equation (
109), the normalized values from the test case are inherited, outputting higher temperatures than were observed from the scaled case.
As mentioned earlier, scaling analyses are recursive and are subject to modifications when less-compatible scaling decisions are made. In this case, the combination to normalize by the chosen reference value and leave the cold-line temperatures resulted in distorted projections. Furthermore, although it may not be obvious, the discharge projected case, in terms of data geometry, performed better than the charge projected case. This suggests that there should be a separate scaling for charge and discharge modes. To modify the scaling analysis to account for the low temperatures and varied scaling performance between both operational modes, it is suggested that the normalizing method be changed. Regardless of what reference value is used—whether it is a temperature value at some time or the temperature difference—the unscaled lower temperatures always trigger scaling distortions. One normalizing method that is capable of considering both maximum and minimum values is the minmax scaler used in machine learning. This is shown in Equation (
110).
Essentially, each data point is resized to range from 0 to 1 where 0 is at the minimum value and 1 is at the maximum value.
When applying the minmax scale,
Figure 18a,b show the difference when considering the minimum and maximum values, regardless of temporal location. Magnitudes are now adjusted to be roughly the same, and this is the correct normalization method for the type of assumptions and scaling decisions imposed. Because it was identified. based on
Figure 17a,b, that the necessity to separate scaling analyses among charge and discharge operations was emphasized, different time ratios were applied to achieve the results in
Figure 18a,b. One remaining shortcoming from the modified scaling analysis is the different transient trends observed at certain time intervals for the charge and discharge modes. For the charge mode, between 4000 and 6000 s and 6500 and 8100 s for centerline and charge-outlet temperatures were observed. For the discharge mode, between 9300 and 12,500 s and 10,300 and 12,200 s for discharge-outlet and centerline temperatures were found. The root cause of these distortions is inadequacies modeling the scaled conditions defined in
Section 11.
Figure 19a,b show the test- and scaled-case mass-flow rates. Immediately, it can be seen that the control sequences between the test and scaled cases are different. Regardless of a change to the timing by a time ratio of 0.855, the on, off, ramping up, and ramping down do not behave as planned. This results in different fluid-injection and ejection sequences, altering the data geometry and trends.
Another aspect that supports the conclusion of a mal-control sequence in the scaled Dymola simulation is the measured separation for inlet, outlet, fluid centerline, and wall centerline temperatures shown in
Figure 20a,b and
Figure 21a,b. It can be shown that relatively high separations are observed around the charge to discharge mode transition, which suggests the heater and mass flow rate controls were not optimal enough to represent the defined scaled-case. Moreover, the initial 1000 s of the charge inlet temperature exhibit unexpected distortions that are possibly rooted from the previously stated issue but will require further investigations to comprehend the source of the anomaly For future references, an improved understanding of the system controls in the Dymola TEDS model is required for better scaling validations concerning the control of heaters and the mass flow rate approaching and after the transition from charge to discharge mode. Overall, the charge mode temperatures show larger distributions of distortions than the discharge mode.
To emphasize the importance of the stated findings, consider an IES case where the stored heat is discharged for chemical processing. Chemical process plants are hard conditioned systems that can only operate with specific temperature, pressure, and other conditions dependent on the given process. If the heat from thermal energy systems is used, precise system controls are required to ensure operating conditions are constantly met, even when transitioning from one heat source to another. The findings based on the scaling analysis applied amplify the control issues found in the scaled Dymola case. Without the scaling analysis, there would be difficulties on justifying the bridge between one case with the other and determining which transients require improvements. By having the capability to project data from test cases, the ideal behavior based on the governing equations, scaling theory, and selected assumptions can be obtained and, to a certain degree, avoid costly retroactive facility changes as identified issues addressed during the pre-design phase. The recursive process of refining the scaling analysis will guarantee the validity, and the current analysis is the first iteration to achieve this goal.
12.2. HTSE
For HTSE, the scaling objective is to increase hydrogen production by adjusting the stack current, voltage, and average temperature. Considering the 1.28 factor increase in stack current, 1.23 factor increase in voltage per cell, and significant increase in the stack average temperature to 1201.270 K, inlet mass-flow rates of 0.264 mol/s, 0.0293 mol/s, and 0.0520 mol/s for steam, hydrogen, and oxygen were determined. Assuming the area-specific resistance remains unchanged,
Table 7 shows the results of the scaled case HTSE simulation. As stated in
Section 10.2.2, the HTSE simulation is a time-dependent steady-state problem. In reality, the startup of the stack should slowly (over approximately 20 s) increase the applied voltage and furnace temperature until operating conditions are achieved for the specified stack current. Instead, the current Dymola HTSE model inversely solves for the stack current assuming thermal-neutral voltage and cell voltage based on furnace temperatures without delay feedback from the cathode- and anode-stack average pressures. The outcome is steady-state stack voltage and furnace temperature. For the applied scaling performance, the steam-inlet mole-flow rate, hydrogen inlet flow rate, and stack voltage increased as anticipated. However, the projected applied temperatures, oxygen mole-flow rate, nitrogen mole-flow rate, and stack current did not behave as in the scaled-case simulation. Despite increasing the scaled-case stack temperature to 1201 K, the applied temperatures did not increase significantly, and the possibility of mistakenly not reflecting other temperature conditions in other modeled components is high. For the scaled-case nitrogen mole-flow rate, oxygen mole-flow rate, and calculated current, it was revealed by the Dymola model code that the physics were being calculated differently from the equation set used in
Section 7. The modeled current density and stack power are [
17]:
Because the current density is calculated by subtracting the operating voltage from the open circuit voltage, and the area-specific resistance is unchanged by the scaling objectives stated in
Section 11. The open-voltage reduced numerator values increase, and as a result, the stack-current decreased compared to the test case values. The corrective action is to either (1) change HTSE scaling equations to adopt Dymola models, or (2) modify the Dymola models to adhere to the set of HTSE equations introduced in
Section 7. Although the cathode mole-flow rates were not affected, a significant reduction in anode mole-flow rates was observed. The HTSE findings in this case suggest that the applied scaling analysis was capable of discovering misplaced physics based on the projected results, as simulation model discrepancy is a large source of error. Unfortunately, due to the steady-state transient nature of HTSE Dymola runs, Equation (
107) cannot be used to visualize distributions of scaling distortions as DSS post-processed parameters require dynamic behavior. Further investigations on the exact cause will be conducted.