This non-constant or unacceptably high discrepancy bias indicates that the model fails to capture some fundamental engineering or physics phenomena. Different modeling assumptions are validated (or rather invalidated) as the model is compared against experiments conducted at different settings within the domain of applicability. Therefore, the fluctuations in the discrepancy bias are attributable to the invalidation of these underlying modeling assumptions. Thus, predictive forecasting to untested regimes should not be attempted. 3.3 Concept of Stabilization In this Section, the assertions of Section 3.1 and Section 3.2 are combined in the concept of stabilization. The concept of stabilization states that each new experiment provides fractional new information for model calibration and thus should provide a fractional reduction in discrepancy bias. Therefore, the discrepancy term gradually converges to the ‘true discrepancy’, henceforth referred to as model form error. The point at which a model stabilizes establishes the usefulness of the model. If the model does not stabilize or stabilizes with too high of a model form error, the model will not be useful as a predictive tool. Figure 1: Comparison of discrepancy estimated using five and 15 experimental measurements to the ‘true discrepancy.’ In Figure 1, the estimated discrepancy bias is compared against the model form error as the number of experimental data points are increased from five to 15. Figure 1 depicts that with 15 experiments, the 437
RkJQdWJsaXNoZXIy MTMzNzEzMQ==