References in periodicals archive ?
The process typically starts with a review of the MVVs, followed by a few general comments, which are then summarized with a fair amount of conventional wordsmithing.
In terms of the MVVs, the result is little more than a procedural formality, and, "too often they end up as a written document sitting on a shelf or a plaque hanging on the wall and are disconnected from the strategic formulation and implementation." (7)
Hence the current situation can be understood, where despite the availability of detailed dimensional (but system-level) models, model-based engine calibration is primarily data-based, and the known physics of the system-level model is unused by calibration engineers, that is, the MVVs are unused.
It is difficult to utilize the MVVs directly because the physics determining MVVs is different from those affecting LVs.
The term "mean-value variables" (MVVs) will refer to one of the 30 GT-Power predictions used to generate the artificial feature spaces.
The only difference is that the one-dimensional model used to generate MVVs was uncalibrated.
Fraction Fraction Fraction extrapolated: N[O.sub.x] extrapolated: extrapolated: VE opacity k-NN Regression Neural k-NN Regression Neural k-NN Regression network network 33% 46% 27% 25% 33% 18% 42% 22% Fraction extrapolated: VE Neural network 18% TABLE 4 Results over interpolation dataset obtained by using MVVs from uncalibrated physical model.
Hence, the test data that was extrapolation in the original experimental feature space of Figure 2 is not an extrapolation in the artificially generated MVV feature space of Figure 4.
The one-dimensional GT-Power model produced 30 MVV predictions that could be suitable candidates for individual features, but these variables could be combined into 736,281 subsets of size 6 or less (each subset is a vector and unique feature space).
Acronyms browser ?
Full browser ?