532
A. del Val et al.
16.3.5 Epistemic Uncertainties
Epistemic uncertainty (also known as reducible uncertainty or incertitude) is defined
as a potential deficiency that is due to a lack of knowledge [42]. It can arise
from assumptions introduced in the derivation of the mathematical model or
simplifications related to the correlation or dependence between physical processes.
It is obviously possible to reduce the epistemic uncertainty by using, for example,
a combination of calibration, further inference from experimental observations and
improvement of the physical models. Epistemic uncertainties can be incorporated in
a Bayesian framework where a calibration of the model parameters can be carried
out. The resulting uncertainties tell us what should be improved in the system to have
significant reliability in our results and how we can achieve that through experiments
by looking at the optimal testing conditions. This calibration of model parameters
represents an important step towards a more unified framework that incorporates
simulation results and experiments and their best possible reliability by giving us
confidence in our results. Typical examples of sources of epistemic uncertainties in
hypersonic flows are surrogate chemical kinetics models and the use of γ models
for catalytic effects.
It is important to notice that in most cases the epistemic uncertainty is not well
characterized by probabilistic approaches. The reason is that it might be difficult
to infer any statistical information due to the nominal lack of knowledge, i.e. the
suitability of the model that is being used to calibrate the parameters is not properly
known. To cope with this, other frameworks are put in place. Simulations using
different plausible models are evaluated and the largest possible confidence margins
are defined.
When we explore the differences between measurements uncertainties and epistemic uncertainties, we have to remark that the former are regarded as frequentist,
meaning that repetitions of a certain measurement give you a measure of the
randomness of the measurement process (producing a Probability Distribution
Function (PDF) that can be used in our tools). Epistemic uncertainties do not have a
basis to define their PDFs. One way of tackling this issue is by adopting a Bayesian
framework, as explained before, in which with sufficient prior knowledge and data, a
PDF can be assigned to the various parameters inferred in this manner in our model
[43].
16.4 Putting It All Together: Extrapolation to Flight
In the design phase of aerospace vehicles some questions need to be addressed
experimentally in order to have a more accurate representation of the reality
encountered during flight. As scaled-down models cannot reproduce all aspects
of atmospheric entry flight, extrapolation methodologies exist for both types of
facilities reviewed in Sect. 16.2.
A. del Val et al.
16.3.5 Epistemic Uncertainties
Epistemic uncertainty (also known as reducible uncertainty or incertitude) is defined
as a potential deficiency that is due to a lack of knowledge [42]. It can arise
from assumptions introduced in the derivation of the mathematical model or
simplifications related to the correlation or dependence between physical processes.
It is obviously possible to reduce the epistemic uncertainty by using, for example,
a combination of calibration, further inference from experimental observations and
improvement of the physical models. Epistemic uncertainties can be incorporated in
a Bayesian framework where a calibration of the model parameters can be carried
out. The resulting uncertainties tell us what should be improved in the system to have
significant reliability in our results and how we can achieve that through experiments
by looking at the optimal testing conditions. This calibration of model parameters
represents an important step towards a more unified framework that incorporates
simulation results and experiments and their best possible reliability by giving us
confidence in our results. Typical examples of sources of epistemic uncertainties in
hypersonic flows are surrogate chemical kinetics models and the use of γ models
for catalytic effects.
It is important to notice that in most cases the epistemic uncertainty is not well
characterized by probabilistic approaches. The reason is that it might be difficult
to infer any statistical information due to the nominal lack of knowledge, i.e. the
suitability of the model that is being used to calibrate the parameters is not properly
known. To cope with this, other frameworks are put in place. Simulations using
different plausible models are evaluated and the largest possible confidence margins
are defined.
When we explore the differences between measurements uncertainties and epistemic uncertainties, we have to remark that the former are regarded as frequentist,
meaning that repetitions of a certain measurement give you a measure of the
randomness of the measurement process (producing a Probability Distribution
Function (PDF) that can be used in our tools). Epistemic uncertainties do not have a
basis to define their PDFs. One way of tackling this issue is by adopting a Bayesian
framework, as explained before, in which with sufficient prior knowledge and data, a
PDF can be assigned to the various parameters inferred in this manner in our model
[43].
16.4 Putting It All Together: Extrapolation to Flight
In the design phase of aerospace vehicles some questions need to be addressed
experimentally in order to have a more accurate representation of the reality
encountered during flight. As scaled-down models cannot reproduce all aspects
of atmospheric entry flight, extrapolation methodologies exist for both types of
facilities reviewed in Sect. 16.2.