Speaker
Description
During the design phases of a spacecraft development, a thermal model of the spacecraft is created that is used to predict the operational temperatures under a defined set of terrestrial and space environments.This thermal model involves a large number of inputs coming from a large number of sources. Some inputs, such as the thermal conductivity of a material are well defined and subject to small variability,especially when modeled as temperature dependent. Other inputs are subject to greater uncertainty,these include:
- Surface optical properties. These are usually taken from handbooks listing emissivities and
absorptivities of different surface finishes. - Conductance across bolted joints or across composite joints. These are often taken from handbook
data that cover very specific cases and applied to a broad range of situations. - MLI effective emissivity. This is typically handled using worst case values (high values for cold case,low values for hot case) as the actual value is a complex function of MLI materials, area, layers,stitching and penetrations
While some recent advances and demonstrations have been made in model correlation techniques, the computational overhead of varying many parameters to find an optimal set still prevents fast and effective model correlation. The most time-consuming step of such a computation is computing the sensitivities of the correlation function to the different combinations of input parameters.
Indeed, the computational effort of many other correlation technologies requires a large number of model runs that can scale anywhere from a factor of N to 2N, depending on the method, where N is the number of design variables that could be changed in a model.
In contrast, by using the adjoint sensitivity method, first-order sensitivities of the correlation function to all the design variables can be computed with just two runs: a so-called “forward calculation” and an “adjoint calculation”. These sensitivities can be used to estimate new design variable conditions at which new forward and adjoint problems can be solved, to minimize a functional which measures the error between the simulation and test or in flight data.
The goal of this paper to present the fundamentals of the adjoint correlation method implemented: solution of governing equations, solution of the adjoint equations, calculation of the gradient vector, and updating the design variables.
Some preliminary results on a few “real-life” steady-state and transient models will be presented together with some metrics of computational gains.
