Resumo:
Several relevant real-world applications rely on an inverse problem, which involves
recovering unknown causes from observing their effects. This differs from the corresponding
direct problem, whose solution involves predicting effects from a complete description of their
causes. Naturally, inverse problems are more challenging than direct problems because, in
general, they are ill-posed, i.e., the solution either does not exist, is not unique or it does not
depend continuously on the input data. To soften this problematic aspect, applied inverse
modeling requires detailed mathematical-physical modeling and well-designed experiments
since the desired parameters are estimated by comparing calculated data with experimental
measurements.
In this PhD thesis, inverse approach was applied to experimentally investigate three case
studies:
• complementary experiments to simultaneously estimate the parameters describing the
temperature-dependent thermal conductivity and specific heat of 304 austenitic stainless
steel. Parameter estimation takes advantage of additional information provided by two
heat-conducting solids with different geometries. It is an alternative approach to
standard thermal characterization techniques, which are often beyond the reach of many
laboratories.
• one-year on-site measurements to estimate various hygrothermal properties and thus
calibrate the simulation model of a lightweight multilayer wall. A 2D fully coupled heat
and moisture transfer model was used to investigate the in-use response of the panel
junction region, which is critical in terms of airtightness. The results enable an accurate
assessment of building operating conditions by reducing uncertainties in material input
data.
• field data to determine the annual heat conduction flux through a wall assembly in an
occupied house. Inverse modeling accounted for the physical interactions between
outdoor environment and indoor occupancy. The methodology and the findings are
useful to support decision-making on energy performance, as there is a lack of longterm
field monitoring and information on dynamic heat flux related to prefabricated
occupied dwellings.
All the above inverse analyzes were based on evaluating the match between data predicted
by numerical simulations in COMSOL Multiphysics and measurements conveying the physical
behavior of the component under study. Numerical and experimental data were processed and
used for inverse estimation purposes in MATLAB environment. After careful analysis of
sensitivity coefficients, different optimization approaches were used to solve the inverse
problems. Bayesian statistical inference was applied to determine the estimates and
corresponding uncertainties of the thermal properties of 304 stainless steel. The Broyden–
Fletcher–Goldfarb–Shanno (BFGS) algorithm, which determines the descent direction by
preconditioning the gradient with curvature information, was used in the second case study.
The wall heat flux was estimated using the sequential function specification method (SFSM),
which expresses temperature as function of heat flux by means of a first-order Taylor series.
The results show that inverse modeling is a reliable tool for obtaining valuable information
about the hygrothermal mechanisms and parameters involved in applied engineering problems.