• Citer

Uncertainty Quantification and Optimization of kinetic mechanisms for non-conventional combustion regimes: Turning uncertainties into possibilities

par Fürst, Magnus
Président du jury Degrez, Gérard
Promoteur Parente, Alessandro ;Frassoldati, Alessio
Publication Non publié, 2020-06-10

Thèse de doctorat

• Accès en ligne
• Détails
• Contenu
• Statistiques

Résumé :

The usage of novel combustion technologies, such as Moderate or Intense Low-oxygen Dilution (MILD) combustion, in the future energy mix provides both a flexible and reliable energy supply, together with low emissions. The implementation though is highly situational and numerical studies can help in the assessment of said technologies. However, the existing uncertainties in numerical modeling of MILD combustion are quite significant, and as detailed kinetics should be considered while modeling MILD combustion, a major part of this uncertainty can be accredited to the kinetics. Combined with the fact that existing detailed mechanisms have been developed and validated against conventional combustion targets, there exists a gap between the performance of existing mechanism and experimental findings. To handle this discrepancy, Uncertainty Quantification (UQ) and Optimization are highly viable techniques for reducing this misfit, and have therefore been applied in this work. The strategy applied consisted of first determining the reactions which showed the largest impact towards the experimental targets, by not only considering the sensitivity, but also the uncertainty of the reactions. By using a so-called impact factor, the most influential reactions could be determined, and only the kinetic parameters with the highest impact factors were considered as uncertain in the optimization studies. The uncertainty range of the kinetic parameters were then determined using the uncertainty bounds of the rate coefficients, by finding the lines which intercepts the extreme points of these maximum and minimum rate coefficient curves. Based on this prior parameter space, the optimal combination of the uncertain parameters were determined using two different approaches. The first one utilized Surrogate Models (SMs) for predicting the behavior of changing the kinetic parameters. This is a highly efficient approach, as the computational effort is reduced drastically for each evaluation, and by comparing the physically viable parameter combinations within the pre-determined parameter space, the optimal point could be determined. However, due to limitations of the amount of uncertain parameters and experimental targets that can be used with SMs, an optimization toolbox was developed which uses a more direct optimization approach. The toolbox, called OptiSMOKE++, utilizes the optimization capabilities of DAKOTA, and the simulation of detailed kinetics in reactive systems by OpenSMOKE++. By using efficient optimization methods, the amount of evaluations needed to find the optimal combination of parameters can be drastically reduced. The tool was developed with a flexibility of choosing experimental targets, uncertain kinetic parameters, objective function and optimization method. To present these features, a series of test cases were used and the performance of OptiSMOKE++ was indeed satisfactory. As a final application, the toolbox OptiSMOKE++ was used for optimizing a kinetic mechanism with respect to a large set of experimental targets in MILD conditions. A large amount of uncertain kinetic parameters were also used in the optimization, and the optimized mechanism showed large improvements with respect to the experimental targets. It was also validated against experimental data consisting of species measurements in MILD conditions, and the optimized mechanism showed similar performance as that of the nominal mechanism. However, as the general trend of the species profiles were captured with the nominal mechanism, this was considered satisfactory. The work of this PhD has shown that the application of optimization to kinetic mechanism, can improve the performance of existing mechanism with respect to MILD combustion. Through the development of an efficient toolbox, a large set of experimental data can be used as targets for the optimization, at the same time as many uncertain kinetic parameters can be used contemporary.