Uncertainty And Sensitivity Analysis In Combustion Kinetics

Reliable chemical kinetic model can help us to understand the combustion process deeply,and potentially help us to design higher-performance engines,increase combustion efficiency and reduce harmful emissions.The development of kinetic model usually focuses on three parts:(1)theory in quantum chemistry and reaction kinetics;(2)fundamental combustion experiments,including the measurements of macro combustion properties and the micro mole fraction of species;(3)developing and improving the kinetic models based on(1)and(2).These three parts are both independent and dependent between each other.For example,for the prediction of the rate coefficient of an unimolecular reaction,theoretical methods have been developed from Lindemann theory to RRKM/Master equation theory.With the development,the theory is coming closer to the real physical and chemical process,which is the independent part of development.Using the measured rate coefficients to modified the parameters of RRKM/Master equation theory and obtaining more accurate results of the temperature and pressure(T-P)dependent rate coefficient,which is the interdependent part.The goal of the development of combustion kinetic models is to reduce the uncertainties of model predictions and ultimately get a kinetic model which has the ability for accurate predictions.The process of the model development is thus the process of reducing the uncertainty of model predictions.This work focuses on the methods which can efficiently reduce the uncertainty of kinetic models.Sensitivity and uncertainty analysis were used as the fundamental tools to explore the mechanism of uncertainty propagation in theoretical calculationkinetic model and experiments.Eventually,the relative methods were developed to construct an efficient methodology for reducing the uncertainty of kinetic models.In this work,firstly,a method was proposed to improve the convergent speed of the traditional global sensitivity analysis method using artificial neural network algorithm(ANN),which couples the advantage of ANN for fast convergence and high-dimensional model representation(HDRM)for direct sensitivity indices calculation,and constructs a double-layer surrogate model,and eventually has a faster convergent speed and is more suitable for the complex model.And then we took a deep look at how uncertainty and sensitivity analysis can help us to develop the combustion kinetic model in above three directions.(1)In the direction of theory reaction kinetics,global uncertainty analysis was conducted for the theoretical rate coefficients of ethanol decomposition calculated by RRKM/ME theory.We found that the competition between dissociation and collisional excitation leads to uncertainties of more than a factor of 100 in the predictions-for the higher energy channel.We also proposed a simple and efficient method to estimate the T-P dependent uncertainty of the rate coefficient of unimolecular reactions,which significantly reduces the difficulty and raises the practicability for estimating the T-P dependent uncertainties of the rate coefficients.(2)In the direction of combustion experiments,we proposed a model-assistant method to increase the detection limit of experimental equipments.By estimating the uncertainties of model prediction under different conditions and analyzing the relationship between the system errors under different conditions,we can measure some experimental targets which cannot be quantitatively detected before.(3)In the direction of model development,sensitivity entropy was proposed to guide the selection of relatively efficient experimental systems for specific targets for model optimization.The practicability of this method was validated in the methanol/O2/Ar laminar premixed flame.The study of this work indicates that it is possible to improve the accuracy of experimental measurements of some specific targets,and also possible to design more effective experiments for the uncertainty minimization of the kinetic models by reasonably analyzing the mechanism of uncertainty propagation in theoretical calculation?kinetic model and experiments using appropriate methods.Eventually,an efficient methodology can be constructed to reduce the uncertainty of the kinetic models.