Aurélie Bellemans

Reduction of a collisional-radiative mechanism for argon plasma based on principal component analysis

This article considers the development of reduced chemistry models for argon plasmas using Principal Component Analysis (PCA) based methods. Starting from an electronic specific Collisional-Radiative model, a reduction of the variable set (i.e., mass fractions and temperatures) is proposed by projecting the full set on a reduced basis made up of its principal components. Thus, the flow governing equations are only solved for the principal components. The proposed approach originates from the combustion community, where Manifold Generated Principal Component Analysis (MG-PCA) has been developed as a successful reduction technique. Applications consider ionizing shock waves in argon. The results obtained show that the use of the MG-PCA technique enables for a substantial reduction of the computational time.

Principal component analysis acceleration of rovibrational coarse-grain models for internal energy excitation and dissociation

The present work introduces a novel approach for obtaining reduced chemistry representations of large kinetic mechanisms in strong non-equilibrium conditions. The need for accurate reduced-order models arises from compression of large ab initio quantum chemistry databases for their use in fluid codes. The method presented in this paper builds on existing physics-based strategies and proposes a new approach based on the combination of a simple coarse grain model with Principal Component Analysis (PCA). The internal energy levels of the chemical species are regrouped in distinct energy groups with a uniform lumping technique. Following the philosophy of machine learning, PCA is applied on the training data provided by the coarse grain model to find an optimally reduced representation of the full kinetic mechanism. Compared to recently published complex lumping strategies, no expert judgment is required before the application of PCA. In this work, we will demonstrate the benefits of the combined approach, stressing its simplicity, reliability, and accuracy. The technique is demonstrated by reducing the complex quantum N2(Σg+1)-N(Su4) database for studying molecular dissociation and excitation in strong non-equilibrium. Starting from detailed kinetics, an accurate reduced model is developed and used to study non-equilibrium properties of the N2(Σg+1)-N(Su4) system in shock relaxation simulations.

Reduction of a collisional-radiative argon model comparing a modified binning method with principal component analysis

Summary form only given. Considerable effort has been carried out to reduce the complexity of detailed chemistry models for plasma flows. Plasma involves many species and different complex reactions, each of them evolving at a different time-scale. Because of this complexity, numerical simulations often remain restricted to simple zero-or one-dimensional calculations. Different reduction techniques have been developed over the years. Time-scale based reductions are possible through rate-controlled constrained equilibrium methods or singular perturbation methods as investigated by the combustion community. Other techniques aim to limit the number of species. Coarse-grain models resulting from binning methods, for example, have been developed for simulating rovibrational nitrogen chemistry. The energy levels of ab initio databases are being lumped into several bins. The number of governing equations is significantly reduced as the considered species equal the amount of bins.This paper compares two techniques for obtaining a reduced model for a 34-species collisional-radiative argon mechanism. The first technique relies on a modified binning method. The excited states are separated from the ground state and averaged through a Maxwell-Boltzmann distribution in a separate bin. Separating the ground state from the bin allows to keep the entire reactive scheme as the excitation reactions are not canceled out. An extra energy equation is solved for the regrouped states. The state model uses an extra temperature to characterize the temperature evolution of the bin. This state to state reduction technique is compared with previously investigated work on Principal Component Analysis. PCA reduces the number of variables of the simulation by projecting the system on a base formed by the principal components. These principal components are retrieved by solving an eigenvalue problem on the covariance matrix of the full dataset and correspond to the eigenvectors with the highest eigenvalues. The advantage of PCA lies in the conservation of detailed chemistry information for every species. Such information is usually lost when applying binning techniques. The present work gives a detailed comparison between both reduction techniques and their resulting model.

MG-local-PCA Method for the Reduction of a Collisional-Radiative Argon Plasma Mechanism

The present paper introduces the use of a locally applied Manifold Generated Principal Component Analysis (MG-local-PCA) on a collisional-radiative mechanism for a 34-species argon mixture. MG-PCA is applied on 1D shock relaxation simulations. The method is validated against shock tube experiments from the University of Toronto (UTIAS). The reduced model is obtained by projecting the original variable set on a reduced basis containing its principal components. An important reduction of simulation cost is obtained as the number of equations has been reduced to the number of principal components.

Development of Reduced Chemistry Models for High Enthalpy and Plasma Flows

This paper considers the development of reduced chemistry models for high enthalpy and plasma flows using Principal Component Analysis (PCA) based methods. Starting from detailed chemistry models, such as multi-temperature and collisional-radiative formulations, a reduction of the variable set (species mass fractions and temperatures) is proposed by projecting the full set on a reduced basis made up of its principal components. Consequently, an important reduction of calculation time is obtained as the governing flow equations are solved for these principal components only. This approach originates from the combustion field, where manifold generated principal component analysis (MG-PCA) has been developed as a successful reduction technique. In this work MG-PCA has been implemented and verified on shock tube simulations for argon plasma based on an electronic specific detailed chemistry model. The method has been validated with experimental results from UTIAS (University of Toronto, Insitute for Aerospace Studies).