Feature extraction and reduced-order modelling of nitrogen plasma models using principal component analysis

Principal component analysis has been presented in recent research as an accurate and efficient method to reduce the complex chemistry and kinetics of large reacting mechanisms. Following the reduction, the original variables are transformed and projected onto a set of independent, orthogonal variables maximizing the total variance in the system: the principal components. However, these new variables are difficult to interpret physically and may introduce instabilities in the low dimensional representation of the manifold. In the present paper we will show the benefits of coupling PCA to a rotation method: the interpretation of the principal components can be related back to the physics. The advantages of rotation are demonstrated on a PCA reduced model for modelling dissociation and excitation processes in nitrogen shock flows.