-
Hierarchical Higher-Order Dynamic Mode Decomposition for Clustering and Feature Selection
Authors:
Adrián Corrochano,
Giuseppe D'Alessio,
Alessandro Parente,
Soledad Le Clainche
Abstract:
In this work, a new algorithm based on the application of higher-order dynamic mode decomposition (HODMD) is proposed for feature selection and variables clustering in reacting flow simulations. The hierarchical HODMD (h-HODMD) performs a reduction of the model order, followed by the iterative selection of the best reconstructed variables thus creating clusters of features which can eventually be…
▽ More
In this work, a new algorithm based on the application of higher-order dynamic mode decomposition (HODMD) is proposed for feature selection and variables clustering in reacting flow simulations. The hierarchical HODMD (h-HODMD) performs a reduction of the model order, followed by the iterative selection of the best reconstructed variables thus creating clusters of features which can eventually be associated with distinct dynamical phenomena. Firstly, h-HODMD is combined with different data pre-processing techniques to assess their influence on the algorithm in terms of reconstruction error. Afterwards, the algorithm is applied to analyze three different databases obtained from numerical simulations of a non-premixed co-flow methane flame, and its performance are compared with the standard HODMD in terms of the achievable degree of reduction as well as in terms of reconstruction error. Results show that h-HODMD improves the reconstruction for all the variables when compared to the standard HODMD algorithm. This condition is achieved thanks to the iterative variables' clustering: finding dedicated modes for a specific group of features does in fact lead to a better reconstruction of the dynamics with respect to the case when the same (global) modes are used to reconstruct the entire set of variables. Finally, the clusters of variables found by means of h-HODMD are analyzed, and it is observed that the algorithm can group chemical species whose behavior is also consistent from a kinetic point of view. In fact, it allows for the possibility to formulate inexpensive reduced dynamical models for predicting flames liftoff, as well as for identifying the formation of local extinction and blowout conditions, to formulate accurate reduced models to describe the formation of pollutants in aviation, and for control purposes.
△ Less
Submitted 22 February, 2023; v1 submitted 19 January, 2023;
originally announced January 2023.
-
Advancing Reacting Flow Simulations with Data-Driven Models
Authors:
Kamila Zdybał,
Giuseppe D'Alessio,
Gianmarco Aversano,
Mohammad Rafi Malik,
Axel Coussement,
James C. Sutherland,
Alessandro Parente
Abstract:
The use of machine learning algorithms to predict behaviors of complex systems is booming. However, the key to an effective use of machine learning tools in multi-physics problems, including combustion, is to couple them to physical and computer models. The performance of these tools is enhanced if all the prior knowledge and the physical constraints are embodied. In other words, the scientific me…
▽ More
The use of machine learning algorithms to predict behaviors of complex systems is booming. However, the key to an effective use of machine learning tools in multi-physics problems, including combustion, is to couple them to physical and computer models. The performance of these tools is enhanced if all the prior knowledge and the physical constraints are embodied. In other words, the scientific method must be adapted to bring machine learning into the picture, and make the best use of the massive amount of data we have produced, thanks to the advances in numerical computing. The present chapter reviews some of the open opportunities for the application of data-driven reduced-order modeling of combustion systems. Examples of feature extraction in turbulent combustion data, empirical low-dimensional manifold (ELDM) identification, classification, regression, and reduced-order modeling are provided.
△ Less
Submitted 5 September, 2022;
originally announced September 2022.
-
Local manifold learning and its link to domain-based physics knowledge
Authors:
Kamila Zdybał,
Giuseppe D'Alessio,
Antonio Attili,
Axel Coussement,
James C. Sutherland,
Alessandro Parente
Abstract:
In many reacting flow systems, the thermo-chemical state-space is known or assumed to evolve close to a low-dimensional manifold (LDM). Various approaches are available to obtain those manifolds and subsequently express the original high-dimensional space with fewer parameterizing variables. Principal component analysis (PCA) is one of the dimensionality reduction methods that can be used to obtai…
▽ More
In many reacting flow systems, the thermo-chemical state-space is known or assumed to evolve close to a low-dimensional manifold (LDM). Various approaches are available to obtain those manifolds and subsequently express the original high-dimensional space with fewer parameterizing variables. Principal component analysis (PCA) is one of the dimensionality reduction methods that can be used to obtain LDMs. PCA does not make prior assumptions about the parameterizing variables and retrieves them empirically from the training data. In this paper, we show that PCA applied in local clusters of data (local PCA) is capable of detecting the intrinsic parameterization of the thermo-chemical state-space. We first demonstrate that utilizing three common combustion models of varying complexity: the Burke-Schumann model, the chemical equilibrium model and the homogeneous reactor. Parameterization of these models is known a priori which allows for benchmarking with the local PCA approach. We further extend the application of local PCA to a more challenging case of a turbulent non-premixed $n$-heptane/air jet flame for which the parameterization is no longer obvious. Our results suggest that meaningful parameterization can be obtained also for more complex datasets. We show that local PCA finds variables that can be linked to local stoichiometry, reaction progress and soot formation processes.
△ Less
Submitted 1 July, 2022;
originally announced July 2022.
-
Higher order dynamic mode decomposition to model reacting flows
Authors:
Adrián Corrochano,
Giuseppe D'Alessio,
Alessandro Parente,
Soledad Le Clainche
Abstract:
In this work, the application of the multi-dimensional higher order dynamic mode decomposition (HODMD) is proposed for the first time to analyse combustion databases. In particular, HODMD has been adapted and combined with other pre-processing techniques (generally used in machine learning), in light of the multivariate nature of the data. A truncation step separate the main dynamics driving the f…
▽ More
In this work, the application of the multi-dimensional higher order dynamic mode decomposition (HODMD) is proposed for the first time to analyse combustion databases. In particular, HODMD has been adapted and combined with other pre-processing techniques (generally used in machine learning), in light of the multivariate nature of the data. A truncation step separate the main dynamics driving the flow from less relevant non-linear dynamics. The method is applied to analyse a database obtained from a Computational Fluid Dynamics (CFD) simulation of an axisymmetric, time varying, non-premixed, co-flow methane flame carried out by means of a detailed kinetic mechanism. Results show that HODMD can reconstruct the main jet dynamics with a reduced number of relevant modes, able to reproduce the system dynamics. These modes are found to be representative for the main flow physics with two main advantages: (i) they provide for the possibility to achieve a strong simplification with respect to the high-dimensional input data, and at the same time (ii) a small reconstruction error with respect to the original dataset is observed. In addition, the method was also validated considering a reduced matrix obtained using Principal Component Analysis (PCA) based feature selection and the Varimax rotation. This validation also reveals that it is not important to have all the variables in the dataset, just a group of them is necessary to obtain the main dynamics of the system. This has an impact on feature selection and on the cost these methodologies for very massive data.
△ Less
Submitted 22 March, 2022;
originally announced March 2022.
