Investigation of different neural network architectures for dynamic system identification: Applications to orbital mechanics

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Machine learning and new AI algorithms inspire the scientific community to explore and develop new approaches for discovery of scientific laws and governing equations for complex physical and nonlinear dynamical systems. The question on how well deep learning approaches can approximate a given set of input data is difficult to answer. Considering the unperturbed two-body problem, this paper investigates the approximation and prediction capabilities of three types of neural networks: Feed-Forward, Residual and Deep Residual. Used in a purely recurrent model, this three architectures are able to produce highly satisfactory performances, very close to numerical integration tolerances. Furthermore, the effect of the mathematical representation (i.e. coordinate system) on the learning process is also investigated. From numerical results, it can be inferred that NN were able to better learn inherent dynamics characteristics in spherical coordinates without any apriori information than in Cartesian coordinate system. It is shown that a simple NN architecture is able to learn the symmetry of the central force and reproduce the conservation of the constants of the motion.

Original languageEnglish (US)
Title of host publicationSpaceflight Mechanics 2019
EditorsFrancesco Topputo, Andrew J. Sinclair, Matthew P. Wilkins, Renato Zanetti
PublisherUnivelt Inc.
Pages1789-1803
Number of pages15
ISBN (Print)9780877036593
StatePublished - Jan 1 2019
Event29th AAS/AIAA Space Flight Mechanics Meeting, 2019 - Maui, United States
Duration: Jan 13 2019Jan 17 2019

Publication series

NameAdvances in the Astronautical Sciences
Volume168
ISSN (Print)0065-3438

Conference

Conference29th AAS/AIAA Space Flight Mechanics Meeting, 2019
CountryUnited States
CityMaui
Period1/13/191/17/19

Fingerprint

orbital mechanics
Nonlinear dynamical systems
system identification
Network architecture
learning
mechanics
Learning systems
Conservation
Identification (control systems)
Mechanics
Dynamical systems
Neural networks
two body problem
machine learning
Cartesian coordinates
spherical coordinates
numerical integration
dynamical systems
dynamic characteristics
symmetry

All Science Journal Classification (ASJC) codes

  • Aerospace Engineering
  • Space and Planetary Science

Cite this

Guého, D., Singla, P., & Melton, R. G. (2019). Investigation of different neural network architectures for dynamic system identification: Applications to orbital mechanics. In F. Topputo, A. J. Sinclair, M. P. Wilkins, & R. Zanetti (Eds.), Spaceflight Mechanics 2019 (pp. 1789-1803). [AAS 19-479] (Advances in the Astronautical Sciences; Vol. 168). Univelt Inc..
Guého, Damien ; Singla, Puneet ; Melton, Robert G. / Investigation of different neural network architectures for dynamic system identification : Applications to orbital mechanics. Spaceflight Mechanics 2019. editor / Francesco Topputo ; Andrew J. Sinclair ; Matthew P. Wilkins ; Renato Zanetti. Univelt Inc., 2019. pp. 1789-1803 (Advances in the Astronautical Sciences).
@inproceedings{2c6c35ff6e184fc8a050dc752e2b6357,
title = "Investigation of different neural network architectures for dynamic system identification: Applications to orbital mechanics",
abstract = "Machine learning and new AI algorithms inspire the scientific community to explore and develop new approaches for discovery of scientific laws and governing equations for complex physical and nonlinear dynamical systems. The question on how well deep learning approaches can approximate a given set of input data is difficult to answer. Considering the unperturbed two-body problem, this paper investigates the approximation and prediction capabilities of three types of neural networks: Feed-Forward, Residual and Deep Residual. Used in a purely recurrent model, this three architectures are able to produce highly satisfactory performances, very close to numerical integration tolerances. Furthermore, the effect of the mathematical representation (i.e. coordinate system) on the learning process is also investigated. From numerical results, it can be inferred that NN were able to better learn inherent dynamics characteristics in spherical coordinates without any apriori information than in Cartesian coordinate system. It is shown that a simple NN architecture is able to learn the symmetry of the central force and reproduce the conservation of the constants of the motion.",
author = "Damien Gu{\'e}ho and Puneet Singla and Melton, {Robert G.}",
year = "2019",
month = "1",
day = "1",
language = "English (US)",
isbn = "9780877036593",
series = "Advances in the Astronautical Sciences",
publisher = "Univelt Inc.",
pages = "1789--1803",
editor = "Francesco Topputo and Sinclair, {Andrew J.} and Wilkins, {Matthew P.} and Renato Zanetti",
booktitle = "Spaceflight Mechanics 2019",
address = "United States",

}

Guého, D, Singla, P & Melton, RG 2019, Investigation of different neural network architectures for dynamic system identification: Applications to orbital mechanics. in F Topputo, AJ Sinclair, MP Wilkins & R Zanetti (eds), Spaceflight Mechanics 2019., AAS 19-479, Advances in the Astronautical Sciences, vol. 168, Univelt Inc., pp. 1789-1803, 29th AAS/AIAA Space Flight Mechanics Meeting, 2019, Maui, United States, 1/13/19.

Investigation of different neural network architectures for dynamic system identification : Applications to orbital mechanics. / Guého, Damien; Singla, Puneet; Melton, Robert G.

Spaceflight Mechanics 2019. ed. / Francesco Topputo; Andrew J. Sinclair; Matthew P. Wilkins; Renato Zanetti. Univelt Inc., 2019. p. 1789-1803 AAS 19-479 (Advances in the Astronautical Sciences; Vol. 168).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

TY - GEN

T1 - Investigation of different neural network architectures for dynamic system identification

T2 - Applications to orbital mechanics

AU - Guého, Damien

AU - Singla, Puneet

AU - Melton, Robert G.

PY - 2019/1/1

Y1 - 2019/1/1

N2 - Machine learning and new AI algorithms inspire the scientific community to explore and develop new approaches for discovery of scientific laws and governing equations for complex physical and nonlinear dynamical systems. The question on how well deep learning approaches can approximate a given set of input data is difficult to answer. Considering the unperturbed two-body problem, this paper investigates the approximation and prediction capabilities of three types of neural networks: Feed-Forward, Residual and Deep Residual. Used in a purely recurrent model, this three architectures are able to produce highly satisfactory performances, very close to numerical integration tolerances. Furthermore, the effect of the mathematical representation (i.e. coordinate system) on the learning process is also investigated. From numerical results, it can be inferred that NN were able to better learn inherent dynamics characteristics in spherical coordinates without any apriori information than in Cartesian coordinate system. It is shown that a simple NN architecture is able to learn the symmetry of the central force and reproduce the conservation of the constants of the motion.

AB - Machine learning and new AI algorithms inspire the scientific community to explore and develop new approaches for discovery of scientific laws and governing equations for complex physical and nonlinear dynamical systems. The question on how well deep learning approaches can approximate a given set of input data is difficult to answer. Considering the unperturbed two-body problem, this paper investigates the approximation and prediction capabilities of three types of neural networks: Feed-Forward, Residual and Deep Residual. Used in a purely recurrent model, this three architectures are able to produce highly satisfactory performances, very close to numerical integration tolerances. Furthermore, the effect of the mathematical representation (i.e. coordinate system) on the learning process is also investigated. From numerical results, it can be inferred that NN were able to better learn inherent dynamics characteristics in spherical coordinates without any apriori information than in Cartesian coordinate system. It is shown that a simple NN architecture is able to learn the symmetry of the central force and reproduce the conservation of the constants of the motion.

UR - http://www.scopus.com/inward/record.url?scp=85072959022&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85072959022&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:85072959022

SN - 9780877036593

T3 - Advances in the Astronautical Sciences

SP - 1789

EP - 1803

BT - Spaceflight Mechanics 2019

A2 - Topputo, Francesco

A2 - Sinclair, Andrew J.

A2 - Wilkins, Matthew P.

A2 - Zanetti, Renato

PB - Univelt Inc.

ER -

Guého D, Singla P, Melton RG. Investigation of different neural network architectures for dynamic system identification: Applications to orbital mechanics. In Topputo F, Sinclair AJ, Wilkins MP, Zanetti R, editors, Spaceflight Mechanics 2019. Univelt Inc. 2019. p. 1789-1803. AAS 19-479. (Advances in the Astronautical Sciences).