IAC-10.C1.7.5 DIRECT TRANSCRIPTION OF LOW-THRUST TRAJECTORIES WITH FINITE TRAJECTORY ELEMENTS



Similar documents
Lecture L17 - Orbit Transfers and Interplanetary Trajectories

Astrodynamics (AERO0024)

Chapter 2. Mission Analysis. 2.1 Mission Geometry

Asteroid Deflection Theory: fundamentals of orbital mechanics and optimal control

Discrete mechanics, optimal control and formation flying spacecraft

Orbital Dynamics: Formulary

Can Hubble be Moved to the International Space Station? 1

Spacecraft Dynamics and Control. An Introduction

Penn State University Physics 211 ORBITAL MECHANICS 1

Coverage Characteristics of Earth Satellites

AUTOMATING INTERPLANETARY TRAJECTORY GENERATION FOR ELECTRIC PROPULSION TRADE STUDIES

Orbital Mechanics and Space Geometry

Solution of the Gaussian Transfer Orbit Equations of Motion

Numerical Analysis An Introduction

Reduction of Low-Thrust Continuous Controls for Trajectory Dynamics and Orbital Targeting

MINIMUM-TIME LOW-EARTH ORBIT TO HIGH-EARTH ORBIT LOW-THRUST TRAJECTORY OPTIMIZATION

ADVANCED TOPICS IN ASTRODYNAMICS GRAVITATIONAL ASSISTED TRAJECTORIES

Lecture L5 - Other Coordinate Systems

Actuator Constrained Optimal Control of Formations Near the Libration Points

Orbital Mechanics. Angular Momentum

G U I D E T O A P P L I E D O R B I T A L M E C H A N I C S F O R K E R B A L S P A C E P R O G R A M

PS 320 Classical Mechanics Embry-Riddle University Spring 2010

The Technical Archer. Austin Wargo

CHAPTER 3 SECURITY CONSTRAINED OPTIMAL SHORT-TERM HYDROTHERMAL SCHEDULING

Mechanics 1: Conservation of Energy and Momentum

2. Orbits. FER-Zagreb, Satellite communication systems 2011/12

Flight and Orbital Mechanics

Numerical Methods for Differential Equations

SOLID MECHANICS TUTORIAL MECHANISMS KINEMATICS - VELOCITY AND ACCELERATION DIAGRAMS

Dynamics of Iain M. Banks Orbitals. Richard Kennaway. 12 October 2005

Small Satellite LEO Maneuvers with Low-Power Electric Propulsion

DIN Department of Industrial Engineering School of Engineering and Architecture

Lecture L14 - Variable Mass Systems: The Rocket Equation

ORBITAL DISPERSION SIMULATION OF NEAR-EARTH OBJECT DEFLECTION/FRAGMENTATION BY NUCLEAR EXPLOSIONS

Machine Learning and Pattern Recognition Logistic Regression

Introduction to Engineering System Dynamics

Physics Notes Class 11 CHAPTER 3 MOTION IN A STRAIGHT LINE

Introduction to Aerospace Engineering

A Mathematical Programming Solution to the Mars Express Memory Dumping Problem

Trajectory design for the Solar Orbiter mission

Chapter 6 Circular Motion

Lecture Notes to Accompany. Scientific Computing An Introductory Survey. by Michael T. Heath. Chapter 10

PHY121 #8 Midterm I

Review of Vector Analysis in Cartesian Coordinates

HybridSail. Hybrid Solar Sails for Active Debris Removal Final Report

Does currently available technology have the capacity to facilitate a manned mission to Mars?

Maximizing Payload Mass Fractions of Spacecraft for Interplanetary Electric Propulsion Missions

KINEMATICS OF PARTICLES RELATIVE MOTION WITH RESPECT TO TRANSLATING AXES

State of Stress at Point

Stability Analysis of Small Satellite Formation Flying and Reconfiguration Missions in Deep Space

Columbia University Department of Physics QUALIFYING EXAMINATION

High-fidelity electromagnetic modeling of large multi-scale naval structures

Georgia Department of Education Kathy Cox, State Superintendent of Schools 7/19/2005 All Rights Reserved 1

Time domain modeling

Lecture L6 - Intrinsic Coordinates

AN INTRODUCTION TO NUMERICAL METHODS AND ANALYSIS

11. Rotation Translational Motion: Rotational Motion:

Homework #1 Solutions

NUMERICAL ANALYSIS OF OPEN CHANNEL STEADY GRADUALLY VARIED FLOW USING THE SIMPLIFIED SAINT-VENANT EQUATIONS

USE OF SCILAB FOR SPACE MISSION ANALYSIS AND FLIGHT DYNAMICS ACTIVITIES

Lecture L3 - Vectors, Matrices and Coordinate Transformations

Halliday, Resnick & Walker Chapter 13. Gravitation. Physics 1A PHYS1121 Professor Michael Burton

APPLIED MATHEMATICS ADVANCED LEVEL

LOW THRUST TRAJECTORY OPTIMIZATION FOR AUTONOMOUS ASTEROID RENDEZVOUS MISSIONS. Anne Schattel, Mitja Echim, Matthias Knauer, Christof Büskens

Trigonometry LESSON ONE - Degrees and Radians Lesson Notes

LINEAR ALGEBRA W W L CHEN

Functional Optimization Models for Active Queue Management

Midterm Exam 1 October 2, 2012

Unified Lecture # 4 Vectors

EN4 Dynamics and Vibrations. Design Project. Orbital Design for a Lunar Impact Mission. Synopsis

Correct Modeling of the Indirect Term for Third-Body Perturbations

Model-based Parameter Optimization of an Engine Control Unit using Genetic Algorithms

How High a Degree is High Enough for High Order Finite Elements?

Orbital Dynamics with Maple (sll --- v1.0, February 2012)

Frequency-domain and stochastic model for an articulated wave power device

Notes: Most of the material in this chapter is taken from Young and Freedman, Chap. 13.

Quasi-static evolution and congested transport

Presentation of problem T1 (9 points): The Maribo Meteorite

WEIGHTLESS WONDER Reduced Gravity Flight

Examination Space Missions and Applications I (AE2103) Faculty of Aerospace Engineering Delft University of Technology SAMPLE EXAM

Halliday, Resnick & Walker Chapter 13. Gravitation. Physics 1A PHYS1121 Professor Michael Burton

Precession of spin and Precession of a top

TOPO Trajectory Operations Officer

Analysis on the Long-term Orbital Evolution and Maintenance of KOMPSAT-2

Chapter 2. Derivation of the Equations of Open Channel Flow. 2.1 General Considerations

Astromechanics Two-Body Problem (Cont)

A New Nature-inspired Algorithm for Load Balancing

A Novel Binary Particle Swarm Optimization

Centripetal Force. This result is independent of the size of r. A full circle has 2π rad, and 360 deg = 2π rad.

This paper is also taken for the relevant Examination for the Associateship. For Second Year Physics Students Wednesday, 4th June 2008: 14:00 to 16:00

Introduction to the Finite Element Method

A Direct Numerical Method for Observability Analysis

ECON20310 LECTURE SYNOPSIS REAL BUSINESS CYCLE

Transcription:

IAC-1.C1.7.5 DIRECT TRANSCRIPTION OF LOW-THRUST TRAJECTORIES WITH FINITE TRAJECTORY ELEMENTS Federico Zuiani PhD Candidate, Department of Aerospace Engineering, University of Glasgow, UK, fzuiani@eng.gla.ac.uk Massimiliano Vasile Reader, Department of Mechanical Engineering, University of Strathclyde, UK massimiliano.vasile@strath.ac.uk Alessandro Palmas Graduate Student, Department of Aerospace Engineering, Politecnico di Torino, Turin Italy, erci_ale@hotmail.com Giulio Avanzini Assistant Professor, Department of Aerospace Engineering, Politecnico di Torino, Turin Italy, giulio.avanzini@polito.it This paper presents novel approach to Low-Thrust trajectory design, based on a first order approximate analytical solution of Gauss planetary equations. The analytical solution is fairly accurate and could be employed for a fast propagation of perturbed Keplerian motion. Thus, it has been integrated in a Direct Transcription Method based one Finite Perturbed Elements in Time (FPET). It has been proved that this method is suitable for the solution of boundary value transfer problem by few iterations of a gradient based local optimizer. This allows in turn to include the FPET Transcription method into a global optimization tool to solve orbital transfer problem where both total V and transfer time need to be minimised with respect to departure and arrival dates. To prove the concept, two transfer problems have been investigated, the first being a direct rendez-vous transfer between Earth and Mars; the second is a spiralling orbit rising for a low Earth orbit. In both cases the results obtain confirmed the soundness of the proposed approach. I. INTRODUCTION The design of a low-thrust (LT) trajectory requires the definition of the thrust profile that satisfies a twopoint boundary value problem. The scope of this work is to provide a computationally efficient way to determine a good approximated solution to this problem with a more accurate representation of the control profile compared to other approximations. In the literature the problem has been tackled in a number of different ways generally classified in two families: indirect methods and direct methods. Indirect methods translate the design of a low-thrust trajectory into the solution of an optimal control problem and derive explicitly the associated first order optimality conditions. The first order optimality conditions are a system of mixed differential-algebraic equations (DAE). Shooting, multiple-shooting, collocation and approximated analytical approaches have been proposed to solve the DAE system and satisfy the boundary conditions. Direct methods, instead, do not derive the optimality conditions but transcribe the differential dynamic equations of motion into a system of algebraic equations and then solves a nonlinear programming problem. Numerical integration and collocation techniques have been proposed to transcribe the differential dynamic equations. Direct methods are generally computationally intensive while indirect methods can display some convergence problem. Both require some form of first guess solution. In the past decade, some low-fidelity approximation techniques have been proposed to generate the first guess solution. However, the use of these low-fidelity solutions is not always straightforward. Sims and Flanagan 1 first proposed a fast direct method based on the transcription of a lowthrust trajectory into a multi-burn transfer to generate a medium-fidelity solution at a low computational cost. This has also been used as a basis for global optimization tools 2,3. Recently, Sukhanov et. al. 4 proposed a method in which multi-revolution LT trajectories are divided into sub-arcs and on each of them an linearised optimal control problem is solved. In this paper, a direct method is presented where the trajectory is decomposed into a number of finite elements. Gauss planetary equations are solved over IAC-1-C1.7.5 Page 1 of 11

each element by means of a perturbative approach, for constant thrust modulus and direction. The trajectory is assumed to be an ε-variation of a Keplerian arc, where ε is a small acceleration term due to the low-thrust. A very fast transcription of the trajectory into a nonlinear programming problem is thus obtained, the accuracy of which is controlled by the number of elements, assuming that every trajectory element remains a first order epsilon-variation of a Keplerian arc. We will show how this approach can be used for fast multiobjective optimization of long low-thrust spirals where both the mass of propellant and the transfer time need to be minimized. II. MODEL AND PROBLEM DEFINITION The equations of perturbed two body motion are usually expressed in terms of Cartesian coordinates or, alternatively, in Keplerian orbital elements but it is also possible to rewrite them in terms of non-singular, equinoctial elements 5. This is a particular formalization of the 6-element state vector which is not affected by singularities for orbit types like those with zeroinclination (undefined line of nodes) or zero-eccentricity (undefined periapsis). This is achieved by replacing e, i, Ω, ω and θ with derived quantities: a P1 e sin ( ω = Ω + ) P2 = e cos( Ω + ω) i X = Q1 = tan sin Ω [1] 2 i Q2 = tan cos Ω 2 L = ( Ω + ω) + ϑ The equations of perturbed two body motion are expressed in terms of a set of 6 Gauss variational equations in Equinoctial parameters: 2 da 2a = ( P2 sin L P1 cos L) cos cos dt h ε β α p + ε cos β sinα r dp1 r p = cos L ε cos β cosα dt h r p + P1 + 1+ sin L ε cos β sinα r P ( Q cos L Q sin L) ε sin β 2 1 2 } dp2 r p = cos L ε cos β cosα dt h r p + P2 + 1+ sin L ε cos β sinα r P ( Q cos L Q sin L) ε sin β 3 dt a h 1 1 2 dq1 r = ( 1+ Q1 + Q2 ) sin L ε sin β dt 2h dq2 r = ( 1+ Q1 + Q2 ) cos L ε sin β dt 2h dl µ r = Q cos L Q sin L ε sin β 1 2 where ε, α and β are modulus, azimuth and elevation of the perturbing acceleration in the radial-tangential reference frame. In a typical LT transfer problem, one wants to find the control law conditions: } [2] which satisfies the boundary ( t ) X = X [3] X( t f ) = X f With t f =t +ToF, where ToF is the time of flight. At the same time, the aim is usually that of minimizing the total V of the transfer: V t f = ε t dt [4] t III. THE PERTURBATIVE APPROACH The system of nonlinear ordinary differential equations (ODE s) expressed in [2] can be easily integrated by means of a numerical approach, but this is rather demanding from the computational point of view, especially when massive computation on many different candidate control profiles is necessary, as in the framework of the solution of orbit transfer optimization problems. On the converse, a perturbative approach, which employs low-thrust non-dimensional acceleration as the perturbation parameter, can be used in order to analytically obtain an efficient, first-order accurate approximate solution for the evolution of orbital parameters under the action of a constant perturbing force in the radial-tangential reference frame 6. Gauss's variational equations [2] are rewritten using true longitude L as the independent variable instead of the time, by means of the derivation chain rule, where dx dx dt = [5] dl dt dl The variable x can be any of the first 5 equinoctial orbital elements, namely, a, P 1, P 2, Q 1, and Q 2, while dt/dl is given by IAC-1-C1.7.5 Page 2 of 11

2 3 dt r h = = dl h P L P L ( 1+ sin + cos ) 1 2 2 [6] This latter expression also provides for the additional equation needed for completing the model, in order to derive time as a function of L. At this point, a first order expansion for each equinoctial orbit parameter (plus time) is introduced in [7], where the zero-order terms obtained for ε = (that is, Keplerian motion) are expressed by constants of the motion, with the only exception of t, which clearly depends on L. a = a + ε a 1 P = P + ε P 1 1 11 P = P + ε P 21 Q = Q + ε Q 1 1 11 Q = Q + ε Q 21 [7] t = t + ε t1 The remaining first-order terms, which approximate the effects of the perturbation action on the trajectory, are determined by means of a technique based on standard perturbation theory 7, applied to the set of first order ODEs [2]. In order to analytically solve the relevant equations, the solution of a set of integrals in the form LF 1 I1 ( LF ) = dl L 3 1+ P sin L + P cos L C F L L 1 2 LF cos L I L = dl [8] ( 1+ P sin L + P cos L) 1 2 LF sin L I L = dl S F ( 1+ P sin L + P cos L) 1 2 is necessary, that must be evaluated between the initial and final angular coordinate values. A closedform solution is available in the complex domain but its derivation is not straightforward and it is omitted here for the sake of conciseness. As an example, the solution for I S is reported in [9]. Once the analytical expressions for a 1 (L), P 11 (L), P 21 (L), Q 11 (L), and Q 21 (L) are available, together with t (L) and t 1 (L), for the considered values of the initial conditions along the transfer arc, it is possible to evaluate all the parameters for a known control force (constant in magnitude and orientation in the radialtangential reference frame), both forward and/or backward in the angular coordinate L. This means that the five variation of orbital parameters and time is known as a first order approximated function of the true longitude L for low-thrust trajectory arcs. Their values can be analytically evaluated from an arbitrary set of 3 3 initial conditions and control force components, expressed in terms of magnitude and two angles. L sin L IS ( L) = dl = 2 1+ P sin L + P cos L 1 2 ( + ) 1 2 1 P P 1 P cos L P1 ln + ( P2 1) sin L + P1 + P2 1 P1 = + 3/2 [9] ( P1 + P2 1) ( + ) 1 2 1 P P 1 P cos L P1 ln + ( 1 P2 ) sin L + P1 + P2 1 + P1 + + ( P1 + P2 1) 3/2 ( P2 + 1) cos L + P1 sin L + P2 + 1 ( P1 + P2 1)( P2 cos L + P1 sin L + 1) The length of the arc along which it is possible to propagate the evolution of a 1, P 11, P 21, Q 11, and Q 21 depends on the requested accuracy. If the accuracy is fixed (e.g. below one percent) the approximated trajectory arc must be smaller when the ratio ε between low-thrust force and gravitational pull grows, so that in case of low-thrust propulsion applied far away from the main body (e.g. in a Heliocentric transfer scenario) the approximated arc must be reduced to 1 to 2 deg for matching the accuracy conditions. On the converse, when low-thrust is applied close to a primary body, as in a geocentric transfer, the ratio ε becomes smaller, and the considered perturbative expansion may provide acceptable results also for propagations over several orbits. IV. FINITE ELEMENTS TRANSCRIPTION The analytical solution found for Gauss planetary equations is included in a Direct Finite Perturbed Elements in Time (FPET) Transcription method. This could be regarded as a variant of the well established Sims & Flanagan Direct Transcription Method 1. The latter used a zero-order approximation of the perturbed Keplerian motion by decomposing the trajectory into n sub-arcs; each subarc was modelled with unperturbed Keplerian motion. The effect of continuous thrust was modelled with V discontinuities at the boundaries of each sub-arc. This had the advantage of being simple and computationally cheap since the solution for Keplerian motion is obviously available in closed form. In the present work, the simple Keplerian model with discrete V impulses is replaced with a perturbed Keplerian model with constant thrust along each sub-arc (see Fig. 1). IAC-1-C1.7.5 Page 3 of 11

Fig. 1: LT Multiple Shooting Method. Every trajectory is divided into n sub-arcs characterised by a constant thrust vector in the radialtransversal frame. The i-th arc of amplitude L i is defined by the following quantities: X m, the 6 Equinoctial parameters at the mid-point of the arc and the 3 control parameters ε, α and β. To obtain the boundary points of the Element the perturbed motion is analytically propagated backward and forward with the same amplitude ± L/2. The mid-point along the arc L was chosen as the base-point for the analytical propagation in the Finite Element for better accuracy, since the error increases superlinearly with arc amplitude. Thus, a two-sided propagation in the form + m L X = f X,, ε, α, β 2 [1] m L X = f X,, ε, α, β 2 provides a better accuracy than that obtained over the same total arclenght L by a forward expansion. The single sub-arcs are then interconnected to each other by imposing matching conditions at their boundaries. To test the accuracy of the Transcription Method with respect to the number of Finite Elements, in the following example we propagated analytically for a fixed time of 1.5 years a constant tangential acceleration -8 2 of 2.5 1 km / s, departing from 1AU. The trajectory was subdivided with a variable number of Finite Elements and for each case, the final state computed was compared with a simple implementation of a Modified Euler Method, on the same thrust profile. The results of a numerical integration with the results of numerical propagation with MatLab ode113 were used as a reference to compute the relative error on the final state. CPU times for all three propagation methods were also considered. Fig. 2: Error on final state for heliocentric orbit propagation. Fig. 3: Error on time of flight for heliocentric orbit propagation. Fig. 4: CPU cost for heliocentric orbit propagation. These test confirmed the first-order behaviour of the error of the analytic formulation and showed that it also guarantees a good accuracy already with a relatively small number of Finite Elements. Moreover, the present IAC-1-C1.7.5 Page 4 of 11

test should be regarded as a worst case since it involves a high perturbation force, roughly equivalent to the thrust of.5 N on a 2 kg spacecraft, continuously active for more than a year. This means that the ratio between the perturbative acceleration and the local gravity is relatively high and thus its accuracy is affected. Finally, the method has a computational cost only marginally higher than the Modified Euler method but still 4 to 7 times lower than the numerical integration with ode113. A similar test has also been carried out propagating a perturbed LEO orbit for 5 days, roughly equivalent to 5 revolutions. Fig. 7: CPU cost for LEO propagation. Fig. 5: Error on final state for LEO propagation. Fig. 6: Error on time of flight for LEO propagation. Here the advantages of the Analytic propagation are even more evident since it completely outperforms the Modified Euler Method, being 1 to 7 times faster that the Numerical Integration with ode113. This is easily explained by the fact that in LEO the gravitational force of the Earth is many times higher than the perturbation force and thus a first order approximation of perturbed Keplerian motion is adequate. It should also be noted that the Analytic propagation is able to give very good accuracy even with only one (or even a fraction of) Finite Element per revolution. This modelling approach can then be employed to solve Orbital transfer optimization problems. The problem under consideration is that of finding a V optimal trajectory from an initial state X and a final state X f, to be covered in a certain time ToF. The problem, formulated with the proposed Transcription method, translates into: min J = u nfpet i= 1 ε t X1 X [11] + s. t. Ceq = Xi Xi+ 1, i = 2,, nfpet 1 = + Xn X FPET f nfpet ToF t i i= 1 εi ε max, i = 1,, nfpet The decision variables are the thrust vector and the first 5 Equinoctial Elements of the midpoint of each arc. Equality constraints are given by matching condition between adjacent sub-arcs; boundary conditions on the i i IAC-1-C1.7.5 Page 5 of 11

initial and final state and condition that the time of flight be equal to a given value. Note that continuity conditions with respect to the Longitude are already automatically satisfied since all the analytical expressions for the variation of Equinoctial elements are already parameterised with respect to L. Therefore the matching constraints apply only to the remaining five Equinoctial Elements. Observe also that in this way the total longitude L tot covered by the trajectory arc could be easily determined as Ltot = Lf L. However, given the periodicity of the Longitude it is also straightforward to model additional revolutions by simply adding multiples of 2π to this figure. This obviously influences the amplitude of the single sub-arcs, given the relationship: L tot n FPET = i= 1 L i. Until now, only a uniform mesh (with respect to L) has been considered. The use of nonuniform grids will be addressed in future work. Limits on maximum engine thrust are formalised as limits on maximum perturbative acceleration: this is not entirely correct since, in fact, while the maximum thrust is constant the maximum achievable acceleration gradually increases with time due to a gradual decrease of spacecraft mass. However as a first approximation this is more than adequate and most of all greatly simplifies the optimization process since the acceleration limit is explicitly expressed as an upper bound on decision variables. This optimization problem could be efficiently solved with a Gradient-like optimizer such as MatLab fmincon. Given n FPET sub-arcs, the problem has 8n FPET decision variables and 5 n + 1 + 1 scalar equality FPET constraints. Given its computational efficiency it is also possible to extend the field of application by using the proposed transcription method to solve a Global, Multi-Objective (MO) Optimization problem for trajectory design. As a first example, a simple direct, rendez-vous transfer problem between a celestial body A and another celestial body B is considered; the aim is to find the optimal transfers with respect to Time of Flight and total V, within a certain range of departure dates and transfer time. The optimization parameters in this case are simply the departure date T, the ToF and the number of additional revolutions around the central body. A stochastic global optimizer then generates a number of candidate decision vectors. Then, for each value of the global decision vector an internal boundary problem (analogous to the one of [11]) is solved with a local optimizer to find the feasible transfer with minimum V. Boundary conditions are given by the ephemeris of body A at T and those of body B at (T +ToF). This internal process makes the Global Optimization rather CPU expensive but this is mitigated by employing the proposed Transcription Method. V. CASE STUDIES To test the proposed Transcription Method on realistic transfer problems, two cases were considered: the first is a direct, rendezvous transfer from Earth to Mars while the second is an orbit rising from a 26 km circular Low Earth Orbit (LEO) to the International Space Station (ISS) orbit. Both problems were solved first as a simple boundary problem and subsequently as a Multi-Objective problem using Epic, a population based Memetic algorithm developed by the authors and already described in previous works 8. V.I Low-Thrust Earth-Mars Spiral For the simple boundary problem solution, the objective is that of finding a V optimal transfer between Earth and Mars departing from Earth at T = 56 MJD2, with a time of flight of 365 3 days, and with 2 complete revolutions. Maximum 8 2 acceleration was set at 2.51 km / s, equivalent to a thrust of.5 N applied to a 2 kg spacecraft. Initial guess was given by a constant, tangential thrust profile of magnitude half the maximum acceleration. The orbit was modelled with 4 Finite Elements. The solution obtained has a total V of 5.64 km/s. To check solution accuracy, optimized thrust profile has been numerically integrated to calculate the final state, giving a relative error for the analytical solution of 3 1-3. Fig. 8: Variation of Equinoctial Elements for Earth- Mars LT transfer: a. IAC-1-C1.7.5 Page 6 of 11

Fig. 9: Variation of Equinoctial Elements for Earth- Mars LT transfer: P 1. Fig. 12: Variation of Equinoctial Elements for Earth- Mars LT transfer: Q 2. Fig. 1: Variation of Equinoctial Elements for Earth- Mars LT transfer: P 2. Fig. 13: Optimised trajectory for Earth-Mars LT Fig. 11: Variation of Equinoctial Elements for Earth- Mars LT transfer: Q 1. Fig. 14: Acceleration modulus ε for Earth-Mars LT IAC-1-C1.7.5 Page 7 of 11

accuracy and CPU cost. The Epic algorithm was run for a maximum number of 8 function evaluations. Fig. 17 and Fig. 18 report the solution points in the parameter space and the Pareto front obtained respectively. Fig. 15: Acceleration azimuth α for Earth-Mars LT Fig. 17: Parameters of the solutions for MO Earth-Mars LT transfer problem. Fig. 16: Acceleration elevation β for Earth-Mars LT Regarding the Thrust profile, is it possible to recognize a typical on-off with thrusting concentrated around the pericenter and apocenter passes. The same problem was solved with DITAN, an optimization tool specifically created for the design of Low-Thrust Trajectories 9. The output solution had a total V of 5.71 km/s, confirming the consistency of the V computed with the Perturbative approach. For the Multi-Objective test, the boundaries for optimization parameters in Table 1 were considered. Lower Upper T [MJD2] 5 5779.94 ToF [days] 1 15 n rev 1 3 Table 1: boundaries for optimization parameters for MO Earth-Mars transfer problem. The trajectory was modelled with 24 Finite Elements, chosen as a good compromise between Fig. 18: Pareto front for MO Earth-Mars LT transfer problem. The Pareto front presents some discontinuities, which are due to the presence of the discontinuous variable n rev. V.II LEO-to-ISS Orbit Rising The second test case considers an hypothetic orbit transfer between the Ariane 5ATV injection orbit and the ISS orbit, requiring an altitude increase of 95 km. The boundary problem is formulated in an analogous way as the Earth-Mars case, although the ISS motion is modelled as a simple, planar, keplerian motion and the injection orbit is assumed to be coplanar to the latter. The parameters of the departure LEO are considered fixed, with the exception of the initial. Departure time is also considered to be fixed, but the Time of IAC-1-C1.7.5 Page 8 of 11

Flight determines the position of the rendez-vous with the ISS. It is therefore essential to define the optimal phasing between the departure from LEO and the encounter with the ISS. As in the previous section, the simple boundary problem is considered first, optimizing the V for a transfer with ϑ = 21, ToF = 3.15 days and n rev = 32. 4 Finite Elements were use in the optimisation. The optimised trajectory has a V of 54.9 m/s. The accuracy is 1-5, a truly remarkable result considering the high number of revolutions. It should also be noted, that convergence was also very fast, with fmincon requiring only 3 iterations. Fig. 21: Variation of Equinoctial Elements for LEO-ISS LT transfer: P 2. Fig. 19: Variation of Equinoctial Elements for LEO-ISS LT transfer: a. Fig. 22: Variation of Equinoctial Elements for LEO-ISS LT transfer: Q 1. Fig. 2: Variation of Equinoctial Elements for LEO-ISS LT transfer: P 1. Fig. 23: Variation of Equinoctial Elements for LEO-ISS LT transfer: Q 2. IAC-1-C1.7.5 Page 9 of 11

number of additional revolutions was fixed to 3. The Epic algorithm was run for a maximum number of 3 function evaluations. Lower Upper [rad] 2π ToF [days] 1.84 1.97 Table 2: boundaries for optimization parameters for MO LEO-ISS transfer problem. Fig. 24: Acceleration modulus ε for LEO-ISS LT Fig. 27: Parameters of the solutions for MO LEO-ISS LT transfer problem. Fig. 25: Acceleration azimuth α for LEO-ISS LT Fig. 28: Pareto front for MO LEO-ISS LT transfer problem. Fig. 26: Acceleration elevation β for LEO-ISS LT The MO problem was formulated by considering two optimization parameters, and the ToF, with the lower and upper boundaries reported in Table 2. The VI. CONCLUSION In this work we have presented a novel numerical approach for Low Thrust trajectory modelling using a first-order analytical solution for Gauss s planetary in equinoctial elements. This allowed for a fast and relatively accurate propagation of perturbed Keplerian motion and was thus integrated in a Direct, Multiple Shooting Method for the solution of two-point boundary transfer problems. Given the computational efficiency of the method it was also possible to solve Global and IAC-1-C1.7.5 Page 1 of 11

Multi Objective Optimisation problems for LT trajectory design which involve the solution of many boundary sub-problems. 1 2 3 4 5 6 7 8 9 Sims, J.A., Flanagan, S. N.: Preliminary Design of Low-Thrust Interplanetary Missions, Paper AAS 99-338, 1999. Vavrina, M.A., Howell, K. C.: Global Low-Thrust Trajectory Optimization through Hybridization of a Genetic Algorithm and a Direct Method, AIAA/AAS Astrodynamics Specialist Conference pp. 18-21, 28. Yam, C.H. and Di Lorenzo, D. and Izzo, D.: Constrained Global Optimization of Low-Thrust Interplanetary Trajectories, Proceedings of the Twelfth conference on Congress on Evolutionary Computation (CEC 21), 21. Sukhanov, A. A., and Prado, A. F. B. de A.: Optimization of Transfers under Constraints on the Thrust Direction: II, Cosmic Research, Vol. 46, No. 1, 28, pp. 51 6. Battin, R.H.: An introduction to the mathematics and methods of astrodynamics, AIAA Education Series, 1987. Palmas, A.: Approximation of Low Thrust Trajectory Arcs by means of Perturbative Approaches, M.Sc. Thesis, Politecnico di Torino, July 21. Kevorkian J., Cole J.D.: Multiple Scale and Singular Perturbation Methods, Springer, New York, 1996. Vasile, M., Zuiani, F.: A Hybrid Multiobjective Optimization Algorithm Applied to Space Trajectory Optimization, Proceedings of the Twelfth conference on Congress on Evolutionary Computation (CEC 21), 21. Vasile, M.: Ditan User Manual Version 5.8.2, 29. IAC-1-C1.7.5 Page 11 of 11