Sie sind auf Seite 1von 96

LOW THRUST INTERPLANETARY

TRAJECTORY OPTIMIZATION
A postgraduate project report submitted to MAHE Manipal in
partial fulfilment of the requirement for the award of the degree of

MASTER OF TECHNOLOGY

in
AEROSPACE ENGINEERING

by

NAVEEN KUMAR S
Reg. No.: 160982003

under the guidance of

External Guide Internal Guide


Mr. Satyendra Kumar Singh Dr. Vidya S. Rao

Scientist, FDG Sr. Assistant prof, ICE Dept

URSC, Bengaluru MIT, Manipal

June 2018
DECLARATION

I hereby declare that the project work entitled Low Thrust Interplanetary

Trajectory Optimization is an original work and has been carried out by

me in the Flight Dynamics Group(FDG), U R RAO satellite centre (URSC)

under the guidance of Mr. Satyendra Kumar Singh, Scientist, FDG, URSC,

Bengaluru. No part of this work has been submitted for the award of a degree

or diploma either to this University or to any other Universities.

Place: Bengaluru

Date: 29-06-2018 [Naveen Kumar S]


Department of Instrumentation and Control Engineering

CERTIFICATE

Place:Bengaluru

Date: 29-06-2018

This is to certify that the project work entitled Low Thrust Interplanetary

Trajectory Optimization is a bonafide work carried out by Mr. NAVEEN

KUMAR S at U R RAO SATELLITE CENTRE (URSC), Bengaluru inde-

pendently under my guidance and supervision for the award of the Degree of

Master of Technology(M.Tech.) in AEROSPACE ENGINEERING of Manipal

Institute of Technology, MAHE, Manipal during the academic year 2017-2018.

Internal guide Head Of Department


Dr. Vidya S. Rao Dr. Dayananda Nayak

Sr. Assistant prof Professor and Head

Department of I&CE Department of I&CE

M.I.T. Manipal-576104 M.I.T. Manipal-576104


This work is dedicated to my parents
Soundar Rajan

and

Rajeswari
Acknowledgments

I take this opportunity to express a deep sense of gratitude for those who have

supported me in completing this project work on time.

I express my sincere thanks to Dr. M Annadurai, Director, URSC, for giving

me this opportunity to carry out this project at URSC. I would like to express

my thanks to Mrs. Dakshayani B P, Group Director, Flight Dynamics Group,

URSC and Mr. B S Kiran, Group Head, Flight Dynamics Group, URSC for

giving me this opportunity and permitting me to utilize the facilities in Flight

dynamics group.

I express the deep sense of gratitude to my guide Mr. Satyendra Kumar Singh,

Scientist, Flight Dynamics Group for guiding and supporting me throughout

the course of the project.

I show my immense gratitude to Dr. D Shrikanth Rao, Director, MIT, Manipal

and Dr.Dayananda Nayak, HOD, Instrumentation and Control Department,

MIT, Manipal for giving me an opportunity and for constant encouragement

to complete the project within specified time.

I also express my sincere thanks to Dr. Vidya S Rao, Sr. Assistant professor,

Instrumentation and Control Dept.,MIT, Manipal for encouraging and sup-

porting me throughout this course of the project.

I am also thankful to Dr. Surekha Kamath, Professor and Dr. Shreesha C,

Professor, Instrumentation and Control Dept., MIT, Manipal for providing me

the opportunity to carry out the project in this prestigious organization.

I would also thank Mrs. Pauline (URSC), FDG. I would also thank Mr. Man-

junath and Mr. Vishwanath.

Finally, I would like to express my thanks to my beloved parents, brother and

friends, who helped me in successfully completing this project.

Naveen Kumar S
Abstract

Trajectory optimization is a powerful technique to analyze mission feasibil-

ity during mission design. High-thrust trajectory optimization problems are

typically formulated as discrete optimization problems and are numerically

well-behaved. Low-thrust systems, on the other hand, operate for significant

periods of the mission time. As a result, the solution approach requires con-

tinuous optimization; the associated optimal control problems are in general

numerically ill-conditioned. In addition, case studies comparing the perfor-

mance of low-thrust technologies for space travel have not received adequate

attention in the literature and are in most instances incomplete. Electrical

propulsion systems broaden a class of spacecraft trajectories than conventional

chemical propulsion. But because low-thrust propulsion systems induce lim-

ited controllability for the spacecraft and increase overall trajectory transfer

duration. The optimization of continuous thrust trajectory remains a terrible

task. A variety of methods for trajectory optimization are available, however

the efficiency of an approach is dependent on the problem scenario it is applied

to.

This work is to design minimum-time and minimum-fuel interplanetary trans-

fer trajectory using optimal control theory employing low-thrust propulsion.

Both direct and indirect approaches are used in the study. several test cases

has been studied before solving the main problem. Different cases have been

studied such as single-link manipulator, Low thrust LEO (Low earth orbit) to

GEO (Geostationary earth orbit) and interplanetary case, Low thrust Earth-

Asteroid optimal transfers have been solved before proceeding to solve the

problem of Low thrust Earth-Venus optimal transfers.


Contents

Declaration i

Certificate ii

Acknowledgments iv

Abstract v

Table of Contents ix

1 Introduction 1

1.1 Trajectory Optimization for Low-Thrust Space Transfers . . . 3

1.2 Project Objective . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.3 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.4 Functional partitioning of project . . . . . . . . . . . . . . . . 5

2 Literature review 7

2.1 Indirect method optimization . . . . . . . . . . . . . . . . . . 8

2.2 Direct method optimization . . . . . . . . . . . . . . . . . . . 9

2.3 Hybrid method optimization . . . . . . . . . . . . . . . . . . . 10

2.4 The Three-Body Problem . . . . . . . . . . . . . . . . . . . . 11

2.5 Runge Kutta integrator . . . . . . . . . . . . . . . . . . . . . . 11

3 Optimal Control 13

vi
3.1 Unconstrained Optimization . . . . . . . . . . . . . . . . . . . 13

3.2 Constrained Optimization . . . . . . . . . . . . . . . . . . . . 14

3.2.1 Optimal Control Problem . . . . . . . . . . . . . . . . 16

3.2.2 Pontryagin’s Minimum Principle . . . . . . . . . . . . . 17

4 Trajectory Optimization Approaches 20

4.1 Mission Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . 20

4.2 Optimization theory . . . . . . . . . . . . . . . . . . . . . . . 21

4.3 Direct and Indirect methods . . . . . . . . . . . . . . . . . . . 22

4.3.1 Direct method . . . . . . . . . . . . . . . . . . . . . . . 22

4.3.2 General Formulation . . . . . . . . . . . . . . . . . . . 23

4.3.2.1 Collocation . . . . . . . . . . . . . . . . . . . 24

4.4 Indirect method . . . . . . . . . . . . . . . . . . . . . . . . . . 27

4.4.1 RK7(8) Integrator with fixed step . . . . . . . . . . . . 29

5 System Modelling 32

5.1 Dynamics Modeling . . . . . . . . . . . . . . . . . . . . . . . . 32

5.2 The N-Body Problem . . . . . . . . . . . . . . . . . . . . . . . 33

5.3 Equations of Motion in an Inertial Frame . . . . . . . . . . . . 35

5.4 Equations Of Relative Motion . . . . . . . . . . . . . . . . . . 40

5.5 Problem Formulation of Low Thrust Trajectory Optimization 41

5.5.1 Minimum-Fuel Problem . . . . . . . . . . . . . . . . . 42

5.5.2 Minimum-Time Problem . . . . . . . . . . . . . . . . . 46

5.6 Earth - Venus Optimal Transfers . . . . . . . . . . . . . . . . 47

5.7 Earth - Asteroid Optimal Transfers . . . . . . . . . . . . . . . 48

5.8 LEO to GEO Low Thrust Transfers . . . . . . . . . . . . . . . 49

5.8.0.1 Assumptions . . . . . . . . . . . . . . . . . . 50

5.9 Single-Link Manipulator Model . . . . . . . . . . . . . . . . . 50

5.9.1 Solution Using Constrained Optimization . . . . . . . . 51


6 Simulation and Result Analysis 53

6.1 Tools used for simulation . . . . . . . . . . . . . . . . . . . . . 53

6.2 Solving Optimal Control Problem With State Constraints (Single-

Link Manipulator) . . . . . . . . . . . . . . . . . . . . . . . . 53

6.2.1 Using Simulation tool (FMINCON) . . . . . . . . . . . 53

6.2.1.1 Simulation Results Using FMINCON . . . . . 54

6.2.2 Collocation . . . . . . . . . . . . . . . . . . . . . . . . 55

6.2.2.1 Simulation Results using collocation . . . . . 56

6.2.2.2 Simulation Results Using Collocation . . . . . 57

6.2.3 MATLAB BVP4C . . . . . . . . . . . . . . . . . . . . 57

6.2.3.1 Simulation Results Using BVP4C . . . . . . . 57

6.3 LEO to GEO transfer . . . . . . . . . . . . . . . . . . . . . . . 58

6.3.1 LEO to GEO transfer Using BVP4C . . . . . . . . . . 58

6.3.1.1 Simulation Results Using BVP4C . . . . . . . 59

6.3.2 LEO to GEO transfer Using Trajectory Optimizer . . . 60

6.3.2.1 Simulation Results Using Trajectory Optimizer 61

6.4 Earth-Asteroid Optimal transfers . . . . . . . . . . . . . . . . 61

6.4.1 Time-Optimal Earth-Asteroid rendezvous mission using

BVP4C . . . . . . . . . . . . . . . . . . . . . . . . . . 62

6.4.1.1 Time-Optimal Simulation results of Earth-Asteroid

Optimal Transfers . . . . . . . . . . . . . . . 63

6.4.2 Fuel-Optimal Earth-Asteroid Optimal Transfers . . . . 65

6.4.2.1 Fuel-Optimal Simulation results of Earth-Asteroid

Optimal Transfers . . . . . . . . . . . . . . . 66

6.5 Earth-Venus Optimal Transfers . . . . . . . . . . . . . . . . . 68

6.5.1 Time-Optimal Earth and Venus optimal Transfers . . . 68

6.5.1.1 Simulation Results Time-Optimal Earth and

Venus Optimal Transfers . . . . . . . . . . . . 69


6.5.2 Fuel-Optimal Earth and Venus Optimal Transfers . . . 71

6.5.2.1 Simulation Results Fuel-Optimal Earth and Venus

Optimal Transfers . . . . . . . . . . . . . . . 72

7 Conclusion and Future Scope 75


List of Tables

4.1 RKF 7(8) coefficients . . . . . . . . . . . . . . . . . . . . . . . 31

6.1 Initial co-states in Earth to asteroid optimal transfers . . . . . 65

6.2 Parameters for Earth to Venus mission . . . . . . . . . . . . . 68

6.3 Initial costates in Earth to Venus optimal transfers . . . . . . 72

x
List of Figures

1.1 Low thrust trajectories . . . . . . . . . . . . . . . . . . . . . . 3

3.1 Gradients of L and f must be colinear at the extrema . . . . . 16

4.1 Low thrust trajectories . . . . . . . . . . . . . . . . . . . . . . 25

5.1 N-Body System . . . . . . . . . . . . . . . . . . . . . . . . . . 35

5.2 (a)Two masses located in an inertial frame. (b)Free body dia-

grams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

5.3 Moving reference frame xyz attached to the mass of m1 . . . . 41

5.4 single link manipulator . . . . . . . . . . . . . . . . . . . . . . 51

6.1 single link manipulator model manipulated in SIMULINK . . 54

6.2 Optimal States . . . . . . . . . . . . . . . . . . . . . . . . . . 55

6.3 Optimal State x1 . . . . . . . . . . . . . . . . . . . . . . . . . 56

6.4 Optimal States . . . . . . . . . . . . . . . . . . . . . . . . . . 56

6.5 Optimal State x1 . . . . . . . . . . . . . . . . . . . . . . . . . 57

6.6 Optimal State x2 . . . . . . . . . . . . . . . . . . . . . . . . . 58

6.7 Low thrust Leo to Geo spiral . . . . . . . . . . . . . . . . . . . 59

6.8 Control history . . . . . . . . . . . . . . . . . . . . . . . . . . 60

6.9 Radius and Velocity Vs time plot . . . . . . . . . . . . . . . . 60

6.10 Low Thrust LEO to GEO Spiral . . . . . . . . . . . . . . . . . 61

6.11 Time-Optimal Trajectory . . . . . . . . . . . . . . . . . . . . . 63

xi
6.12 Switching vs Time(years) . . . . . . . . . . . . . . . . . . . . . 63

6.13 Co-state vs Time(years) . . . . . . . . . . . . . . . . . . . . . 64

6.14 Mass vs Time(years) . . . . . . . . . . . . . . . . . . . . . . . 64

6.15 Fuel-Optimal Trajectory . . . . . . . . . . . . . . . . . . . . . 66

6.16 Thrust and Switching vs Time(years) . . . . . . . . . . . . . . 66

6.17 Mass vs Time(years) . . . . . . . . . . . . . . . . . . . . . . . 67

6.18 Co-States vs Time(years) . . . . . . . . . . . . . . . . . . . . . 67

6.19 Time-Optimal Trajectory . . . . . . . . . . . . . . . . . . . . . 69

6.20 Switching vs t(years) . . . . . . . . . . . . . . . . . . . . . . . 70

6.21 mass vs t(years) . . . . . . . . . . . . . . . . . . . . . . . . . . 70

6.22 Co-states vs t(years) . . . . . . . . . . . . . . . . . . . . . . . 71

6.24 Switching and T vs t(years) . . . . . . . . . . . . . . . . . . . 72

6.23 Fuel-Optimal Trajectory . . . . . . . . . . . . . . . . . . . . . 73

6.25 Mass vs t(years) . . . . . . . . . . . . . . . . . . . . . . . . . . 73

6.26 Co-states vs Time(years) . . . . . . . . . . . . . . . . . . . . . 74


Nomenclature

m Mass of Body, kg

J Performance Index

H Hamiltonian

z Dynamic variable

r,v Position, km and Velocity Vectors, km/s

x,y,z Components of Position Vector, km

vx , vy , vz Components of Velocity Vector, km/s

λ Lagrange Multiplier

µ Mass parameter, km3 /s2

c Exhaust velocity, km/s

g Gravitational constant, km3 kg −1 s−2

h Step size

α Thrust magnitude

u Thrust vector

Tmax Maximum thrust, newton(N )

t0 , tf Initial time and Final time, s


Abbreviations

EP Electric Propulsion

Isp Specific Impulse

TBVP Two Point Boundary Value Problem

IVP Initial Value Problem

LEO Low Earth Orbit

GEO Geosynchronous Earth Orbit

GTO Geo Transfer Orbit

OCP Optimal Control Problem

PMP Pontrygian’s Maximum Principle

STM State Transition Matrix

ODE Ordinary Differential Equation

NLP Non-linear programming

EA Earth-Asteroid

EV Earth-Venus
Chapter 1

Introduction

Trajectory design for low-thrust propulsion systems presents unique new chal-

lenges to the mission design community. Perhaps not surprisingly, one ap-

proach to addressing these challenges is incorporating optimization methods

into the process[1]. The benefits that optimization methods provide are nu-

merous, but they also bring with them new challenges. The field is vast with

applications far beyond trajectory design, therefore selecting the appropriate

optimization method for a given problem can be challenging. A complete un-

derstanding of both the problem and the potential optimization approaches

is necessary to ensure an efficient design process. The optimization problem

is obtained via variation of an augmented performance index that includes

equations of motion and necessary state and control variable constraints[2][3].

The subject of spacecraft trajectory optimization has a long and inter-

esting history. The problem can be simply stated as the determination of a

trajectory for a spacecraft that satisfies specified initial and terminal condi-

tions, that is conducting a required mission, while minimizing some quantity

of importance. The most common objective is to minimize the propellant re-

quired or equivalently to maximize the fraction of the spacecraft that is not

1
Introduction

devoted to propellant. Of course, as is common in the optimization of con-

tinuous dynamical systems, it is usually necessary to provide some practical

upper bound for the final time or the optimizer will trade time for propellant.

There are also spacecraft trajectory problems where minimizing flight time is

the important thing, or problems, for example those using continuous thrust,

where minimizing flight time and minimizing propellant use are synonymous.

In recent years, pressure to reduce the costs of interplanetary missions has

led to an emphasis on designing missions with shorter flight times, smaller

launch vehicles and simpler flight systems and these requirements have re-

newed interest in low-thrust propulsion systems because of their high propel-

lant efficiencies. Thus the need to optimize their flight paths has posed certain

challenges.

Low-thrust systems possess little impulse power but may be operated con-

tinuously. Conventional chemical propulsion systems, on the other hand, pro-

vide short, but intense, bursts of thrust.

Electric propulsion systems use high specific impulse (Isp) values but can

only provide a few Newton of thrust due to power limitations of the space-

craft. Furthermore, the cost savings from using high Isp, low-thrust electric

propulsion instead of chemical propulsion is considerable

Using a spiraling, low-thrust trajectory will take more time than a Hohmann

type transfer. However, the high Isp values for electric thrusters means that

for given amount of fuel, more change in velocity can be provided than when

using chemical [4]. Because of the low thrust provided by electric thrusters

(less than 1 N typically), accelerations tend to be very low (on the order of

10−4 to10−6 g0 ), and so thrust times are long. For example, a transfer may be

on the order of weeks to months for performing a significant orbit change,

and the thrusters are fired for the entire duration of the transfer. Moreover,

a direct trajectory cannot be implemented using such low thrust, and so a

Aerospace, I&CE, MIT 2 URSC, Bengaluru


Introduction

spiraling transfer trajectory must be used[5]

Figure 1.1: Low thrust trajectories

The Figure 1.1 shows two types of spiraling low-thrust trajectories as well as

a high-thrust Hohmann transfer. One type of low-thrust trajectory shown is

a spiraling orbit raising in which the orbit remains relatively circular from an

initial low Earth orbit to some higher orbit. The other type of low thrust

trajectory starts on a highly elliptical orbit (provided by launch capabilities),

and uses a continuous low-thrust burn to spiral down to the lower, more cir-

cular orbit. Contrastingly, the chemical propulsion Hohmann transfer directly

inserts the spacecraft from the low orbit to the higher orbit without having to

orbit the Earth multiple times.

1.1 Trajectory Optimization for Low-Thrust

Space Transfers

One of the most important tasks during the analysis and design of space mis-

sions is the design and optimization of suitable mission trajectories. This the-

sis addresses the problem of designing optimal interplanetary trajectories for

Aerospace, I&CE, MIT 3 URSC, Bengaluru


Introduction

spacecraft equipped with low-thrust propulsion. Generally, optimality is de-

fined with respect to a set of mission constraints such as mission time and over-

all propellant consumption. For the case of electric propulsion, optimization

problems are in general less straightforward to formulate because for typical

applications both mission time and propellant consumption need to be taken

into account but are (in general) subject to competing mission constraints[3][6].

Compared to low-thrust spacecraft optimization problems, optimal control

problems for high-thrust systems are relatively straightforward. With high-

thrust spacecraft the duration of (control) thrust arcs is usually short in com-

parison to the mission time.

Low-thrust propulsion systems, on the other side, operate for a significant

part of the overall mission time. Consequently, the control variables need

to be modelled as continuous functions requiring tools based on continuous

optimization to solve optimal control problems adequately.

Several strategies have been suggested and used in the literature to solve con-

tinuous optimization problems for spacecraft transfer applications. Indirect

approaches based on the Calculus of Variations (COV) provide a means to

formulate optimal control problems as low-dimensional, discrete optimizations

problems. However, convergence characteristics of trajectory optimization

techniques based on the Calculus of Variations heavily depend on the “qual-

ity” of the initial guess of optimization parameters as well as the skill level of

the individual operating the optimization tool. As a consequence, the search

for optimized trajectories for continuous low-thrust spacecraft is typically a

rather tedious and time-consuming process.

Aerospace, I&CE, MIT 4 URSC, Bengaluru


Introduction

1.2 Project Objective

To design minimum-time, and minimum fuel low-thrust spacecraft transfers

for interplanetary mission using optimization approach. Before proceeding to

the main objective several case studies based on both the direct and indirect

approaches has been carried out. While solving case studies different methods

has been incorporated and after better understanding, the project objective

has been solved.

1.3 Motivation

The high impulse in low-thrust propulsion systems makes low-thrust trajectory

quite appealing. Also, the optimisation of low thrust trajectories is a demand-

ing task for the mission. However, low-thrust propulsion systems are, so far,

the most practical and efficient way to travel into space, they present limited

manoeuvrability. Their use increases mission duration. Hence this makes use

of low thrust propulsion complicated and challenging. Because of more com-

plexity involved in problems made it more interesting to solve interplanetary

trajectory optimization problem (optimal control).

1.4 Functional partitioning of project

1. mathematical modeling of the given system and definition of the suitable

performance index for the optimal control problem.

2. The suitable performance index, constraints and boundary conditions

are defined as per the given problem.

3. Using Pontryagin’s Minimum Principle that involves the formation of a

Hamiltonian H function.

Aerospace, I&CE, MIT 5 URSC, Bengaluru


Introduction

4. Optimality conditions are then applied on Pontryagin’s H function to

obtain the state and co-state equations.

5. The optimal control formulation of a dynamical system leads to a Two-

Point Boundary Value Problem (TPBVP). Methods like Shooting method

and the Collocation method are used to solve TPBVP.

6. Simulation of results.

Aerospace, I&CE, MIT 6 URSC, Bengaluru


Chapter 2

Literature review

The trajectory optimization problem is a problem which relates to designing

a trajectory that minimize or maximize an objective function. This function

can be the transfer time, the final mass along a path. Moreover, a solution tra-

jectory must satisfy the equations of motion for the spacecraft.In this chapter

we briefly summarize the works related in the field of trajectory optimization.

Generally, there are three different ways of generating optimal or near-

optimal low-thrust trajectories:

(1) Indirect optimization methods

(2) Direct optimization methods

(3) Hybrid optimization methods,

Each method has its own advantages and disadvantages.

Continuous low thrust propulsion for this work is more effective because

the gravity field of the several bodies is considered for trajectory optimization.

7
Literature review

2.1 Indirect method optimization

For solving the objective of the problem mainly focused on the work done

by Tieding Guo, Fanghua Jiang and Junfeng Li (Homotopic approach and

pseudospectral method applied jointly to low thrust trajectory optimiation).

The homotopic approach and the pseudospectral are two popular techniques

for low thrust trajectory optimisation. A hybrid scheme is proposed in this

paper [7]. where other work done by Chen Zhang, Francesco Topputo, Franco

Bernelli-Zazzera and Yu-Shan Zhao (Low-thrust minimum-fuel optimization in

the circular restricted three body problem) have solved low thrust minimum

energy, minimum fuel, and minimum time problem using indirect method. In-

direct methods have been successfully applied to a wide variety of low-thrust

trajectory optimization problems. Peng Zhan, JunfengLi, HexiBaoyin, and

GeshiTang (A low-thrust transfer between the Earth–Moon and Sun–Earth

systems based on invariant manifolds) low-energy,low-thrust transfer between

two halo orbits associated with two coupled three-body systems is studied in

this paper.The transfer is composed of a ballistic departure,a ballistic inser-

tion and a powered phase using low-thrust propulsion to connect these two

trajectories.It is based on the calculus of variation and Pontryagin’s minimum

principle. The optimization problem is obtained via variation of an augmented

performance index that includes equations of motion and necessary state and

control variable constraints. Shooting method is a well known indirect method,

where the initial values are guessed and numerical integration is performed to

arrive at the solution.

Fanghua Jiang, Hexi Baoyin, and Junfeng Li (Practical Techniques for

Low-Thrust Trajectory Optimization with Homotopic Approach) have solved

the problem which solves the fuel-optimal problem of low thrust trajectory

by starting from the related and easier energy-optimal problem. To this end,

Aerospace, I&CE, MIT 8 URSC, Bengaluru


Literature review

some effective techniques are presented to reduce the computational time and

increase the probability of finding the globally optimal solution[7].

Francesco Topputo and Franco Bernelli-Zazzera (Optimal Low-Thrust In-

variant Manifold Trajectories via Attainable Sets) has solved the problem

based on Indirect method, Planar low-energy low-thrust transfers to the moon,

as well as spatial low-thrust stable manifold transfers to halo orbits in the

Earth–moon system, are presented[8].

2.2 Direct method optimization

Early research efforts to develop trajectory optimization tools based on direct

methods date back to the 1980’s.The paper by Betts gives a brief overview

of the most common and popular numerical methods to solve trajectory opti-

mization problems which is of direct method. A Survey of Numerical Methods

for Trajectory Optimization paper which gives the depth knowledge on the op-

timal control problem. Direct optimization strategies discretize the continuous

optimal control problem thereby reformulating it as a nonlinear programming

problem(NLP), and, thus making it tractable to a wider range of numerical op-

timization approaches. Collocation methods are one of the primary techniques

for solving the direct optimization problems[5][9].

Peter J. Edelman and Kshitij Mall has worked on Optimal Low-Thrust

LEO to GEO Circular Orbit Transfer, in which they have found the optimal

transfer trajectory from low earth orbit (LEO) to geosynchronous-Earth orbit

(GEO), and they have assumed this problem we assume a planar transfer

with a spherical, non-rotating Earth, and neglect other external forces due to

the Moon, the Sun, and solar radiation pressure they have solved using the

BVP4C of Matlab code which is of using collocation method which solves the

problem using higher order polynomials. Direct collocation methods work by

Aerospace, I&CE, MIT 9 URSC, Bengaluru


Literature review

approximating the problem dynamics and control variable using polynomial

splines. These methods sometimes referred to as direct transcription [10].

In 1995, Sean Tang and Bruce Conway has solved the problem on Low-

Thrust interplanetary trajectory where using the collocation and nonlinear

programming, where this above method is applied to determine the minimum

time low thrust inter planetary trajectory[11].

Ashish Tewari book on advanced control aircraft, spacecraft and rockets

where he has mentioned in detail of simple collocation method for solving

optimal control problems using collocation method[12].

2.3 Hybrid method optimization

The direct methods discussed in Section 2.2 are based on the idea of locally

approximating polynomials for the state, costate, and control variables. In

recent years the concept of using globally orthogonal polynomials has gained

in importance and received increasing attention in the literature. Globally

orthogonal polynomials offer an alternative approach of transforming trajec-

tory optimization problems into nonlinear programming problems and provide

the means to conveniently reformulate optimal control problems as problems

involving systems of purely algebraic equations. One of the benefits of using

globally orthogonal polynomials functions such as the Legendre and Cheby-

shev polynomials lies in the fact that the same order of accuracy for the states

and costates is guaranteed, which is usually not the case for transcription

methods using a locally approximating polynomial function description. As a

result, costate information is in general available to relatively high accuracy

providing a “good” initial set of optimization parameters for subsequent usage

in indirect methods[13].

Aerospace, I&CE, MIT 10 URSC, Bengaluru


Literature review

2.4 The Three-Body Problem

A description of the time-dependent behaviour of three gravitationally inter-

acting bodies was first mathematically formalized in 1687 by Isaac Newton.

Overtime, the formulation of this problem has come to be denoted the three-

body problem Curtis and Bate1971[14].

The optimal control problem can be stated as finding the optimal control

history that satisfy the system dynamics, while satisfying the constraints.

In the circular restricted three-body problem, two massive bodies move in

circular orbits around their common centre of mass, and the third mass is

negligible with respect to the other two with respect to a rotating reference

frame, the two co-orbiting bodies are stationary, and the third can be station-

ary as well at the Lagrangian points, or move around them, for instance on a

horseshoe orbit. It can be useful to consider the effective potential[15].

The equations of motion of a circular restricted three-body problem is given

by

µ1 µ2
(ẍ − 2ω ẏ − ω 2 x) = − 3
(x + π2 r12 ) − 3 (x − π1 r12 )
r1 r2
µ1 µ2
(ÿ + 2ω ẋ − ω 2 y) = − 3 y − 3 y (2.1)
r1 r2
µ1 µ2
z̈ = − 3 z − 3 z
r1 r2

2.5 Runge Kutta integrator

In numerical analysis, the Runge–Kutta methods are a family of implicit and

explicit iterative methods, which include the well-known routine called the

Euler Method, used in temporal discretization for the approximate solutions

of ordinary differential equations.

T. Feagin in his work the difference between the results of orders eight

Aerospace, I&CE, MIT 11 URSC, Bengaluru


Literature review

and ten can be used to estimate the local truncation error and thus to vary

the step size. Numerical experiments demonstrate that the method compares

favourably with other high-order embedded methods.

Runge-Kutta-Nystrom formulas of the eigth, seventh, sixth, and fifth order

were derived for the general second order (vector) differential equation written

as the second derivative of x = f(t, x, the first derivative of x). The formulas

include a stepsize control procedure, based on a complete coverage of the

leading term of the local truncation error in x, and they require no more

evaluations per step than the earlier Runge-Kutta formulas[16] for the first

derivative of x = f (t, x).

Aerospace, I&CE, MIT 12 URSC, Bengaluru


Chapter 3

Optimal Control

We consider first simple unconstrained optimization problems, and then demon-

strate how this may be generalized to handle constrained optimization prob-

lems. With the observation that an optimal control problem is a form of con-

strained optimization problem, variational methods are used to derive an op-

timal controller, which embodies Pontryagin’s Minimum Principle[11][17][18].

3.1 Unconstrained Optimization

Consider a function

L:<→< (3.1)

We want to find,

min L(u) (3.2)


u

Let us assume that L is sufficiently smooth, and consider the Taylor ex-

pansion:

dL dL2
L(u) = L(u0 ) + |(u=u0 ) (u − u0 ) + 2 |(u = u0 )2 + ... (3.3)
du du

13
Optimal Control

Then we have a necessary condition,

dL
|(u=u0 ) = 0 (3.4)
du
and a sufficient condition

dL2
|(u=u0 ) > 0 (3.5)
du2
Note that these are only conditions for a local minimum. Additional con-

ditions are required to find the global minimum if the function is non-convex.

If we have a function with more than one variable, that is L = <n → < we

have the following conditions.

( )
∂L ∂L ∂L
, , ..... |( u = u0 ) (3.6)
∂u1 ∂u2 ∂u1
and

 
∂2L ∂2L
∂U12
... ∂U1 ∂un 
∂ 2L

 
2
|( u = u0 ) = 
 ... ...  |( u = u0 ) > 0
...  (3.7)
∂u  
∂2L ∂2L
∂Un ∂u1
... ∂Un2

i.e., it is a positive definite.

3.2 Constrained Optimization

Consider the problem

minx L(x), f (x) = 0 (3.8)

with L : <n → <andf : <n → <m . Then this problem is equivalent to

min L(x) + λf (x) (3.9)


u,λ

Aerospace, I&CE, MIT 14 URSC, Bengaluru


Optimal Control

The function

H(x, λ) = L(x) + λf (x) (3.10)

is called the Hamiltonian of the optimization problem. The coefficients

λ ∈ <m are called Lagrange multipliers of the system.

Without loss of generality we consider x ∈ <2 . The necessary conditions

for a minimum are

∂H
=0 (3.11)
∂x, λ
or
∂H ∂L ∂f
= +λ
∂x1 ∂x1 ∂x1
∂H ∂L ∂f
= +λ (3.12)
∂x2 ∂x2 ∂x2
∂H
= f (x1 , x2 )
∂λ
The third condition is equivalent to the boundary conditions of the original

problem being satisfied.

The first two conditions are equivalent to saying that the vectors

   
∂L ∂f
 ∂x1   ∂x1 
 ;  (3.13)
∂L ∂f
∂x2 ∂x2

are parallel or collinear. If these vectors are parallel, then the matrix

 
∂L ∂f
 ∂x1 ∂x1 
  (3.14)
∂L ∂f
∂x2 ∂x2

has rank less than 2, which means that the linear system obtained by

equating to zero the derivative of the Hamiltonian has a non trivial solution

on λ. With the help of a diagram in Figure 3.1 it is easy to understand that

where we have a minimum or maximum the two gradients (with the red vector

Aerospace, I&CE, MIT 15 URSC, Bengaluru


Optimal Control

representing the gradient of L and the black vector representing the gradient

of f ) have to be parallel, as otherwise one can increase or decrease the value

of L while satisfying the constraint f (x) = 0.

Figure 3.1: Gradients of L and f must be colinear at the extrema

3.2.1 Optimal Control Problem

The optimal control problem is to find an optimal control input u∗ ∈ U ⊂ Rm

for a set of generally nonlinear, coupled differential equations of the form[19][11],

ẋ = f (x, u, t), t ∈ [t0 , tf ] (3.15)

subject to the boundary conditions

ψ(x(t0 ), x(tf ), t0 , tf ) = (ψ0 (x(t0 , t0 ), ψf (x(tf ), tf ))T = 0 (3.16)

and such that associated cost function


Z tf
J = ψ(x(tf ), tf ) + L(x, u, t)dt (3.17)
t0

is minimized. In general, constraints on state and control variables have to

be considered, as well. These constraints enter the optimal control analysis in

the form of inequality and equality constraints

σ(x, u, t) ≥ 0 (3.18)

Aerospace, I&CE, MIT 16 URSC, Bengaluru


Optimal Control

For example, for spacecraft applications the maximum available thrust (or

equivalently, the magnitude of the control input) presents a typical control

variable constraint, that is 0 ≤k u k≤ umax . Note that this two-sided control

variable constraint has to be rewritten as two separate one-sided constraints

to match the form of equation For a detailed analysis on the implementation

of state and control variable constraints we refer to Bryson[19].

3.2.2 Pontryagin’s Minimum Principle

Consider the system [17]

ẋ = f (x, u), x(0) = x0 (3.19)

with associated performance index

Z T
J[x0 , u(·)] = φ(x(T)) + L((x(t), u(t))dt (3.20)
0

and final constraint

ψ(x(T)) = 0 (3.21)

The following terminology is customary:

• J[x0 , u(·)] is called cost function.

• φ(x(T)) is called end constraint penalty.

• L(x, u) is called running cost.

• H(x, u, λ) = L(x, u) + λf (x, u) is called hamiltonian.

The Optimal Control Problem is: Find the control function

u : [0, T ] 7→ <m (3.22)

Aerospace, I&CE, MIT 17 URSC, Bengaluru


Optimal Control

such that the performance index is minimized and the final state constraint

and the system equations are satisfied.

Solutions of the Optimal Control Problem also solve the following set of

differential equations:

State Equation : ẋ = Hλ = f (x), (3.23)

∂L ∂f
Co − State Equation : −λ̇ = Hx = +λ (3.24)
∂x ∂x

∂L ∂f
Optimality Condition : 0 = Hu = +λ (3.25)
∂u ∂u

State initial condition : x(0) = x0 (3.26)

Co − state f inal condition : λ(T ) = (φx + ψx υ)|x (T ) (3.27)

where υ is the Lagrange multiplier corresponding to end condition given

by Equation 3.17.

We use the Lagrange multipliers to eliminate the constraints. Since the

main constraints are now given by a dynamical system ẋ = f (x, u) it is clear

that they must hold for all t and thus the vector function λ : [0, T ] → <n is a

function of time.

Using the notation H(x, u, λ) = L(x, u) + λf (x, u),the unconstrained min-

imization problem can be written as

Z T
J(x, u, λ) = φ(x(T )) + ψ(x(T ))υ + [L(x, u) + λ(f (x, u) − ẋ)]dt
0
Z T
(3.28)
= φ(x(T )) + ψ(x(T ))υ + [H(x, u, λ) − λẋ]dt
0

The differentials are written as

Aerospace, I&CE, MIT 18 URSC, Bengaluru


Optimal Control

J = [φ(x) + ψx (x)υ]δx|x=x(T ) +
Z T (3.29)
˙
[Hx δx + Hu δu + Hλ δλ − λδ˙ẋ + δλ(x)]dt + ψ(x)|x(T ) δυ
0

Note that by integrating by parts:

Z T Z T
− λδ ẋ = −λδx|t=T + λδx|t=0 + λ̇δxdt (3.30)
0 0

Furthermore, since x(0) = x(0) is constant, it holds that λδx|t=0 = 0. So

we can rewrite the previous expression as

δJ = [φx (x) + ψx (x)υ − λ]δx|x=x(T ) +


Z T (3.31)
[(Hx + λ̇)δx + (Hλ − ẋ)δλ + Hu δu]dt + ψ(x)|x(T )δv
0

Now for the function u : [0, T ] → <m to minimize the cost function,,

δJ must be zero for any value of the differentials have to be zero for every

t ∈ [0, T ]. This observation gives the equations as required.

Aerospace, I&CE, MIT 19 URSC, Bengaluru


Chapter 4

Trajectory Optimization

Approaches

In this chapter, we are discussing about the different optimization techniques

such as direct method and the indirect for low thrust propulsion, different

mission scenarios where this approaches will be used. before approaching the

problem some initial guesses are required while using the indirect approach.

to obtain those initial guesses we use differential evolution.

4.1 Mission Scenarios

We analyse optimization algorithms from the perspective of the general minimum-

time optimal control problem and with particular focus on time-optimal inter-

planetary transfers. Depending on specific mission objectives various solution

families can be identified. Of particular interest are the intercept problem, the

orbit transfer problem, and the rendezvous problem:[6]

P 0 Intercept problem: Minimum-time intercept with target object.

P 1 Orbit transfer problem: Minimum-time transfer between an initial

Keplerian and a target Keplerian orbit. There is no additional angular

20
Trajectory Optimization Approaches

end-point constraint.

P 2 Double orbit transfer problem: P 1 transfer with subsequent P 1

return transfer to initial orbit. There are no additional angular interior-

and end-point constraints.

P 3 Rendezvous problem: Minimum-time rendezvous with a target ob-

ject. The angular end-point constraint depends on the initial angular

separation between initial and target object, and the synodic period of

the target object.

P 4 Double rendezvous problem: Minimum-time transfer including a

target object rendezvous and subsequent rendezvous with initial object.

Problem families P 0 through P 4 represent optimal control problems with

increasing levels of complexity. While the intercept problem only involves a

constraint on the final position of the spacecraft, for rendezvous problems both

the final position and velocity of the spacecraft have to match final position

and velocity of the target planet. Accordingly, additional midpoint constraints

have to be satisfied for the double rendezvous problem.

4.2 Optimization theory

Continuous trajectory optimization problems are traditionally solved via direct

or indirect methods. A distinguishing feature of direct methods is that they

are numerically more robust than indirect methods. The “quality” of the

initial guess (closeness of initial guess from global minimum) is therefore not as

crucial as for indirect methods, which is often the decisive factor when opting

for direct methods. Both optimization approaches require an initial set of

optimization parameters an initial guess to initiate the optimization procedure.

Aerospace, I&CE, MIT 21 URSC, Bengaluru


Trajectory Optimization Approaches

The generation of a proper initial set of optimization parameters is usually the

most computationally expensive phase of an optimization algorithm.[20][3]

4.3 Direct and Indirect methods

Analytical solutions are available only for some special classical optimal con-

trol problems. In general, it is necessary to resort to numerical methods to

obtain solutions to the optimization problem. In the literature these methods

are classified into two categories: direct and indirect methods. As the names

suggest, direct methods directly solve for the unknown control variables. With

indirect methods the control variables are solved for indirectly via the associ-

ated Two-Point Boundary Value Problem.

4.3.1 Direct method

In direct algorithms the optimal control problem is transformed into a non-

linear programming problem, which is solved either via a penalty function

method or methods of augmented Lagrangian functions.

The basic idea behind direct approaches as realized in collocation methods

is to introduce a discretization of the time interval t ∈ [t0 , tf ]. The control

parameters of the nonlinear programming problem – the unknowns of the op-

timization problem – are the values of the state and control variables at the

grid points. For collocation methods, piecewise linear interpolating functions

between the grid points are chosen for the controls. The states are chosen to

be continuously differentiable and piecewise cubic functions. More advanced

discretization schemes use higher-order polynomial approximations and other

finite sum expansions to improve accuracy of the solution trajectory. However,

using more sophisticated discretization models also significantly increases nu-

merical difficulties, especially in the presence of path constraints. For a more

Aerospace, I&CE, MIT 22 URSC, Bengaluru


Trajectory Optimization Approaches

detailed discussion on the theory of direct methods we refer to Betts[5].

The major advantage of direct methods lies in their numerical robustness;

there is no equivalent to the numerically sensitive costate system as present

in indirect methods. One of the drawbacks of direct methods, on the other

hand, is that only approximate solutions are obtained. Increasing the number

of control parameters yields, in general, an improvement in the accuracy of

the solution. The increased number of unknowns, however, leads to a con-

siderable increase in computational complexity and computation time. An-

other disadvantage of direct methods is the existence of multiple minima (also

called pseudo-minima) as a result of the discretization process.Even though

pseudo-minima satisfy all necessary conditions for the optimal solution, the

corresponding parameter set may not be “close” to the parameter set of the

actual minimum.

4.3.2 General Formulation

• Formulate problem as collection of N phases, where

(k) (k)
t0 ≤ t ≤ tf0 (4.1)

• Dynamic variables  
(k)
y (t)
Z(k) =   (4.2)
(k)
u (t)

made of state variables y(k) (t) and control variable u(k) (t)

• Parameters p(k) independent of t.

• State equations

ẏ = f [y(t), u(t), p, t] (4.3)

Aerospace, I&CE, MIT 23 URSC, Bengaluru


Trajectory Optimization Approaches

• Boundary conditions

ψol ≤ ψ[y(t0 ), u(t0 ), p, t0 ] ≤ ψou ,


(4.4)
ψof ≤ ψ[y(tf ), u(tf ), p, tf ] ≤ ψof

• Algebric path constraints

gl ≤ g [y(t), u(t), p, t] (4.5)

• Variable bounds

yl ≤ y(t) ≤ yu ,

ul ≤ u(t) ≤ uu , (4.6)

pl ≤ p ≤ pu ,

• Optimal Control Problem: Find dynamic variables z and parameters p

subject to constraints and bounds, which minimize

(1) (1) (1) (1)


J = φ [y(t0 , t0 , y(tf , p(1) , t0 , .......
(4.7)
(N ) (N ) (N ) (N )
y(t0 , t0 , y(tf , p(N ) , t0 ]

4.3.2.1 Collocation

The collocation method approximates the solution by a linear combination of

a number of piecewise continuous polynomials that are usually defined on a

mesh of collocation points. The approximate solution is then substituted into

the system of ordinary differential equations such that the system is exactly

satisfied at each collocation point. The number of collocation points plus the

number of boundary conditions must equal the number of unknown coefficients

in the approximate solution[11].

Aerospace, I&CE, MIT 24 URSC, Bengaluru


Trajectory Optimization Approaches

Figure 4.1: Low thrust trajectories

The Figure 4.1 illustrates the discretization and collocation points for a

scalar differential equation. The most common choice of approximation is a

linear combination of spline functions. In order to achieve a given accuracy,

the collocation points must be carefully selected. Let us now consider a simple

example of the collocation method applied to solve a TPBVP.

consider a simple TPBVP for a second-order system:

ÿ = f (t, y, ẏ) (4.8)

y(t0 ) = 0, y(tf ) = yf (4.9)

We devise a simple collocation technique based upon a polynomial solution,

y(t) = a0 + a1 t + ....... + an tn (4.10)

where the coefficients,a0 , a1 , ..., an , are constants, and apply this approximate

solution over the single interval,t0 ≤ t ≤ tf , such that the boundary conditions

at both ends are met and the governing differential equation is satisfied at both

the ends as well as at an interior point y(t1 ), t0 ≤ t1 ≤ tf . Thus we write

Aerospace, I&CE, MIT 25 URSC, Bengaluru


Trajectory Optimization Approaches

y(t0 ) = a0 + a1 t0 + ...... + an tn0 = y0 ,

y(tf ) = a0 + a1 tf + ...... + an tnf = yf ,

y(t1 ) = a0 + a1 t1 + ...... + an tnf = y1 , (4.11)

˙ = a1 + 2a2 t + ...... + nan tn−1 ,


y(t)
¨ = 2a2 + 6a3 t + ........ + n(n − 1)an tn−2 = f (t, y, ẏ).
y(t)

The solution, [y(t1 ), y(t˙ 1 )], at interior point t = t1 (called the collocation

point), is obtained by solving the IVP with initial condition [y(t0 ), y(t˙ 0 )], in

the interval t0 ≤ t ≤ t1 , in order to determine the values of the polynomial

coefficients. One can choose a fifth-order polynomial (n = 5) which is forced

to satisfy the boundary conditions, slopes, and differential equation at both

ends of the control interval:

   



 a0 






 y0 




 
 
 

   


 a1




 y˙ 
0


 
 
 


   
a2  y¨0 
   
=C (4.12)
a3  y¨f 

  
 

 
 
 


 
 
 

   



 a4






 y˙f




   
  
a5 
  y˙f 
 

where

 
1 t0 t20 t30 t40 t50 
 
0
 1 2t0 3t20 4t30 5t40 

 
0 0 2 6t0 12t20 20t30 
 
C=


 (4.13)
0
 0 2 6tf 12t2f 3
20tf 
 
0
 1 2tf 3t2f 4t3f 5t4f 

 
1 tf t2f t3f t4f t5f

Aerospace, I&CE, MIT 26 URSC, Bengaluru


Trajectory Optimization Approaches

By using an IVP solver that generates the solution, y(t1 ), at an interior

point, t = t1 , from the initial condition, [y0 , y˙0 ], we can evolve an iterative

solution strategy wherein the error in satisfying the differential equation at

the single collocation point, t = t1 ,

 = [t¨1 − f [t1 , y(t1 ), t˙1 ]] (4.14)

is brought to within a specified tolerance by suitably changing the initial

condition.

4.4 Indirect method

In indirect methods, an optimal solution is found by satisfying optimality con-

ditions instead of minimizing a cost criterion directly as in direct methods.

Depending on the given optimal control problem, the optimality conditions

lead to a two-point or multi-point boundary value problem (MPBVP). It con-

sists of differential equations for the continuous state and adjoint variables,

an algebraic equation for the continuous control, and boundary conditions for

the continuous states, the adjoint variables, and time. In some very special

cases with purely continuous, linear systems and quadratic costs, an analytic

solution of the BVP can be found. In most cases, however, it is necessary to

iteratively approach a solution with numerical methods.

Indirect methods are based on the calculus of variation and Pontryagin’s min-

imum principle. The optimization problem is obtained via variation of an

augmented performance index + that includes motion equations (constraints)

and – when necessary state and control variable constraints. The motion equa-

tion constraints are associated with the vector of co-states λ ∈ Λ ⊂ <n ; the

infinite-dimensional optimization problem is transformed into a n-dimensional

(and therefore discrete) optimization problem for the unknown co-states. Note

Aerospace, I&CE, MIT 27 URSC, Bengaluru


Trajectory Optimization Approaches

that for the unconstrained (with respect to path and control constraints)

minimum-time problem n = n[5].

Assuming the absence of path and control variable constraints for the time

being, the Hamiltonian function H . is obtained from the augmented perfor-

mance index as

H = L (x, u, t) + λT f (x, u, t) (4.15)

Under certain smoothness conditions the following first-order necessary

conditions for the state and control vectors are then obtained from the first

variation of + .

∂H
ẋ = = f (x, u, t) (4.16)
∂λ
and

∂H ∂L (x, u, t) ∂f (x, u, t)
λ̇ = − =− − λ (4.17)
∂x ∂x ∂x

which represent a Boundary Value Problem (BVP). In general, the boundary

conditions for the state variables are given; the boundary conditions for the

co-states, or more precisely, their initial λ(t0 ) or final values λ(tf ) are the

unknowns of the optimization problem. The optimal control law is determined

by minimizing the Hamiltonian function with respect to the control vector;

that is,

∂H
= 0, and
∂u
(4.18)
∂ 2H
> [0]
∂u2

Aerospace, I&CE, MIT 28 URSC, Bengaluru


Trajectory Optimization Approaches

∂2H
where ∂u2
> [0] denotes positive definiteness of a matrix. The optimal

control law u∗ can then be written as

u∗ = arg min H (x∗ , λ∗ , u), ∀t ≥ 0 (4.19)


u∈U

There exists a wide variety of techniques to solve the general Boundary Value

Problem. The most frequently used approach for trajectory optimization prob-

lems is based on the multiple-shooting method.

Contrary to most other techniques, the multiple-shooting method has the

distinct advantage that all kinds of constraint scenarios can be implemented

conveniently and exact solutions to the optimal control problems can be ob-

tained.

As mentioned in the previous section, the system of co-state equations is

extremely sensitive to variations in the initial conditions. Consequently, the

successful application of indirect methods heavily depends on the availability

of a “good” initial guess of the optimization parameters. Therefore, the first

phase of every optimization algorithm, the generation of an acceptable set

of optimization parameters is typically the most work-intensive and mathe-

matically complex stage of the optimization procedure. Another drawback of

indirect methods is that the switching structure of the constraints has to be

known a-priori. For a detailed review of solution techniques for constrained

optimal control problems[5][4].

4.4.1 RK7(8) Integrator with fixed step

In numerical analysis, RK method is an explicit method for solving ordinary

differential equations (Dormand and Prince 1980). The method is a member

Aerospace, I&CE, MIT 29 URSC, Bengaluru


Trajectory Optimization Approaches

of the Runge-Kutta family of ODE solvers. The difference between solutions

obtained from 7th and 8th order RKF 7(8) integrator is then taken as an error

estimate of the solution. This error estimate is used in adaptive step size inte-

gration algorithms. Other similar integration methods can also be used such

as RK4(5), RK5(6). The fixed step size h has been used in this RK7(8) inte-

grator. Embedded RKF 7(8) method with a fixed step size evaluates thirteen

functions of differential equations and are described below[16].

f1 = f (x0 , y0 )

k
X
fk = f (x0 + αk h, y0 + h βkλ fλ )(k = 2, 3, 4, 5, 6, ...., 13)
λ=0

k=6
X
y = y0 + h ck fk + 0(h8 )
k=1

k=6
X
ŷ = y0 + h ĉk fk + 0(h9 )
k=1

Error in the solution is estimated as

∆ = ŷ − y

where, λ, k, βkλ , ck , ĉk and αk are RKF 7(8) coefficients and are given in Table

4.1. The coefficients appearing in this table are not unique. The tables below

gives coefficients proposed by Fehlberg himself. f1 and fk are function eval-

uations of differential equations and are calculated using coefficients listed in

Table 4.1

Aerospace, I&CE, MIT 30 URSC, Bengaluru


Table 4.1: RKF 7(8) coefficients
k αk βkλ ck cˆk
41
1 0 0 840
0
2 2
2 27 27
0 0
1 1 1
3 0 0

Aerospace, I&CE, MIT


9 36 12
1 1 1
4 6 24
0 8
0 0
5 5 −25 25
5 12 12
0 16 16
0 0
1 1 1 1 34 34
6 2 2
0 0 4 5 105 105

31
5 −25 25 −65 125 9 9
7 6 108
0 0 108 27 54 35 35
1 31 61 −2 13 9 9
8 6 300
0 0 0 225 9 300 35 35
2 −53 704 −107 67 9 9
9 3
2 0 0 6 45 9 90
3 280 280
1 −91 23 −976 311 −19 17 −1 9 9
10 3 108
0 0 108 135 54 60 6 12 280 280
2383 −341 4496 −301 2133 45 45 18 41
11 1 4100
0 0 164 1025 82 4100 82 164 41 840
0
3 −6 −3 −3 3 6 41
12 0 205
0 0 0 0 41 205 41 41 41
0 840
−1777 −341 4496 −289 2193 51 33 12 41
13 1 4100
0 0 164 1025 82 4100 82 164 41
0 1 0 840

λ 1 2 3 4 5 6 7 8 9 10 11 12

URSC, Bengaluru
Trajectory Optimization Approaches
Chapter 5

System Modelling

5.1 Dynamics Modeling

Prior to any analysis and development of trajectory control strategies for the

motion of a spacecraft, a dynamical model must be constructed. A dynamical

model offers a mathematical description of the laws that govern the motion

and interaction of bodies. Over time, mathematicians and physicists have

improved the accuracy and efficiency with which dynamical models describe

the motion. However, the purpose is not always to describe the motion of

bodies with the greatest degree of accuracy; rather, simplified models that

roughly approximate the motion of bodies are often useful because their sim-

plicity allows for greater insight into the essential interactions occurring within

a system. Simplified models, for example the two or three body problems, are

constructed given a set of reasonable assumptions. Once a dynamical model

is available, examination of the system for integrals and equilibrium solutions

is useful for understanding the underlying structure of the solution space[14].

32
System Modelling

5.2 The N-Body Problem

The most general dynamical model to incorporate all gravitational forces as

point mass sources is the N-Body problem. This model was formally intro-

duced in 1687 by Issac Newton in his ground breaking work PhilosohphaeNat-

uralis Principae Mathematica.

Newton introduced his three laws of motion that serves as the founda-

tion for much of modern dynamics. The law of motion states that the force

impressed on a body is proportional to, and in the same direction, as the

derivative of the body’s momentum. The law is expressed mathematically in

vector form as,[15]

F = m r̈ (5.1)

where F is the vector sum of all forces acting on the particle mass m and r̈ is

the vector acceleration of the mass as observed from an inertial reference frame.

Note in above equation, that m is a constant scalar, thus, this relationship

applies only to fixed-mass systems. Elsewhere within the Principia, Newton

formulated his Universal Law of Gravitation.


GM md
|F| = −
(5.2)
d3
This model for the gravitational force on a single particle mass m due

to the existence of mass M when the relative distance between the bodies

is d. A single particle mi is located within a system of N other bodies, as

demonstrated in Figure below. Note that the position of mi relative to particle

mj is denoted dji . Thus, the force on mi due to the existence of particle j is

directed as described by the direction −dji /dji The total gravitational force

acting on particle i is then obtained by summing the individual gravitational

forces. This generalized form of Newton’s Universal Law of Gravitation is,

Aerospace, I&CE, MIT 33 URSC, Bengaluru


System Modelling

n
X mi mj
Fi = −G 3
dji (5.3)
j=1,j6=1
d ji

where j = i is excluded from the summation because the body obviously

cannot exert a force on itself.

Newton’s model for gravity and the law of motion are combined to produce

the differential equation of motion for particle i in a system of N bodies,

n
X mi mj
mi r̈ = −G dji (5.4)
j=1,j6=1
d3ji

as in Figure 5.1, the vector r̈ in equation 5.4 is the position vector from

an inertially fixed origin to the mass mi , while the vector dji is the vector

extending from mj to mi . Thus, the relative displacement dji is given by

dji = ri − rj .

Collecting differential equations of the form in equation 5.4 for each par-

ticle in an N-body system can become quickly intractable when N is large.

The simplest non-trivial case of this is the two-body problem, a focus for

mathematicians for hundreds of years. This simple model allows for analytical

solutions, some of which were described by Johann Kepler and his predecessors

even before the time of Newton.

The closed form solutions in the two body problem are readily applicable

to celestial mechanics and render reasonably accurate approximations for the

motion of many celestial bodies. Additionally, these solutions were especially

useful when computational capabilities were much more limited. However, the

rapid advancement of computing power over the last 75 years, has enabled fea-

sible examination of motion in more complex dynamical models and a new and

diverse set of tools for understanding gravitational interactions has emerged.

Aerospace, I&CE, MIT 34 URSC, Bengaluru


System Modelling

Figure 5.1: N-Body System

5.3 Equations of Motion in an Inertial Frame

Figure 5.2 shows two point masses acted upon only by the mutual force of

gravity between them. The positions R1 and R2 of their centres of mass

are shown relative to an inertial frame of reference XY Z . In terms of the

coordinates of the two points

R1 = X1 Iˆ + Y1 Jˆ + Z1 K̂
(5.5)
R2 = X2 Iˆ + Y2 Jˆ + Z2 K̂
The origin O of the inertial frame may move with constant velocity (relative

to the fixed stars), but the axes do not rotate. Each of the two bodies is acted

upon by the gravitational attraction of the other. F12 is the force exerted on

m1 by m2 , and F21 is the force exerted on m2 by m1 .

Aerospace, I&CE, MIT 35 URSC, Bengaluru


System Modelling

The position vector RG of the centre of mass G of the system in Figure 5.1

is defined by the formula:

m1 R1 + m2 R2
RG = (5.6)
m1 + m2
Therefore , the absolute velocity and the absolute acceleration of G are:

m1 Ṙ1 + m2 Ṙ2
vG = ṘG =
m1 + m2
(5.7)
m1 R̈1 + m2 R̈2
aG = R̈G =
m1 + m2
The adjective “ absolute ” means that the quantities are measured relative

to an inertial frame of reference.

Let r be the position vector of m2 relative to m1 . Then,

r = R2 + R1 (5.8)

Figure 5.2: (a)Two masses located in an inertial frame. (b)Free body diagrams

Or, using equation (5.3)

Aerospace, I&CE, MIT 36 URSC, Bengaluru


System Modelling

r = (X2 − X1 )Î + (Y2 − Y1 )Î + (Z2 − Z1 )Î (5.9)

Furthermore , let ûr be the unit vector pointing from m1 towards m2 , so

that

r
uˆr = (5.10)
r
where r is the magnitude of r,

p
r= (X2 − X1 )2 + (Y2 − Y1 )2 + (Z2 − Z1 )2 (5.11)

The body m1 is acted upon only by the force of gravitational attraction

towards m2 . The force of gravitational attraction, Sg , which acts along the

line joining the centers of mass of m1 and m2 ,

Therefore, the force exerted on m1 by m2 is:

Gm1 m2
F12 = uˆr (5.12)
r2
where uˆr accounts for the fact that the force vector F12 is directed from

m1 towards m2 . (Do not confuse the symbol G , used in this context to

represent the universal gravitational constant, with its use elsewhere in the

book to denote the center of mass.) By Newton’s third law (the action-reaction

principle), the force F12 exerted on m2 by m1 is −F12 , so that:

Gm1 m2
F21 = − uˆr (5.13)
r2
Newton’s second law of motion as applied to body m1 is F12 = m1 R̈1 ,

where R̈1 is the absolute acceleration of m1 . Combining this with Newton’s

law of gravitation (Equation 5.12) yields:

Gm1 m2
m1 R̈1 = − uˆr (5.14)
r2
Aerospace, I&CE, MIT 37 URSC, Bengaluru
System Modelling

Likewise , by substituting F21 = m2 R̈2 into Equation 5.12 we get:

Gm1 m2
m2 R̈2 = − uˆr (5.15)
r2
It is apparent from forming the sum of Equations 5.14 and 5.15 that m1 R̈1 +

m2 R̈2 = 0. According to Equation 5.7, that means the acceleration of the

center of mass G of the system of two bodies m1 and m2 is zero. Therefore,

as is true for any system that is free of external forces, G moves in a straight

line through space with a constant velocity vG .

The gravitational potential energy V of the force of attraction F between

two point masses m1 and m2 separated by a distance r is given by:

Gm1 m2
V = − (5.16)
r
A conservative force like gravity can be obtained from its scalar potential

energy function V by means of the gradient operator,

F = −∇V (5.17)

where, in Cartesian coordinates,

∂ ∂ ∂
∇= î + ĵ + k̂ (5.18)
∂x ∂y ∂z
For the two-body system in Figure 5.1 we have, by combining Equations

5.11 and 5.16

Gm1 m2
V = −p (5.19)
(X2 − X1 )2 + (Y2 − Y1 )2 + (Z2 − Z1 )2
The attractive forces F12 and F21 in Equations 5.11 and 5.12 are derived

from Equation (5.19) as follows

∂V ˆ ∂V ˆ ∂V

F12 = − ∂X2
I + ∂Y 2
J + ∂Z 2

∂V ˆ ∂V ˆ ∂V

F21 = − ∂X1
I + ∂Y 1
J + ∂Z 1

Aerospace, I&CE, MIT 38 URSC, Bengaluru


System Modelling

The gravitational force, outside of a sphere with a spherically symmetric

mass distribution M is the same as that of a point mass M located at the

center of the sphere. Therefore, the two-body problem applies not just to

point masses but also to spherical bodies (as long, of course, as they do not

come into contact!).

Let us return to Equations (5.14) and (5.15), the equations of motion of the

two-body system relative to the XYZ inertial frame. We can divide m1 out of

Equation (5.14) and m2 out of Equation (5.15) and then substitute Equation

(5.10) into both results to obtain

r
R̈1 = Gm2
r (5.20)
r
R̈2 = −Gm1
r
These are the final forms of the equations of motion of the two bodies in

inertial space. With the aid of Equations (5.5), (5.12) and (5.13) we can express

these equations in terms of the components of the position and acceleration

vectors in the inertial XYZ frame:

X2 − X1 Y2 − Y1 Z2 − Z1
Ẍ1 = Gm2 Ÿ1 = Gm2 Z̈1 = Gm2
r3 r3 r3
X1 − X2 Y1 − Y2 Z1 − Z2
Ẍ2 = Gm1 Ÿ2 = Gm1 Z̈2 = Gm1
r3 r3 r3
(5.21)
p
where, r = (X2 − X1 )2 + (Y2 − Y1 )2 + (Z2 − Z1 )2

The position vector r and velocity vector v of a particle are referred to

collectively as its state vector. The fundamental problem before us is to find

the state vectors of both particles of the two-body system at a given time given

the state vectors at an initial time.

Aerospace, I&CE, MIT 39 URSC, Bengaluru


System Modelling

5.4 Equations Of Relative Motion

Let us differentiate Equation (5.8) twice with respect to time in order to obtain

the relative acceleration vector,[15]

r̈ = R̈1 + R̈2 (5.22)

Substituting Equations (5.20) into the right side of this expression yields:

G(m1 + mz )
r̈ = − r (5.23)
r3
The gravitational parameter µ is defined as:

µ = G(m1 + m2 ) (5.24)

The units of µ are km3 s−2 . Using Equation (5.24) we can write Equation

(5.23) as:

µ
r̈ = − r (5.25)
r3
This fundamental equation of relative two-body motion is a non-linear

second order differential equation that governs the motion of m2 relative to

m1 . It has two vector constants of integration, each having three scalar

components. Therefore, Equation (5.25) has six constants of integration.

Aerospace, I&CE, MIT 40 URSC, Bengaluru


System Modelling

Figure 5.3: Moving reference frame xyz attached to the mass of m1 .

To begin, we imagine a non rotating Cartesian coordinate system attached

to m1 , as illustrated in Figure 5.3 . Resolve r̈ = − rµ3 r into components in this

moving frame of reference to obtain the relative acceleration components:

µ µ µ
ẍ = − x ÿ = − y z̈ = − z (5.26)
r3 r3 r3

5.5 Problem Formulation of Low Thrust Tra-

jectory Optimization

The dynamics of a spacecraft subject to sun’s gravity and low thrust propulsion

system and equations of motion is given by [10]

   
 ṙ   v 
   
ẋ = f (x, u,t) ⇒  = µ u (5.27)

   − r3
r + m


   
ṁ −uTmax /(Isp g0 )

State vector denoted by x = [x, v, m]T , the system dynamics in above

Aerospace, I&CE, MIT 41 URSC, Bengaluru


System Modelling

equation can be written as,

ẋ = f (x, u, t) (5.28)

where r and v are the spacecraft’s position and velocity vectors in the heliocen-

tric frame (HERF), respectively. The spacecraft’s mass and thrust are denoted

by m and u, respectively. Note all the above equations are in dimensionless

form, where dimension of space, time and mass are

1. Length unit, DU = 1AU ≈ 1.5 × 108 Km,

2. Time unit, T U = 1year = 3.15576 × 107 s,

3. Mass unit, M U = M0 (kg)

Here M0 is the initial mass of spacecraft and thus the dimensionless initial

mass m(t0 ) = 1. Other constants including µ, Isp , g0 are sun’s gravitational

constant, engine’s specific impulse and acceleration of gravity at sea level,

respectively, are all transformed to be consistent with the above specified di-

mensions.

The thrust vector u from (equation 5.26) is also denoted by u = (Tmax u)α

, where 0 ≤ u ≤ 1, ||α|| = 1,

where Tmax , u, α denote the maximum thrust magnitude, normalized thrust

magnitude and the direction of thrust.

5.5.1 Minimum-Fuel Problem

Typically the fuel optimal performance index J0 is written as

J0 = −m(tf ) (5.29)

The optimal thrust u should make the above index as small as possible.

In other words residual fuel m(tf ) should be large as possible. Note the per-

formance index J0 can be recasted in the following equivalent form (with a

Aerospace, I&CE, MIT 42 URSC, Bengaluru


System Modelling

constants ignored) if the mass of dynamics system in equation (5.26) is inte-

grated.

Z tf
Tmax
J0 = udt (5.30)
c ti

where ti and tf denotes initial and final times, To make the optimal tra-

jectory design problem well- defined, various boundary conditions should also

be specified, i.e.,initial and terminal conditions.For the optimal rendezvous

problem ,the initial conditions are

r(ti ) − ri = 0 v(ti ) − vi = 0 m(ti ) − 1 = 0 (5.31)

r(tf ) − rf = 0 v(tf ) − rf = 0 (5.32)

where where r0 , v0 and rf , vf are the initial and terminal states(position

and velocity),respectively.

Thus the basic task now is to design an optimal trajectory satisfying various

differential constraints in Eqs.(5.27), algebraic constraints in Eqs.(5.31),and at

the same time making the performance index in Eq.(5.30) as small as possible.

Optimal control and TPBVP

To derive the optimal control, the Hamiltonian H in the present problem

is written as

µ Tmax u  Tmax u  Tmax


H = λr .v + λv . 3 r + α + λm − + u (5.33)
r m Isp g0 Isp g0

where λ(t) = [λr , λv , λm ]T are the co-states associated with the state

equations (5.27). According to PMP, the optimal control can be denoted by

(u, α) = arg min H (5.34)


||α||=1,u∈[0 1]

Aerospace, I&CE, MIT 43 URSC, Bengaluru


System Modelling

for u ≥ 0, the optimal thrust direction α is thus obtained directly,

λv
α=− (5.35)
||λv ||
Substituting above thrust direction into the Hamiltonian H in Eq. (5.32),

one obtains the simplified Hamiltonian as

Tmax  ||λv ||Isp g0  µ


H= 1 − λm − u + λr .v − λv . 3 r (5.36)
Isp g0 m r

Eq. (5.33) and the fact that the state equations (5.27) and the Hamiltonian

H in Eq. (5.35) are all linear with respect to the normalized thrust magni-

tude u, whose admissible domain is closed, i.e., u ∈ [0, 1]. Thus the optimal

normalized thrust magnitude u is derived as bang-bang control

Switching detection technique

It has been mentioned that the right-hand side of Eq. (5.27) is discon-

tinuous. When numerical integration is performed, discontinuities cause the

integration error to accumulate at the switching points if these are not explic-

itly determined, which in turn affects the entire shooting process. A hybrid

switching detection technique combined with a variable-step integration has

been implemented to precisely detect the switching times [21].

where S is called the switching function is, defined as

||λv ||Isp g0
S = 1 − λm − (5.37)
m





 1 S<0


u= 0 S>0 (5.38)





u ∈ [0, 1]
 S=0
The co-state equations associated with state equation (5.27) are then de-

rived by

Aerospace, I&CE, MIT 44 URSC, Bengaluru


System Modelling

 
 
µ 3µ(λv .r)
 λ̇r  λv r3 − r5 r
∂H    
λ̇ = − → λ̇ v
= −λ r
 (5.39)
∂x   
  


Tmax ||λv ||u
λ̇m − m2
The co-state equation (5.39) are closely related with optimality and distin-

guish the optimal control approach from various direct methods in trajectory

optimization. Actually the state equation (5.27),and the co-state equation

(5.38) together define the complete dynamics of the optimal control problem.

The loss of clear optimality for most direct methods might be rooted in the

absence or ignorance of the co-state equations above.

Various boundary conditions for the co-states,i.e.transversality conditions,

should also be derived to make the problem well-defined. For the present

fuel-optimal rendezvous problem,

Let ψ0 (x(t0 )) = x(t0 ) − x0 and ψf (x(tf )) = x(tf ) − xf , v(tf ) − vf ]T

denote the initial and terminal constraint functions, respectively.

The boundary conditions for the co-states are all unknown except for λm .

As the final mass is free, its associated co-state must be zero at tf i.e.,

λm (tf ) = 0 (5.40)

Note that, in minimum-fuel problems ( = 0), the magnitude of u depends

only upon the sign of S, and a bang-bang profile is generated. Once the

optimal control variables α and u are determined as functions of the states

and co-states, through Eqs. (5.34) and (5.36), the motion can be integrated

implicitly with dynamics

Aerospace, I&CE, MIT 45 URSC, Bengaluru


System Modelling

   
 ṙ   v 
   
 v̇   − µ r + u 
   r 3 m 
   
 ṁ  −uTmax /(Isp g0 )
   
ẏ = F (y) → 
 =
   (5.41)
λ˙r   λv µ3 − 3µ(λ5v .r) r 

   r r 
   
λ˙   −λr 
 v  
   
Tmax ||λv ||u
ṁ − m2

where y = [x, λ]T is a 14-dimensional canonical variable. A two point

boundary value problem is defined by Eq. (5.40) together with the boundary

conditions [Eqs. (5.31) and (5.39)]. If the initial co-state vector λi was given,

one could integrate Eq. (5.40) and check if the final conditions are verified.

Let [x(t), λ(t)]T = ψ([xi , λi ]T , ti , t) be the solution of Eq. (5.40) in-

tegrated from[xi , λi ]T , from the initial time ti to the generic time t. The

optimization problem is stated as follows

Find λi such that [x(tf ), λ(tf )]T




r(tf ) − rf = 0





= ϕ([xi , λi ], ti , tf ) satisf ies v(tf ) − vf = 0 (5.42)





λm (tf ) = 0

5.5.2 Minimum-Time Problem

A minimum-time has to be solved to infer the minimum possible transfer time

tfmin for each Tmax ; The minimum-fuel problem [Eq. (5.41)] has to be solved

then for tf ≥ tfmin . The performance index of minimum-time problems is[21]

Z tf
Jt = 1 dt (5.43)
ti

therefore, the Hamiltonian reads

Aerospace, I&CE, MIT 46 URSC, Bengaluru


System Modelling

µ Tmax u  Tmax u 
Ht = λr .v + λv . 3 r + α + λm − +1 (5.44)
r m Isp g0
where the minimum time switching function is

c
St = −λv − λm (5.45)
m
when the transversality condition sets Ht (tf ) to zero.

The minimum time problem is stated as follows ;

Find (λi , tf ) such that [x(tf ), λ(tf )]T






 r(tf ) − rf = 0




v(tf ) − vf = 0

= ϕ([xi , λi ], ti , tf ) satisf ies (5.46)




 λm (tf ) = 0




H (t ) = 0

t f

5.6 Earth - Venus Optimal Transfers

From the equation (5.27) two body dynamics has been considered for the

Earth - Venus rendezvous mission where, the problem is of 3 dimensions, the

spacecraft position vectors r = [x, y, z]T and the spacecraft velocity vectors

v = [vx , vy , vz ]T . The control variables are the throttle factor, u ∈ [0, 1] , and

the thrust direction α. where, r in state and co-state equations is substituted

by below equation for this particular problem.[10]

p
r= x2 + y 2 + z 2 (5.47)

Consider Eq. (5.30) as the performance index for the Earth-Venus optimal

transfers, where the initial and terminal conditions of the mission will be given,

where the Hamiltonian for the 3 dimensional state and velocity vectors has to

implemented in Eq. (5.32).

Aerospace, I&CE, MIT 47 URSC, Bengaluru


System Modelling

In this case of Earth-Venus rendezvous there will be 7 state variables and

7 co-state variables, the values of all the constants are mentioned in Eq. 5.40,

where the initial conditions and final conditions of this case will be mentioned

in result analysis section.

For minimum time problem, The performance index for the minimum time

problem is mentioned in Eq. (5.42) where for this mission for both fuel optimal

and time optimal all the conditions remains same but only change will be with

the Hamiltonian and the switching function. The throttle factor u = 1 and no

switching will be detected in the minimum time problem, the thrusters will be

switched on in every condition.

5.7 Earth - Asteroid Optimal Transfers

Eq. (5.27) two body dynamics has been considered for the Earth - asteroid

optimal transfers where, in this case the problem is of 2 dimensions, for this

case position and velocity vectors of the spacecraft is given by r = [x, y]T and

v = [vx , vy ]T . The control variables are the throttle factor , u ∈ [0, 1] and the

thrust direction α.

where r for this case is given in below equation

p
r= x2 + y 2 (5.48)

In this case of Earth - Asteroid optimal transfer problem, there will be 5

state variables and 5 co-state variables and where the two conditions dynamic

system, performance index, initial and terminal conditions and switching func-

tion of both fuel optimal and time optimal along with the shooting function

has been discussed in detail in the above subsection of Minimum fuel and

Minimum time.

Aerospace, I&CE, MIT 48 URSC, Bengaluru


System Modelling

5.8 LEO to GEO Low Thrust Transfers

The problem which is of Optimal Low-Thrust LEO to geosynchronous-Earth

orbit(GEO) Circular Orbit Transfer. In this case we want to find the optimal

transfer trajectory to get from a 300 km altitude, low-Earth orbit (LEO) to

a 35,786 km altitude, geosynchronous-Earth orbit (GEO). In this problem we

assume a planar transfer with a spherical, non-rotating Earth, and neglect

other external forces due to the Moon, the Sun, and solar radiation pressure,

etc. Additionally we assume a constant thrust magnitude and a constant mass

flow rate [4].

The optimization problem is posed as follows: Maximize:

J = r(tf ) (5.49)

subject to:

ṙ = u
v2 µ T sinα
u̇ = r
− r2
+ m0 −|ṁ|t
(5.50)
v̇ = − uv
r
+ T cosα
m0 −|ṁ|t
v
θ̇ = r

r = Radius,km

u = Radial component of velocity, km/s

v = Tangential component of velocity, km/s

θ = Longitudinal angle, rad

α = Thrust angle, rad

T = Thrust magnitude, N

m0 = Initial vehicle mass, kg

ṁ = Mass flow rate, kg/s

µ = Gravitational parameter, km3 /s2

Aerospace, I&CE, MIT 49 URSC, Bengaluru


System Modelling

The optimization problem is posed as a maximum radius transfer rather

than a minimum time transfer with known initial and final orbit radii (due to

numerical difficulties associated with the latter).We obtain the time of flight

for the LEO to GEO transfer via iteration, by checking to see if the final orbit

radius achieved matches the radius at GEO at the end of each iteration.

5.8.0.1 Assumptions

1. The initial orbit is 300km altitude equatorial circular orbit (LEO).

2. The final target orbit is a 35785km altitude equatorial circular orbit

(GEO)

3. Hohmann transfer burns are impulsive and take zero time.

4. Low thrust burn is constant for the entire transfer duration.

5. All burns are planar in the equatorial plane.

6. The initial mass of each vehicle is 1500kg.

5.9 Single-Link Manipulator Model

Consider the following model of a single-link manipulator[22], which is illus-

trated in figure (5.4):

Aerospace, I&CE, MIT 50 URSC, Bengaluru


System Modelling

Figure 5.4: single link manipulator

g v 1
θ̈(t) = − sin(θ(t)) − 2 θ̇(t) + 2 u(t) (5.51)
l ml ml
where θ is the angular position, m is the mass of the end-of-rod element, l

is the length of the rod, is the friction coefficient at the pivot point, and u is

the applied torque at the pivot point.

Define x1 = θ, x2 = θ̇ and suppose that m = 2kg and l = 1m and

v = 6kgm2 /s . Then the model can be written as

x˙1 (t) = x2 (t)


(5.52)
x˙2 (t) = −9.8sin((x1 (t)) − 3x2 (t) + 0.5u(t).

5.9.1 Solution Using Constrained Optimization

Formulation as a Constrained Optimal Control Problem : For example, us-

ing Euler’s method for discretization with a step h = 0.05s it is possible to

define the following optimal control problem:

Aerospace, I&CE, MIT 51 URSC, Bengaluru


System Modelling

39
X 1
min J = 50(x1 (40) − 0.4)2 + [(x1 (k) − 0.42 ) + (u(k) − 7.632 )] (5.53)
u(k)
k=0
2

subject to

x1 (K + 1) = x1 (k) + hx2 (k)

x2 (k + 1) = x2 (k) + h[−9.8sin(x1 (k)) − 3x2 (k) + 0.5u(k)], k = 0, ...., 39

x(0) = [0 0]T

x2 (k) − 0.4 ≤ 0, k = 0, ......, 40

−10 ≤ u(k) ≤ 10, k = 0, ......, 39

Aerospace, I&CE, MIT 52 URSC, Bengaluru


Chapter 6

Simulation and Result Analysis

6.1 Tools used for simulation

The software used to simulate the results for the given problem statement

and for the different case studies is Matlab and Traj-Opt optimizer. Matlab’s

BVP4C has been used to obtain solution for some cases.

Software Toolbox ⇒ Matlab

Version ⇒ R2014a

6.2 Solving Optimal Control Problem With

State Constraints (Single-Link Manipula-

tor)

6.2.1 Using Simulation tool (FMINCON)

MATLAB Implementation Using Constrained Optimization : The implemented

solution is based on MATLAB’s function, which is part of the Optimization

Toolbox. Function f mincon implements a sequential quadratic programming

algorithm to solve non-linearly constrained optimization problems.

53
Simulation and Result Analysis

The M AT LAB BV P 4C function is used to compute the objective func-

tion f(x). A simulation of the system is carried out to evaluate the objective

function. Note that ’slm’ is a SIMULINK model of the single link manipulator,

which is illustrated in Figure 6.1. In Figure 6.1, notice the use of input and

output ports.

Figure 6.1: single link manipulator model manipulated in SIMULINK

The input port corresponds to the control signal, and the output ports

correspond with the states of the system. The sorting order of the output

ports is important. The solver used for integration was a fixed-step, fifth-

order Runge–Kutta method, with a step size of h = 0.05s.

6.2.1.1 Simulation Results Using FMINCON

In Figure 6.2 shows the plot of x1 and x2 versus time in seconds. Red line

indicates x1 and blue indicates x2.

Aerospace, I&CE, MIT 54 URSC, Bengaluru


Simulation and Result Analysis

Figure 6.2: Optimal States

6.2.2 Collocation

The collocation method approximates the solution by a linear combination of

a number of piecewise continuous polynomials that are usually defined on a

mesh of collocation points. The approximate solution is then substituted into

the system of ordinary differential equations such that the system is exactly

satisfied at each collocation point. The number of collocation points plus the

number of boundary conditions must equal the number of unknown coefficients

in the approximate solution.

The most common choice of approximation is a linear combination of spline

functions. In order to achieve a given accuracy, the collocation points must

be carefully selected. Let us now consider a simple example of the collocation

method applied to solve a TPBVP.

This problem has been solved by taking initial time t0 and final time tf

along with a and b , where y(t0 ) = a and y(t0 ) = b.

Aerospace, I&CE, MIT 55 URSC, Bengaluru


Simulation and Result Analysis

6.2.2.1 Simulation Results using collocation

This problem is solved using collocation scheme as mentioned in subsection(4.3.2.1)

with t1 = 0.5 and a tolerance,  = 10−7 by taking the initial and final time as

t0 = 0 and tf = 2sec and by taking a = 0 and b = 0.4. In figure (6.3), shows

of x1 versus time plot and in figure (6.4), shows x2 versus time plot.

Figure 6.3: Optimal State x1

Figure 6.4: Optimal States

Aerospace, I&CE, MIT 56 URSC, Bengaluru


Simulation and Result Analysis

6.2.2.2 Simulation Results Using Collocation

6.2.3 MATLAB BVP4C

The function bvp4c is a finite difference code that implements the three-stage

Lobatto IIIas formula. This method uses a collocation formula and the col-

location polynomial provides a C 1 -continuous solution that is fourth-order

accurate uniformly in [to, tf ]. Mesh selection and error control are based on

the residual of the continuous solution.

6.2.3.1 Simulation Results Using BVP4C

In Figure 6.5 and Figure 6.6 shows simulated results of x1 vs time in seconds

and x2 versus time in seconds.

Figure 6.5: Optimal State x1

Aerospace, I&CE, MIT 57 URSC, Bengaluru


Simulation and Result Analysis

Figure 6.6: Optimal State x2

6.3 LEO to GEO transfer

6.3.1 LEO to GEO transfer Using BVP4C

To find the optimal transfer trajectory to get from a 300 km altitude, low-

Earth orbit (LEO) to a 35,786 km altitude, geosynchronous-Earth orbit (GEO).

In this problem we assume a planar transfer with a spherical, non-rotating

Earth, and neglect other external forces due to the Moon, the Sun, and solar

radiation pressure, etc.

We require 2n + 2 boundary conditions, where n is the number of process

equations, for a well-defined TPBVP. For our problem, we require 2(4) + 2 =

10, and currently know 8. The known boundary conditions are:

initial conditions:
q
µ
t0 = 0s, r0 = 6678km, u0 = 0km/s v0 = r0
= 7.726km/s, θ0 =

0rad

Final conditions:

Aerospace, I&CE, MIT 58 URSC, Bengaluru


Simulation and Result Analysis

q
µ
tf = known, uf = 0km/s, ψ1 = V − f − rf
=0

where we have used µ = 3.986105km3 /s2 .

6.3.1.1 Simulation Results Using BVP4C

figure (6.7) shows the optimal trajectory from LEO to GEO.

Figure 6.7: Low thrust Leo to Geo spiral

figure (6.8) shows the control time history through out the mission.

Aerospace, I&CE, MIT 59 URSC, Bengaluru


Simulation and Result Analysis

Figure 6.8: Control history

figure (6.9) shows the change in radius and velocity with respect to the

time.

Figure 6.9: Radius and Velocity Vs time plot

6.3.2 LEO to GEO transfer Using Trajectory Optimizer

In optimizer the problem is solved using direct method. Trapezoidal colloca-

tion has been incorporated for propagating from the initial time to final time

by maximizing the radius. The accuracy obtained through this method is very

low when compared to which is obtained through the matlab bvp4c.

For the comparison purpose trajectories are obtained from two solvers and

Aerospace, I&CE, MIT 60 URSC, Bengaluru


Simulation and Result Analysis

analyzed. Figure 6.7 and Figure 6.10 shows the trajectories obtained using

BVP4C and trajectory optimizer.

6.3.2.1 Simulation Results Using Trajectory Optimizer

Figure 6.10: Low Thrust LEO to GEO Spiral

6.4 Earth-Asteroid Optimal transfers

This problem is to design a fuel-optimal optimal transfer trajectory from the

Earth to an asteroid. In dimensions introduced i.e.,DU, T U , and M U , the

initial and final states(position and velocity) are

Aerospace, I&CE, MIT 61 URSC, Bengaluru


Simulation and Result Analysis

x0 = [1.0, 0.0, 0.0, 2π],

xf = [−0.694328, −0.790895, 4.602579, −4.040613], respectively.

The transfer time is 240days.The specific impulse, maximum thrust

magnitude and initial total mass are 3000s, 0.135N and 1500kg, respectively.

6.4.1 Time-Optimal Earth-Asteroid rendezvous

mission using BVP4C

In this Time Optimal case switching function will be ,

k λv k Isp g0
S = −λm − (6.1)
m
In time optimal case thrusters will be switched on continuously, that means

control u = 1;

where in this case problem is solved by taking guess values for the co-states,

this problem is solved using matlab built in two point boundary value

problem solver BV P 4C.

The time optimal trajectory shown in Figure 6.11 in which red color

indicates the thrust arc, switching plot has been mentioned in Figure 6.12 it

always remains negative (i.e., thrusters ON throughout mission). How mass

changes with respect to time has shown in Figure 6.14 along with co-states

plot in Figure 6.13.

Aerospace, I&CE, MIT 62 URSC, Bengaluru


Simulation and Result Analysis

Figure 6.11: Time-Optimal Trajectory

6.4.1.1 Time-Optimal Simulation results of Earth-Asteroid

Optimal Transfers

Figure 6.12: Switching vs Time(years)

Aerospace, I&CE, MIT 63 URSC, Bengaluru


Simulation and Result Analysis

Figure 6.13: Co-state vs Time(years)

Figure 6.14: Mass vs Time(years)

Aerospace, I&CE, MIT 64 URSC, Bengaluru


Simulation and Result Analysis

6.4.2 Fuel-Optimal Earth-Asteroid Optimal Transfers

The switching function for Earth-Asteroid transfer is mentioned below

k λv k Isp g0
S = 1 − λm − (6.2)
m

In this case the problem is solved using Indirect approach where the co-states

values were taken from the reference paper to solve the problem. In this case

RK7(8) integrator has been used to propagate and obtain the result. The

fuel optimal trajectory in Figure 6.15 in which red colour indicates the coast

arc and green colour indicates thrust arc, switching plot has been mentioned

in Figure 6.16 it always remains negative (i.e., thrusters ON throughout

mission). How mass changes with respect to time in Figure 6.17 along with

co-states plot in Figure 6.18.

The co-state values for Fuel-Optimal Earth-Asteroid Optimal Transfers are

mentioned in Table 6.1

Table 6.1: Initial co-states in Earth to asteroid optimal transfers


λx -0.483033

λy 0.148713

λvx 0.050381

λvy -0.139474

λm 0.031785

Aerospace, I&CE, MIT 65 URSC, Bengaluru


Simulation and Result Analysis

Figure 6.15: Fuel-Optimal Trajectory

6.4.2.1 Fuel-Optimal Simulation results of Earth-Asteroid

Optimal Transfers

Figure 6.16: Thrust and Switching vs Time(years)

Aerospace, I&CE, MIT 66 URSC, Bengaluru


Simulation and Result Analysis

Figure 6.17: Mass vs Time(years)

Figure 6.18: Co-States vs Time(years)

Aerospace, I&CE, MIT 67 URSC, Bengaluru


Simulation and Result Analysis

6.5 Earth-Venus Optimal Transfers

A fuel-optimal and time-optimal low-thrust rendezvous mission from the

Earth to Venus, one typical three dimensional low thrust trajectory with

multi-revolutions, will be solved here. All the state dynamics have been

mentioned in section 5.5. The parameters for the mission are listed in Table

6.2.

Table 6.2: Parameters for Earth to Venus mission


Parameter value units

flight time 1000.0 days

Initial position [9.708322 × 10−1 , 2.375844× 10−1 , -1.670155× 10−6 ] DU

Initial velocity [-1.598191, 6.081958, 9.443368× 10−5 ] DU/TU

Final Position [-3.277178× 10−1 , 6.389172× 10−1 , 2.765929× 10−2 ] DU

Final velocity [-6.598211, -3.412933, 3.340902× 10−2 ] DU/TU

Isp 3800.0 s

Tmax 0.33 N

m0 1500.0 Kg

6.5.1 Time-Optimal Earth and Venus optimal

Transfers

In time optimal cases there switching will not be detected, and the control

will be always u = 1 that means thruster will be ’ON’ through out the

mission. The time optimal problem for this Earth-Venus transfers is solved

using the matlabBV P 4C by taking the random initial co-state values, if the

initial co-state values are not taken appropriately, the problem encountered

is singular Jacobian error. This error can be avoided by taking the scaled

co-state values.

The time optimal trajectory shown in Figure 6.19 in which red colour

Aerospace, I&CE, MIT 68 URSC, Bengaluru


Simulation and Result Analysis

indicates the thrust arc, switching plot has been mentioned in Figure 6.20 it

always remains negative (i.e., thrusters ON throughout mission). How mass

changes with respect to time has shown in Figure 6.21 along with co-states

plot in Figure 6.22.

6.5.1.1 Simulation Results Time-Optimal Earth and Venus

Optimal Transfers

Figure 6.19: Time-Optimal Trajectory

Aerospace, I&CE, MIT 69 URSC, Bengaluru


Simulation and Result Analysis

Switching vs t(years)
0

-50
S

-100

-150

-200
0 0.2 0.4 0.6 0.8 1 1.2
t(years)

Figure 6.20: Switching vs t(years)

Mass vs t(years)
1

0.95

0.9
m(kg)

0.85

0.8

0.75
0 0.2 0.4 0.6 0.8 1 1.2
t(years)

Figure 6.21: mass vs t(years)

Aerospace, I&CE, MIT 70 URSC, Bengaluru


Simulation and Result Analysis

Figure 6.22: Co-states vs t(years)

6.5.2 Fuel-Optimal Earth and Venus Optimal

Transfers

In fuel optimal cases the switching will be detected, the controller will be

u ∈ [0, 1] Thrusters will be switched to ’ON’ and ’OFF’ conditions according

to the switching values if switching value is positive, thrusters will be turned

’OFF’, switching values if switching value is negative thrusters will be turned

’ON’. The initial co-state value for this fuel-optimal problem was referred

from the reference paper[10].

The fuel optimal trajectory shown in Figure 6.23 in which red colour

indicates the coast arc and green colour indicates thrust arc, switching plot

has been mentioned in Figure 6.24 it always remains negative (i.e., thrusters

ON throughout mission). How mass changes with respect to time has shown

Aerospace, I&CE, MIT 71 URSC, Bengaluru


Simulation and Result Analysis

in Figure 6.25 along with co-states plot in Figure 6.26.

The co-state values for Fuel-Optimal Earth-Venus Optimal Transfers are

mentioned in Table 6.3

Table 6.3: Initial costates in Earth to Venus optimal transfers


λx 0.604140

λy 0.154961

λz 0.248134

λvx -0.024124

λvy 0.095894

λvz -0.020234

λm 0.141416

6.5.2.1 Simulation Results Fuel-Optimal Earth and Venus

Optimal Transfers

Figure 6.24: Switching and T vs t(years)

Aerospace, I&CE, MIT 72 URSC, Bengaluru


Simulation and Result Analysis

Figure 6.23: Fuel-Optimal Trajectory

Figure 6.25: Mass vs t(years)

Aerospace, I&CE, MIT 73 URSC, Bengaluru


Simulation and Result Analysis

Figure 6.26: Co-states vs Time(years)

Aerospace, I&CE, MIT 74 URSC, Bengaluru


Chapter 7

Conclusion and Future Scope

The detailed study of the optimal control problem has been carried out in

this work. The role of optimal control has been studied by solving some

problems based on low thrust Interplanetary transfers.

Two different types of solution approach has been considered in this work.

One is the direct method, where the optimal control problem is converted to

NLP and then solved. Second is the indirect method, where initial guesses of

co-states are used to arrive at the solution.

Recent research works on optimal control problem are based on the indirect

approach, because of the accuracy of the solution. In this work, the study of

a low thrust minimum time and low thrust minimum fuel trajectory designs

have been carried out. Using indirect approach low thrust minimum fuel

problems are solved and using direct method low thrust minimum time

problem are solved.

Before proceeding to the interplanetary transfer problem, to understand the

direct and indirect approaches to solve optimal control problem some case

studies have been done to solve and analyze the solution obtained. Several

complexities are involved in the problem such as the determining the

switchings in minimum fuel optimal trajectory design and are taken into

75
Conclusion

consideration appropriately. The fixed step RK7(8) has been used to detect

switching points and integrate the dynamics.

The future scope of this work

1). Low thrust time optimal transfer and fuel optimal interplanetary transfer

problem can be solved using the hybrid method because it will avoid the

complexity involved in selecting initial co-states values.

2). Detailed study of differential evolution can be carried out to obtain initial

co-states values.

Aerospace, I&CE, MIT 76 URSC, Bengaluru


Bibliography

[1] R. E. Pritchett, “numerical methods for low-thrust trajectory

optimization,” Ph.D. dissertation, Purdue University, 2016.

[2] G. LANAVE, “Waypoints Zemzev feedback space guidance for

multi-spiral, long-duration low thrust transfers,” Ph.D. dissertation,

University of Arizona, 2016.

[3] M. Kim, “Continuous Low-Thrust Trajectory Optimization: Techniques

and Applications,” Ph.D. dissertation, Virginia Polytechnic Institute

and State University, 2005.

[4] J. M. Longuski, J. J. Guzmán, and J. E. Prussing, Optimal Control with

Aerospace Applications. New York: Springer, New York, NY, 2014.

[5] J. T. Betts, “Survey of Numerical Methods for Trajectory

Optimization,” Journal of Guidance, Control, and Dynamics, vol. 21,

no. 2, pp. 193–207, 1998.

[6] J. Olympio, “Optimisation and optimal control methods for planet

sequence design of low-thrust Interplanetary transfer problems with

gravity-assists,” Ph.D. dissertation, 2009.

[7] F. Jiang, H. Baoyin, and J. Li, “Practical Techniques for Low-Thrust

Trajectory Optimization with Homotopic Approach,” Journal of

Guidance, Control, and Dynamics, vol. 35, no. 1, pp. 245–258, 2012.

77
References

[8] G. Mingotti, F. Topputo, and F. Bernelli-Zazzera, “Optimal Low-Thrust

Invariant Manifold Trajectories via Attainable Sets,” Journal of

Guidance, Control, and Dynamics, vol. 34, no. 6, pp. 1644–1656, 2011.

[9] J. T. Betts, “Very low-thrust trajectory optimization using a direct SQP

method,” Journal of Computational and Applied Mathematics, 2000.

[10] T. Guo, F. Jiang, and J. Li, “Homotopic approach and pseudospectral

method applied jointly to low thrust trajectory optimization,” Acta

Astronautica, 2012.

[11] B. A. Conway, Spacecraft trajectory optimization. New York:

Cambridge university press, 2010.

[12] A. Tewari, Advanced Control of Aircraft, Spacecraft and Rockets, 2011.

[13] D. P. Bertsekas, Dynamic Programming and Optimal Control, 2007.

[14] R. Bate, D. Mueller, and J. White, Fundamentals of Astrodynamics,

1971.

[15] H. D. Curtis, Orbital Mechanics for Engineering Students, 2014.

[16] P. J. Prince and J. R. Dormand, “High order embedded Runge-Kutta

formulae,” Journal of Computational and Applied Mathematics, 1981.

[17] D. S. Naidu, Optimal Control Systems, 2003.

[18] D. E. Kirk, Optimal Control Theory: An Introduction, 2004, vol. 17,

no. 3.

[19] A. E. Bryson and Y.-C. Ho, Applied Optimal Control: Optimization,

Estimation and Control. New York: Taylor and Francis group, 1975.

[20] S. Sreesawet, “a new algorithm to generate low-thrust spacecraft

trajectories,” Ph.D. dissertation, Kasetsart University, 2009.

Aerospace, I&CE, MIT 78 URSC, Bengaluru


References

[21] C. Zhang, F. Topputo, F. Bernelli-Zazzera, and Y.-S. Zhao,

“Low-Thrust Minimum-Fuel Optimization in the Circular Restricted

Three-Body Problem,” Journal of Guidance, Control, and Dynamics,

vol. 38, no. 8, pp. 1501–1510, mar 2015.

[22] V. Becerra, “Solving optimal control problems with state constraints

using nonlinear programming and simulation tools,” Education, IEEE

Transactions on, 2004.

[23] M. Kim, “Continuous Low-Thrust Trajectory Optimization: Techniques

and Applications,” Ph.D. dissertation, Virginia Polytechnic Institute

and State University, 2005.

[24] J. M. L. J. G. E. Prussing, Optimal Control with Aerospace

Applications. New York: Springer, New York, NY, 2014.

[25] R. Bertrand and R. Epenoy, “New smoothing techniques for solving

bang-bang optimal control problems - Numerical results and statistical

interpretation,” Optimal Control Applications and Methods, vol. 23,

no. 4, pp. 171–197, 2002.

[26] Z. Artstein, “Discrete and Continuous Bang-Bang and Facial Spaces Or:

Look for the Extreme Points,” SIAM Review, vol. 22, no. 2, pp.

172–185, apr 1980.

[27] R. P. Russell, “Primer vector theory applied to global low-thrust trade

studies,” in Advances in the Astronautical Sciences, vol. 124 I, 2006, pp.

857–876.

[28] K. C. Howell and M. T. Ozimek, “Low-thrust transfers in the

earth-moon system including applications to libration point orbits,” in

Aerospace, I&CE, MIT 79 URSC, Bengaluru


References

Advances in the Astronautical Sciences, vol. 129 PART 2, 2008, pp.

1455–1481.

[29] G. Strang, Introduction to linear algebra, 2003.

[30] J. T. Betts, “Survey of Numerical Methods for Trajectory

Optimization,” Journal of Guidance, Control, and Dynamics, 1998.

[31] C. Zhang, F. Topputo, F. Bernelli-Zazzera, and Y.-S. Zhao, “Low-Thrust

Minimum-Fuel Optimization in the Circular Restricted Three-Body

Problem,” Journal of Guidance, Control, and Dynamics, 2015.

Aerospace, I&CE, MIT 80 URSC, Bengaluru


PROJECT DETAILS

STUDENT DETAILS
Student Name Naveen Kumar S
Register No. 160982003 Section/Roll No. 3
Email Address naveensk7593@gmail.com Phone No. (M) 9036465607
PROJECT DETAILS
LOW THRUST INTERPLANETARY
Project Title
TRAJECTORY OPTIMIZATION
Project Duration 10 months Date Of Reporting 28/07/2018
ORGANIZATION DETAILS
Organization
U.R. RAO SATELLITE CENTRE (URSC)
Name
Full Postal Address
HAL Airport Road,Vimanapura Post, BENGALURU-560017
with Pincode
Website Address https://www.isac.gov.in
SUPERVISOR DETAILS
Supervisor Name Mr. Satyendra Kumar Singh
Designation Scientist
Full Contact
FDG, URSC, HAL Airport Road, Vimanapura Post,
Address
BENGALURU-560017
with Pincode
Email Address satyendra.0086@gmail.com Phone No. (M) 8971802350
INTERNAL GUIDE DETAILS
Faculty Name Dr. Vidya S Rao
Designation Sr. Assistant Prof
Full Contact Dept. of Instrumentation and control,
Address Manipal Institute of Technology,
with Pincode Manipal, Karnataka-576104
Email Address rao.vidya@manipal.edu Phone No. (M) 9686915089

Das könnte Ihnen auch gefallen