Repor-£06 Linear Quadratic Regulator design for an unpowered, winged re-entry vehicle February 1997 Ir. E. Mooij | ss ee oman f . J a =! se te _ TU D elft Faculty of Aerospace Enginooring Delft University of TechnologyLinear Quadratic Regulator design for an unpowered, winged re-entry vehicle Ir. E, MooijCopyright © 1997, by Deltt University of Technology, Faculty of Aerospace Engineering, Delt, The Netherlands. All rights reserved. No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without the prior written permission of the Delft University of Technology, Faculty of ‘Aerospace Engineering, Delft, The Netherlands. Publisher: Delft University of Technology ; Faculty of Aerospace Engineering P.O. Box 5058 2600 GB Delft The Netherlands. tel.: 015-2782058 fax: 015-2781822 Date February 1997 Report LR - 806 ISBN: 90-5623-040-9Organization: TUDILA/AZR Document code: LA-806 Date: February 1997_ Page: i Title Linear Quadratic Regulator design for an unpowered, winged re-entry vehicle Author(s) E. Mooij Abstract This report describes the design of an attitude controller for an unpowered, winged re-entry vehicle. The decoupling of the sym- metric and asymmetric motion makes it possible to design two separate controllers, one for the pitch motion and one for the lateral motion, The design of the controller, a Linear Quadratic Regulator, is based on linearisation of the equations of motion and feedback of the attitude and angular rates. The gains appearing in the control laws are computed by defining a quad- fatic cost criterion and then solving the matrix Riccati equation. Results of the study include the step and ramp response of the two separate controllers and the flight along the nominal trajec- tory with the integrated controller. The deviations from the nominal trajectory are acceptable, so this controller can be used for a detailed sensitivity analysis. Keyword(s) control-system design, linear state feedback, gain scheduling, Riccati equation, re-entry, HORUS-2B Date February 1997 Prepared Verified E. Mooi ass P.Ph. van den Broek ae ¥ Approved P.Th.L.M. van Woerkom z gto — Authorized EB P.G. Bakker p. £ hAter, etna entree ee(Organization: TUD/LAYAZR Document coda: LA-806 CHANGE RECORD Date: February 1997_, Page: i Issue | Rev. | Date Pages Topics Introduced No. | No. Changed! Added! Deleted ot or | 08-03-95 | - First version 02 | 10-01-96 | all Comments processed 03 | 02-04-96 | all Final draft o4 | 14-06-96 | All Comments processed 05 | 20-02-97] all Comments processed Rbintintent eect eeNotations ... Chapter 1 - Introduction .......-.++. 1.4. Background, 1.2. Attitude control of winged re-entry vehicles. 1.3. Attitude-control concepts. Table of Contents 1.4. Control-system design cycle. ......... Chapter 2 - The Motion of a Vehicle in a Planetary Atmosphere 2.1. The general form of the equations of motion. . 2.2. Linearisation of the equations of motion. 2.3. The state-space form of the equations of motion Chapter 3 - Open-loop Behaviour of the Re-entry Vehicle... 3.1 Chapter 4 - Design of the Controller . . . 44. 44, Introduction. 3.2. Nominal trajectory. 3.3. Eigenvalues and eigenmotion Introduction 4.2. The matrix Riccati equation. 4.3. Longitudinal controller. 43.4. 43.2. 43.3, 434, 435. Reduced system for symmetric motion. Root locus of the reduced system. Selection of pitch controls. Control laws ‘Computation of the feedback gains. Lateral controller. .. 44a. 442. 443. 444, Reduced system for asymmetric motion, Root locus of the reduced system. . Selection of the controls. Control laws. . .. 65 musenenenseesOrganization: TUDALR/AZR Date: February 1997_ Document code: LR-806 Page: iv 4.4.5. Computation of the feedback gains. ....... - 88 pee oS Chapter 5 - Verification of the Controller .......-..+++ 5.1 Introduction, 93 5.2. Root loci of the closed-loop systems 94 5.2.1. Longitudinal controller. 94 5.2.2. Lateral controller. 94 5.3. Step response. 99 5.3.1. Introduction. . . 99 5.3.2. Longitudinal controller. en Bosessseuoees CJ 5.3.3. Lateral controller. . oe 100 5.4. Ramp response. . .. pupsubcHosoqponeDs - + 103 5.4.1. Longitudinal controler. «=... cence ees 108 5.4.2. Lateral controller... . cic ee ees 104 5.5. Flight along nominal trajectory with integrated controller, | -.. 107 Chapter 6 - Conclusions and Recommendations ............ References «2.1.26. eceeeeeseeeeereeees Appendix A - Definition of State Variables ... Appendix B - Linear Stability Mode! of HORUS-2B . . . Appendix C - Selected Controller Gains ............45ROR RET Te =z 58 Notations element of state matrix speed of sound state or system matrix element of control matrix aerodynamic reference length control input matrix aerodynamic reference length output matrix drag-force coefficient roll-moment coefficient lift-force coefficient pitch-moment coefficient yaw-moment coefficient side-force coefficient drag direct transmission matrix acceleration due to gravity in radial direction acceleration due to gravity in meridional direction height identity matrix moment (product) of inertia cost criterion control gain control-gain matrix litt roll moment mass Mach number pitch moment Petetremmannnnenncenvarit {unit) ms rad Nm kg Nm(Organization: TUD/LF/AZR, Date: February 1997 _ Document code: LA-805 Page: vi7 nN’ yaw moment Nm Pp roll rate radis P period s P solution of matrix Lyapunov equation q pitch rate rad/s dyn dynamic pressure Nim? @ control deviation weight matrix r yaw rate radis R modulus of position vector m R control effort weight matrix Re equatorial radius m s side force N s square root of Rr ‘Sa aerodynamic reference area m? t time s t% half or doubling time s z Ty; roll-thruster moment Nm es pitch-thruster moment Nm qt yaw-thruster moment Nm u control vector uf independent control vector v modulus of velocity vector ms v Lyapunov function x state vector y output vector z modulus of a complex number : XYZ axes Greek a angle of attack rad B angle of sidestip rad Y flight-path angle rad 3 geocentric latitude rad ba aileron deflection angle rad 8, body-flap deflection angle rad 8 elevator deflection angle rad 3, rudder deflection angle rad By wing-flap deflection angle rad A perturbation : 5 damping ratio : 8 argument of complex number rad a eigenvalue -Organization: TUDILFVAZR Date: February 1997_ Document code: LA-B06 Page: vi u gravitation parameter m/s? ” eigenvector Pp (atmospheric) density kg/m? o bank angle rad t geocentric longitude rad x heading rad Oy rotational rate of the central body rad/s Oy eigenfrequency rad/s o rotation vector rad/s Indices oO nominal state a aileron b body flap B body frame c ‘commanded value cb central body e elevator f final, finite Pp in direction of roll rate q in direction of pitch rate r in direction of yaw rate r rudder rR in direction of position vector v in direction of velocity w elevon KZ along X-, Y- and Z-axis a in direction of angle of attack B in direction of angle of sideslip Y in direction of flight-path angle é in direction of latitude o in direction of bank angle x in direction of heading Abbreviations ACS Aerodynamic Control System com. centre of mass cPU Central Processing Unit dof. degree of freedom FCs Flight Control System nc Guidance, Navigation and Control eeOrganization: TUD/LRVAZR : Date: February 1997, Document code: LA-805 Page: vi Laa LaR MIMO MRAC PID RCS. START STR Linear Quadratic Gaussian Linear Quadratic Regulator Multiple Input, Multiple Output Model Reference Adaptive Control Proportional, integral and Derivative Reaction Control System Simulation Tool for Atmospheric Re-entry Trajectories Self-Tuning Regulator aChapter 1 Introduction During all phases in the design and operations of space vehicles, computer simulation of the flight performance plays an important role. Models of the vehicle and the environment can be simple, to get a first impression of the feasibility of a vehicle design or a particular mission, or can be very detailed when, for instance, the influence of aeroelasticity on the performance of an attitude-control system has to be studied. Within the framework of an ongoing research at the Faculty of Aerospace Engineering, Delft University of Technology, the development of a flight-simulation tool has been initiated (Mooij, 1994) with which, amongst others, the guided and controlled ascent of air-breathing space planes subjected to disturbances and model uncertainties can be analysed, The Simulation Too! for Atmospheric Re-entry Trajectories (START) has been selected to serve as a basis for further development. The original version of START did not include any guidance and control models, nor propulsion systems and the related variable mass properties. For a gradual development of START to a tool, capable of analysing the ascent missions that we mentioned before, several phases have been defined (Mooij, 1994). Current research focuses on the guided and controlled flight of unpowered, winged re-entry vehicles. A major step herein is the development of an attitude-control system. In this report we describe the design of such an attitude-control system for a selected unpowered, winged re-entry vehicle. Before we come to a discussion of attitude-control systems in this chapter, we begin by giving some background information on re-entry missions and how to increase mission success by incorporating guidance, navigation and control (Section 1.1). In the succeeding Section 1.2, we will present the attitude control of winged re-entry vehicles. Section 1.3 introduces several attitude-control concepts and one will be selected to employ as ‘our design (linear state feedback with gain scheduling). In Section 1.4, finally, the design process of the selected control system is detailed and an overview of the chapters of this report is given.‘Organization: TUD/LRVAZR. Date: February 1997 Document code: LR-808 Page: 1.4. Background. Two important aspects of the entry and descent of space vehicles are the tactical aspect of having control over the time and location of landing and the severe mechanical and thermal loading on the vehicle. For manned missions, for instance, the maximum deceleration is usually limited by an upper bound of 3 g to save the occupants from discomfort or worse, Also in case of unmanned (scientific) missions a too strong deceleration might be harmful to the on-board instruments. The thermal load, e.g., the maximum heat flux, the wall temperature in the stag- nation point and the integrated heat load, define to a large extent the design and therefore also the mass of the thermal protection system, so obviously we want to have the most favourable load on the vehicle. However, first of all a winged re-entry vehicle like the American Space Shuttle is supposed to land on an air strip, so its trajectory should be targeted to the landing location right on from entry. And, in case of the parachute descent of the Apollo capsules, it ‘was important that it would splash down in the ocean near the recovery ships. During the process of mission analysis and mission design, an optimal trajectory is computed which usually satisfies trajectory constraints (e.g., a maximum allowable thermal load) and end conditions (e.g,, the landing place in case of a re-entry mission). Once this so-called nominal trajectory has been defined it must be verified that the vehicle can actually fly this, trajectory, or, in other words, whether the vehicle can execute the required manoeuvres without violating any constraints. Furthermore, it must be guaranteed that the vehicle will still be able to fulfl its mission when it encounters (unforeseen) disturbances which make it deviate from its nominal path. To ensure mission success, the space vehicle is equipped with a so-called Guid- ance, Navigation and Control system (GNC system). The task of the guidance system is to generate steering commands, e.g., a commanded attitude or thrust level, taking a reference state, trajectory constraints and/or a final state into account. For this task, the system needs input from the outside worid, for instance the current actual state. These data have to be provided by the navigation system, using sensor informa- tion and predefined theoretical models. The control system has to take care that the steering ‘commands are carried out, such that, for example, the actual attitude equals the commanded attitude in a reasonably short time and that this attitude is dynamically stable (trim stability). To achieve this, the control system may drive aerodynamic contro! surfaces, reaction-control wheels and thrusters, etc. ‘The design of a GNC system is usually centred around a nominal mission that is free from disturbances. The environment is modelled at a certain level of complexity only and also the description of the vehicle is of course not infinitely accurate. An important question is whether the GNC system will be able to steer and control the vehicle in the presence of all kinds of uncertainties that it is bound to encounter during the actual mission. One way to assure this is. to design a very robust system’. But the next question that arises then is: how robust should " Robustness is defined as the particular property that a control system must possess for it to operate properly in realistic situations (Shahian and Hassul, 1993). Mathematically, it means that the controller must not ‘only work forthe system that it has been designed for, but for a whole family of (similar) systems. I the controller is stable fora set of systems whose parameters deviate substantially from the nominal system, then the controller is said to have robust stability, tOrganization: TUDILAVAZR Date: February 1997_ Document code: LR-B06 Page: 3° our system be? A too robust system might negatively influence the flying qualities and manoeuvrability and if the system is not robust enough, we can still end up with a severe control problem. It is therefore very important that already during the design process we study as many uncertainties as possible to see how our guidance and control system will deal with them. Usually what one does is, once the GNC system has been developed for the nominal mission, simulating a number of test cases with different error sources included, and with all dynamic, vehicle and environment models as accurate as possible. In Mooij (To BE PUBLISHED) such a sensitivity analysis of a GNC system is described. That analysis is centred around the HORUS-2B, an unpowered, winged re-entry vehicle that re- sembles the Space Shuttle (see also Fig. 1.1). Initially the HORUS was designed as an upper stage of the Ariane launcher. Later on, the concept was changed and it became the rocket- powered second stage of the German Sanger Two-Stage-To-Orbit space plane. A brief descrip- tion of the mission will follow below. f — Tl Fig. 1.1 - The HORUS-28 (MBB, 19882), After launch and orbital operations, it re-enters the atmosphere at an altitude of 120 km and the vehicle will begin its voyage back to its landing site, the European launch base in Kourou, French Guyana. The re-entry phase ends when HORUS is at a distance of about 80 km from the runway. It is said that the vehicle has reached the Terminal Area, which marks the begin- ning of the next mission phase in case of a winged re-entry vehicle. We will not study that phase, however. i tt | i i } terrane tenet rntententieentatOrganization: TUDILR/AZR Date: February 1997, Document code: LR-805 Page: The re-entry guidance system should track the prescribed trajectory as well as possible, while responding to disturbances and model errors by performing the appropriate manoeuvres. But the prescribed trajectory should not be followed at any cost because reaching the landing area without violating the constraints is the major mission objective. So another design goal is to limit the demands on the flight control system while being applicable to any reasonable ref- erence trajectory. Since HORUS is basically an unpowered vehicle, only two steering variables are available, i.e., the angle of attack a and the bank angle o (the angle of sidestip B is considered to be a disturbance, and will always be commanded to zero). For an efficient guid- ance, the nominal trajectory must provide sufficient margins for varying a and . This is also true for the path constraints, of course, and especially the thermal loads should remain well below the critical limits, A fundamental functional separation in the HORUS guidance is the subdivision of the hor- izontal and vertical entry guidance, corresponding with the symmetric and asymmetric motion of the vehicle. The vertical flight path is controlled by adjusting the angle of attack and the absolute value of the bank angle, while the sign of the bank angle is provided by the horizontal guidance (MBB, 1988b). To begin with the latter, this guidance controller steers the HORUS towards a targeting point at some 80 km from the runway. As a result from vertical guidance by means of the absolute bank angle, a lateral (or asymmetric) motion is introduced that needs to be compensat- ed for, such that the average bank angle is zero and the vehicle will keep on heading towards the target. The reference parameters for the horizontal guidance logic are the actual heading of the vehicle and the heading of a direct trajectory towards the targeting point. The difference between the two is the so-called heading error, which has to be kept sufficiently small. The controller uses a predefined heading-error dead band to change the sign of the bank angle the moment the heading error exceeds the dead-band value. The corresponding manoeuvre is known as a bank reversal, and results in so-called S-tuns as is also the case with the Space Shuttle. The objectives of the vertical guidance are to: + arrive at the Terminal-Area interface with a prescribed total energy and altitude (or velocity), + meet the flight-path constraints during the flight. To meet these objectives, the vertical guidance is divided into a separate energy and altitude control. The total energy, the sum of altitude-dependent potential and velocity-dependent kinetic energy, will be controlled such that only the final value at the Terminal-Area interface will be met with no direct effect on the constraints during the flight. The internal sharing of potential and kinetic energy, on the other hand, will affect the constraints through the altitude-velocity relation. The decrease of total energy is due to the working of the atmosphere on the vehicle, in the form of the drag force. So to influence the difference between actual and reference energy, the dissipated energy due to drag can be changed. Since one of the control parameters to change the drag of the vehicle is the angle of attack, energy control is accomplished by varying this, parameter. Altitude control is realised by variation of the vertical component of the lift force.(Organization: TUDILRVAZR Date: February 1997_ Document code: LA-806 Page: 5” Whereas the angle of attack influences the lift force in absolute sense, by changing the bank angle this force can be rotated about the velocity vector, thus resulting in two components per- pendicular to the trajectory, i.e., one vertical and one lateral. The vertical component defines the descent rate, so by changing the (absolute value of the) bank angle we can control the variation in altitude. Note that the lateral component of the lift results in a motion in the horizontal plane and thus a variation in heading. For this reason, bank reversals are required to keep the vehicle headed at the target without affecting the altitude control. 1.2. Attitude control of winged re-entry vehicles. Until today, there has only been one winged re-entry vehicle that has actually returned from orbit to Earth: the American Space Shuttle. Its Russian counterpart, the Buran, only made one unmanned (atmospheric) test flight before the project was cancelled due to budget problems. Also in Europe budget cuts were the reason that Hermes, a smaller version of the Space Shuttle, did not leave the drawing board and was cancelled. But whether the vehicle has actually been built or not, also the many publications on space vehicles can usually teach us about applied subsystems, such as the attitude-control system. Unfortunately, we could not lay hands on any literature concerning either the Buran control system, or the proposed Hermes attitude controller. Publicly-available documentation on the Space-Shuttle attitude controller was only sparsely available to us, and then in principle only for the ascent phase. McHenry et al. (1979) and Schleich (1982) give quite a detailed discussion on the Space-Shuttle ascent guidance, naviga- tion and control. McDermott et al. (1982) do a linearised stability analysis for this control sys- tem, and Schletz (1982) discusses the use of quaterions in the GNC system. The attitude con- troller for the entry and descent phase is only marginally mentioned in those publications. The Space Shuttle enters the atmosphere with a large angle of attack of about 40°, in order to minimise the heat load. Further down the trajectory, the angle of attack is reduced to meet with the cross- and downrange requirements. Range control throughout the entry is accomp- lished by control of the bank angle. The Flight Control System (FCS) must guarantee a safe and stable flight and thereby take into account wide variations in flight conditions and large model-data tolerances, next to the large attitude changes. To perform its tasks, the FCS can use a Reaction Control System (RCS) and aerodynamic control surfaces. Hamilton (1982) states five features of the Space Shuttle that present unique stability problems in combination with the large velocity range: i) the Shuttle is an unpowered vehicle, i) the control of the aileron, udder and the RCS jets is blended, il) the gains of the FCS are scheduled, iv) the rigid-body stability margins are small, with strong bending modes within an octave of the cross-over frequency, and v) the control system is multi-rate digital. Klinar et al. (1975) give a general overview of the Space Shuttle Flight Control System. However, this overview cannot give the final details of the FCS, since the first Space-Shuttle flight was in 1981 and the FCS design was continuously updated when new data became avail- able. Since the general idea did not change, we will use this paper (and others) to give the reader a global impression of the FCS design. The system design goal was to maximise the‘Organization: TUD/LAVAZR. Date: February 1997 _ Document code: LR-806 Page: 6° use of the aerodynamic control surfaces, of course within their power limitations, and to minimise the RCS propellant consumption while satisfying the handling qualities required for manual operation. The longitudinal and lateral controllers of the Space Shuttle are described as follows. The longitudinal automatic and manual FCS designs are conventional pitch-rate feedback control- lers, with outer loop closure accomplished by surface-position feedback or by the pilot. Actua- tors for both trim and control are the symmetric elevons, the body flap and for the low dynamic- pressure region the pitch jets. Also in case of the lateral automatic and manual FCS, the design exists of conventional rate-feedback controllers. Operations are depending on the angle of attack. For the higher angle-of-attack operation, the rudder is ineffective leaving only differential elevons (ailerons) and the RCS thrusters for control. The rudder is only activated below Mach numbers of 3.5. The yaw thrusters are used to control banking whereas the ailerons are used to damp sideslip. At lower angles of attack, there is a conventional aileron/rudder crossfeed for turn coordination. ‘A major concem during the design of the FCS was the flexible body interaction. As we can see in Fig. 1.2, which shows us the simplified entry FCS configuration in the all-aerodynamic phase, there are several bending filters included in the design. These filters were added because due to the high loads the Shuttle cannot be treated as a rigid body. The bending mode stabilisation was considered to be a problem that drove the design of the controller. What we can also see from this figure, is that the longitudinal and lateral motion are not completely decoupled, since the longitudinal controller has a feedback compensation of the yaw rate, which isa latera-motion component. Note that the commanded attitude that is computed by the guid- ance logic, is translated into commanded angular rates which are fed to the attitude controller (attitude-rate control instead of attitude control). a sy of Lefora (SOG comes jue "s Tas Lets tnue 2 come ron fas 7 cre conassio aa ae 1_ feset |B Acco Jaume | como See a come er tiem as a fone fm pale Momo RATE “ rem pj MAND aL fawn tore Fig. 1.2 - Entry FCS configuration for Mach = 3.5 (based on Epple and Altenbach, 1983).Organization: TUDILA/AZR Date: February 1997_ Document code: LA-806 Page: 7 Gruner w a = to THe ; | 4 ag z | le! fe] lelgl Me) level fete) |e E bl Is se (al EET ER Le LE Fig. 1.3 - The Space-Shutte lateral entry Fight Control System (based on Hamilton, 1982). eee(Organization: TUD/LRVAZR . Date: February 1997 _ Document coda: LR-808 Page: 8° Hamilton (1982) discusses the flexible body stabilisation for the aerosurface control loops. In Fig. 1.3, we have included the simplified block diagram that Hamilton used for the discussion ‘on the lateral FCS. We see that the basic idea is based on rate feedback, with several filters added to account for elasticity and to improve the response of the system. Furthermore, we find scheduled gains. The forward-loop gains are inversely proportional to the dynamic pressure, which raises the gains where aerosurface effectiveness is low. The aileron loop contains a yaw- rate feedback that is scheduled with the angle of attack. This gain is proportional to COT a. Analysis of the entry FCS has been described by a number of authors. Stone and Powell (1976) do an analysis of the entry guidance and control system to determine the sensitivity of the Shuttle to off-nominal stability and control aerodynamic parameters. Besides, they identified the boundary values for each of these parameters. The Space Shuttle entry flight control off- nominal design considerations are also the topic of a paper by Bayle (1984). He presents the sensitivity of the flight control stability margins to aerodynamics, discusses the flight control verification process and compares the predicted performance with the flight-test results of the first Space Shuttle fight, STS-1 Nguyen et al. (1990) describe the testing methodology that was used for verification of the Shuttle FCS, using simulation (software: linear stability analysis and off-line non-linear simulation programs; hardware: MIL engineering and verification simulators) and flight tests (the first four orbiter missions). Epple and Altenbach (1983) describe the dynamic stability testing of the Space Shuttle Columbia FCS and flexible-body interaction. Input stimuli were applied to the Shuttle vehicle controllers to excite bending while the FCS is powered-up, and the measur- ed responses from the operating FCS were compared with the predicted responses from the flex FCS flight mode! suitably modified to represent the ground-test configuration. Myers et al (1982), finally, assess the FCS and the flying qualities of the Space Shuttle during approach and landing. 1.3. Attitude-control concepts. A control-system design can be based on a number of underlying theories. The oldest and most, widely used concept is the one of feedback, of either the output or state of a process. A technique which is being used more and more in industrial control systems is adaptive control. Modern, robust control techniques that are subject of many studies are based on H., or ji- synthesis. Of course, there are many variations on the different concepts. Since we do not want to do an extensive survey of different control techniques, we shall limit ourselves to the three mentioned concepts that have known flight applications, and then only briefly. Our selection criteria of a particular technique for the control of an unpowered, winged re- entry vehicle are quite simple and straightforward. ‘+ We have a need for a control system in order to develop an analysis technique for testing Quided and controlled flight of atmospheric space vehicles. This means, that we should be able to control the vehicle but that it does not have to be the best possible controller which guarantees mission success under all circumstances. In fact, it is no problem if the | i 1 tl '‘Organization: TUD/LRVAZR 5 Date: February 1997 _ Document code: LA-806 Page: 9° sensitivity analysis will show the limitations of the controller. To put it in other words, not the performance of the controller is the study objective, but the analysis technique. + The design should be as simple and transparent as possible, so that we can gain in the related flight dynamics. + The controller has to be implemented in an existing flight-simul its implementation should not be the cause of great difficulties. + Preferably, but this does not have to be a necessity, we want to apply to HORUS a control technique that has not been applied before. * Last but not least, for an actual flight application the algorithm should be easy to be ‘embedded in an on-board computer. This is an aspect that we do not consider here, however, ight n software package, so Feedback control systems have found widespread use in, amongst others, aeronautical engineering (Bryson, 1985). Simple forms of feedback are (a combination of) Proportional, Integral and Derivative (PID) feedback of the output of the system or plant that has to be regulated. Classical control theory of linear systems was based on frequency response and root-locus techniques, see, for instance the books by Kuo (1987) and D’Souza (1988). A set of general performance requirements, that were not optimal in a mathematical sense but rather aimed at a reasonable performance, were commonly used. Initially, the older concepts were not easy to apply to multi-variable plants, State feedback systems, however, are particularly suitable for systems with Multiple Inputs and Multiple Out- puts, so-called MIMO systems. The parameters that define the control-system performance, the gains, can be obtained by pole placement or, alternatively, based on mathematically defined optimisation criteria. The Linear Quadratic cost criterion is well known in this respect, resulting in the so-called Linear Quadratic Regulator (LOR), see, for example, the books by Bryson and Ho (1975), Lewis (1986) and Gopal (1989). More recent trends in feedback design are given by Kokotovic (1984), who gives an overview dealing with non-linear feedback, .e., adaptive and ‘composite control, and with external linearisation. A recent application of the LQR using the method of extended linearisation is given by Wang and Sundarajan (1995), who describe a non- linear longitudinal flight controller for the F-8 aircraft, An advantage of LOR is that it is a systematic method for designing MIMO systems. Furthermore, the implementation of the control laws in fight-simulation software is fairly simple, and the computational load for on-line simulation is low. The problems dealing with pole assign- ment linked with MIMO systems have been replaced by an optimisation problem, and pole se- lection is now changed to the selection of the optimisation parameters (weighting matrices). However, when not all the states of the controlled system are available, then most of the attractive properties of the LAR methodology are lost. In that case an estimator is introduced to estimate the unavailable states, see Shahian and Hassul (1993), which then results in a Linear Quadratic Gaussian (LQG) controller. However, the LOR seems to be a very appealing concept for our purposes, i.e., designing- and implementation-wise, so we keep this method in mind for selection. ‘When a process is dynamic, i.e., the system parameters vary strongly with time, or in case there are many disturbances, a controller with constant parameters is not likely to perform well i tl i(Organization: TUDAL/AZR Date: February 1997_ Document code: LA-806 Page: 10 over the entire operational range. In that case we want in some way to adapt our controller parameters to the changing circumstances. Adaptive control, a special type of non-linear feed- back control, found its way into use in the earty fifties, as an autopilot for high performance aircraft (Astrom and Wittenmark, 1989), later on followed by applications in the F-94, F-101 and X-15 research aircraft (Boskovich and Kaufmann, 1966). Throughout the succeeding years, several different adaptive techniques were developed, of which three are more common: Gain Scheduling, Self-Tuning Regulation and Model Reference Adaptive Control. These forms of adaptive control are discussed by many authors. We have already mentioned the book by Astrém and Wittenmark, wt gives an excellent treatment with many examples and applications. We will add here the survey of adaptive feedback control, given by Astrém (1987). Itfocuses on the three mentioned concepts of adaptive control, but gives also a list with more than 350 references. In case of gain scheduling, auxiliary variables, that relate well to the characteristics of the process dynamics, are used to change the controller parameters. However, since there is no feedback from the performance of the closed-loop system, which compensates for an incorrect schedule, Astrém and Wittenmark (1989) do not regard this scheme as truly adaptive. With Self- Tuning Regulation, the system to be controlled is described by a model with (partially) unknown, parameters. During operation, these parameters are estimated with a recursive estimation method. The estimated parameters are treated as the best ‘guess’ of the system and used to caloulate new controller parameters. The problem of self adjusting the parameters of a controller in order to stabilise the dy- namic characteristics of a feedback control system when drift variations in the plant parameters occur, was the origin of Model Reference Adaptive Control or MRAC (Landau, 1974). With this, technique, a reference model serves as the basis to generate the steering commands for the (unknown) plant. The parameters of the controller are adjusted in such a way that the difference between the model output and the plant output are minimised. The performance of the controller is in this way less sensitive to environmental changes, modelling errors and nor-linearities within the system. A drawback might be, however, that a large control effort is required to make the plant follow the model (Messer ef al, 1994). Furthermore, the mathematical foundation of the original MRAC is quite large and may withstand a quick design and implementation. A survey of model reference adaptive techniques, both in theory and applications, is given by Landau (1974). This survey includes over 250 references. A recent work on direct adaptive control algorithms, and especially a simplified form of Mode! Reference Adaptive Control, is given by Kaufman et al, (1994). This latter methodology seems to be promising with respect to ease of use and computational requirements, so in principle we will keep this method in mind. However, due to time constraints and lack of practical applications we will not apply this concept here, but keep it as a focus point for further research ACRI/LAN (1992) discusses guidance and adaptive-control techniques of moderate lif-to- drag vehicles, and applies them to the problem of atmospheric transfer. The attitude of the Apollo-like vehicle can be regulated by a pulsating reaction-control system. Three possible attitude-control concepts are discussed, i.e., gain scheduling, STR and MRAG, in order of ‘mathematical complexity. Each of the controllers was based on a PID-type of control law and(Organization: TUDILR/AZR Date: February 1897 _ Document code: LR-808 Page: 11 gave satisfactory results, although the STR and MRAC increased the fuel consumption because the required extra signals generated oscillations in the angles of attack and sideslip. ‘The last of the three control concepts that we discuss here is robust control, or, to be more specific, application of H., i.e., the minimisation of the o--norm? of some transfer function, and 1. controller synthesis and analysis. These types of controller designs were developed for multi- variable feedback systems in the face of uncertainties (Doyle and Stein, 1981), since the per- formance of feedback systems is then ultimately limited. The name H, refers to the space of ‘stable and proper transfer functions. The objective in H_, control is to minimise the »=-norm of some transfer function, which will increase the robust stability margin of the system (Shahian and Hassul, 1993). As is the case with LG problems, also H., uses a state estimator and feeds back the estimated states. The controller and estimator gains are computed from two Riccati equations. Differences can be found in the coefficients of the Riccati equations, the weights. In principle, transfer-function weights are used to shape the various measures of performance in the frequency domain. In H-control problems, they are also used to satisfy the so-called rank conditions, that are frequently violated in case of inappropriate weights. Proper selection of the weights depends primarily on the experience of the user, and his understanding of the physics of the problem and other engineering constraints. Because of this, H., control is a complex method to apply. The capturing of both the performance of feedback and uncertainty aspects has been presented by Doyle et al. (1982). It involves a generalisation of the ordinary Singular Value Decomposition, and it provides a reliable, non-conservative measure to determine whether both the performance and robustness requirements of a feedback loop are satisfied. This measure is called the Structured Singular Value, denoted by the symbol p. Necessary and sufficient conditions to handle bounded structured uncertainty, that result from unmodelled system dy- namics, are given by Doyle (1985). Since the theory is far too complicated to describe in a nutshell, we suffice by giving the above references and the book by Doyle ef al. (1992), that gives an excellent introductory treatment of the robust performance problem. ‘Two applications of H.,-control and u-synthesis are of interest. Doyle et al. (1987) apply u- synthesis to the Space Shuttle lateral axis FCS during re-entry. Their conclusion was that the use of iis a very promising and powertul tool, if only for analysis. The produced results were very encouraging, and they yielded important information about the performance and robustness of the controllers. The second application concerns attitude control of hypersonic space planes. Since the mid eighties there was an ever-growing interest in guidance and control of space planes. Because of the large flight regime and the uncertainties in the dynamics, the application of robust control to this class of vehicles has been studied by several authors. As an example 2 The «-norm of a transfer function G(s) is defined by IGI. = sup |GUo)| where s is the Laplace variable and « the radial trequency (rad/s). Graphically, the e-norm is simply the peak in the Bode magnitude plot of the transfer function,‘Organization: TUD/LA/AZR. Date: February 1997 _ Document code: LA806 Page: 12 we mention the work of Gregory et al. (1994), They applied the concepts of H,, and u toa longitudinal model of a winged cone configuration at Mach 8. Their conclusion was that the ‘addition of using j1 provides robust performance, much more than H., in itself. Although they restricted to linear analysis, the results are very promising to stimulate further research. With respect to robust control, we found several references that give comparisons of dif- ferent control techniques. Grocott et al. (1994) make a comparison between § different robust- control techniques for uncertain structural systems, i.e., Sensitivity Weighted Linear Quadratic Gaussian, Maximum Entropy, Multiple Model, H,, and u-synthesis. The techniques were evaluat- ‘ed on computational requirements, the degree to which performance suffers from achieving robustness, and the maximum performance that can be experimentally achieved. For their low- ‘order benchmark problem, they found that because of the (very conservative) guarantees of robustness (H_), a large performance penalty can result. Compensators based on p-synthesis are much less conservative than H., designs, but are computationally infeasible for large-order plants, because the order of the control system increases drastically. Vincent et al. (1994) compared the Linear Quadratic Regulator with H., applied to a lateral- directional control-system design for a Mach = 0.9 flight condition. Both controller designs demonstrated excellent model-following performance, although each of the controllers had some individual strong points. The LR control law was simpler than the Hone, and could easily be implemented with in-line computer code. For the H,, design, some numerical points of concern were identified, i.e., a possible need for increased precision numerical representation (more memory required and a higher CPU load), and the fact that model-order reduction proved to be ‘a delicate numerical problem. In conclusion, we can state that H., and y-synthesis have promising features, although there are still practical problems with respect to performance and computational load. Further- more, the mathematical foundation of both methods is complex and quite some experience is required to develop controllers. For this reason, we will not pursue these methods. ‘Summarised, we have introduced three different control techniques, in order of increasing complexity: 1) Output or state feedback, with the controller parameters (gains) obtained by * pole placement, or * optimal control theory (LOR). 2) Adaptive control, with three of the more common techniques given by * Gain scheduling, + Self-Tuning Regulation, and + Model Reference Adaptive Control 3) Robust control, with two different schemes: + H.-control, and + u-synthesis. Of these three techniques, for reasons of simplicity while still having a reasonable performance, we will select state feedback where we will compute the gains by means of the optimal control‘Organization: TUDALFVAZR. Date: February 1997_ Document code: LR-806 Page: 137 theory. Furthermore, to cope with the large flight regime we will apply gain scheduling. In short, we will apply the Linear Quadratic Regulator with gain scheduling. Despite its old (but also proven) concept, and the introduction of many new control theories and techniques, the LOR is still widely in use, of which many examples can be found in literature (although they do not always focus on aircraft and space vehicles). Gawroriski (1994), for instance, proposes a linear quadratic design procedure for NASA's deep space network antennas, thereby dividing the antenna model into tracking and flexible subsystems and designing controllers for each of these parts separately. (Because of the separation the controller design showed a significant performance improvement.) Furthermore, Collins and Richter (1995) applied a Linear Quadratic Gaussian design as a possibility for the Hubble Space Telescope. The use of this control concept showed that it met all specifications, and that the precise attitude control required for Hubble was possible. Last but not least, we already mentioned the application to the F-8 aircraft. One of the disadvantages of linear state feedback control is its limited robustness in the presence of model uncertainties and non- linearities. Wang and Sundarajan (1995) used the extended linearisation approach together with the LR, thus removing some of the difficulties of gain scheduling, namely, that the scheduling variable should vary slowly and also that the scheduling variable should capture the plant non- linearity. Their (preliminary) conclusion was that the non-linear controller performed much better than the conventional gain-scheduled controller. The choice of this control scheme is based on the motivations given at the beginning of this section. Furthermore, to our knowledge this scheme has not been applied to a vehicle like HORUS, although the controller is based on similar contro! laws provided by MBB (1988b). However, they used pole placement to compute the gains instead of a quadratic cost criterion. Furthermore, the documentation of their controller was not complete (and not published in open literature), and there were no numerical values of the gains available. Last but not least, the design methodology for the LOR will be set up in a general way so that it can also be used for other vehicles and missions, and possibly also for a refinement of the controller design as to increase robustness and performance. In this respect we mention that the design methodology has been applied to a re-entry test vehicle with a triangular cross section, that is controlled only aerodynamically by three aerodynamic surfaces (Mooi et al, 1995). 1.4. Control-system design cycle. While studying the flight behaviour of conventional aircraft and designing autopilots for this class of vehicles, itis common practice, depending on the type of trajectory, to separate the longitu- dinal and lateral motion. This is usually allowed, because the two motions are decoupled, i.e., ‘a symmetric manoeuvre only has a marginal effect on the asymmetric motion and vice versa This decoupling implies a simplification of the autopilot design, and moreover, it gives an increased understanding of the natural aircraft motions. For similar reasons, it would be useful to do the same for hypersonic vehicles. However, for air-breathing space planes the coupling between engine and aerodynamic effects results in violation of the separability conditions, although aerodynamic effects basically satisfy these(Organization: TUDILR/AZR : Date: February 1997_ Document code: LR-806 Page: 1 conditions (Sinai, 1990)°. Since the control-system design in this report is focused on an unpowered re-entry vehicle, one might expect that separation of the longitudinal and lateral motion is possible. But, HORUS will fly with a large angle of attack and bank angle, indicating strong lateral motion, so the question: Is separation still possible? is worth asking. The design cycle can be divided into a number of successive steps, which will be briefly described below. In Fig. 1.4, we have schematically depicted the design process. Once more, it should be noted that we do not aim at designing the optimal control system for this vehicle and mission, Therefore, some iteration loops which are indicated in Fig. 1.4 will not be exe- cuted, although they should be considered when starting from scratch and aiming at the best Possible controller. In order to verify the controller and to set up a more general methodology of controller design, we take the following steps: 1) Wewill analyze the full six-degrees-of-treedom (6-d.0..) motion of the re-entry vehicle (both symmetric and asymmetric flight). The state of the vehicle is defined by three position variables (modulus of the position vector A, longitude t and latitude 8), three variables for the velocity (groundspeed ¥, flight-path angle y and heading x), three angular rates (roll rate p, the pitch rate q and the yaw rate 1) and three aerodynamic angles for the attitude (the angle of attack a, the angle of sideslip B and the bank angle a). The corresponding flight-dynamics model consists of 12 first-order differential equations. The related equations are given in Section 2.1. 2) The 12 coupled differential equations are non-linear and time varying, which makes it im- Possible to design the linear state-feedback control system with classical control theory. To apply this theory, the equations of motion have to be linearised and to be made time in- variant, To make this process as easy as possible, some assumptions will be made to sim- plify the starting equations. The linearisation is discussed in Section 2.2. 3) To study the open-loop behaviour of the vehicle flying its nominal trajectory, or in other words, the stability of steady flight, the linearised equations of motion have to be written in a special form, the so-called state-space form. This matrix equation will be presented in the final section (2.3) of Chapter 2. The nominal trajectory is divided into a number of discrete points, the so-called time points in which the vehicle is considered to be equilibrium. Per time point, a so-called Linear Time Invariant (LT!) system can be obtained, To address the time-varying character of the re-entry mission, each of the LTI systems will be combined in a series, that serves as the basis for the next step. 4) To study the open-loop behaviour of the vehicle it is sufficient to look at the eigenvalues of the system. More complete information about the characteristics modes is provided by 3 The performance of the propulsion system is depending on the angle of attack, in the form of pre-com- pression of the air with increasing angle of attack. This shows as an increment in the net installed thrust. In a similar manner, flying with a non-zero angle of sidestip will have its effect on the performance, but now in a negative sense. The in-coming airflow in the inlet ofthe propulsion system can decrease and shock waves can originate. Both phenomena decrease the available thrust. Since the thrust force is one of the major extemal forces in symmetric motion, the asymmetric angle of sideslip has established a coupling between the symmetric and asymmetric motion,(Organization: TUDILA/AZR . Date: February 1997_ Document code: LA-806 Page: 15 the corresponding eigenvectors. This so-called eigenmotion is studied in Chapter 3. 5) Chapter 4 describes the actual controller design. From the study of the eigenmotion around the nominal trajectory, it appeared that the symmetric and asymmetric motion are de- coupled. For this reason, the controller is divided into two parts, i.e., a pitch controller (Section 4.3) and a lateral controller (Section 4.4). For each controller, the corresponding reduced system of equations of motion is discussed, as well as the eigenvalues of this system. Furthermore, the selected control laws and the computation of the feedback gains ate presented. The underlying theory for gain computation, i., the application of optimal control theory, is given in Section 4.2. 6) The verification of the designed controller(s), consisting of three parts, is discussed in Chapter 5. For both the pitch and lateral controller, the response of the closed-loop system to both a step and a ramp input will be considered. Also, the flight along the nominal trajec- tory with the integrated controller is presented, Nota bene: at this place (before testing the controller in the non-linear flight environment), it would be possible to do a sensitivity analysis of the linear model. When the conclusion is that the linear model is not robust enough, a redesign has to be done. This sensitivity analysis is skipped in this report, because we are not doing an optimal design of an attitude controller. Step 6 marks the end of the design process, as discussed in this report. Since we know from MBB (1988b), that a similarly developed controller gives a fair performance, not further improving the performance of the attitude controller seems a reasonable thing to do. However, the design of a controller is usually not that straightforward. After the feedback-gain computation and some response tests, one might have to go back one or two steps in the design process, as can be seen in Fig. 1.4. It is possible that the time points, which have been selected, are not sufficient to cover the whole trajectory. In that case, more time points need to be selected. ‘On the other hand, it is quite well possible that for none of the time points the required behaviour can be achieved. Then, one has to reconsider the choice of the contro! laws. ‘When the two separately developed controllers prove to have done what they are supposed to do, they have to be integrated and cooperate so that the nominal mission for which the controllers were designed, can be simulated. At this stage, we can check whether the sim- plifying assumptions, which we made during the design process, were justified. Only after suc cessfully completing this test, we can proceed with the next step: a sensitivity analysis, which should give us insight in the behaviour of the vehicle and the controller under other than the nominal conditions. Again, it is possible that we have to redesign the controller. The outcome of the analysis might even imply that a linear state-feedback controller with gain scheduling is not suitable at all. Whether that is the case, remains an open question within the framework of this research.ween enna = == sci ses Date: February 1997_ Page: 16 ‘oma tT ? Organization: TUD/LR/AZR Document code: LA-805 a { j we,| [ome oe ame. | [ame — woz | | ome i coz | | oe a f t : oS - Fig. 1.4 - The design process ofthe linear state-feedback controllerChapter 2 The Motion of a Vehicle ina Planetary Atmosphere To study the motion of a vehicle in a planetary atmosphere, it is necessary to derive a math- ematical model of this motion. Starting with the Laws of Newton, the translational and rotational mation of a vehicle can be described by a system of 12 coupled, first-order differential equa- tions. These (non-linear) differential equations can be numerically integrated to gives us the variation of position, velocity, attitude and angular rate with time. The general form of these equations is introduced in Section 2.1. However, to design a state-feedback control system while applying classical control theory, it is necessary that the system of differential equations is linear in all its state variables. This process of linearisation is discussed in Section 2.2. The ‘matrix form of this linearised system, also called the state-space form, is described in Section 2.3. This form enables us in the first place to study the characteristic motion (or open-loop behaviour), and in the second place to design our control system. These topics will be described in Chapters 3 and 4, respectively 2.1. The general form of the equations of motion. The control-system design is centred around an unpowered, winged re-entry vehicle. Detailed discussions on the equations of motion of such a vehicle can be found in Mooij (1994a). We will summarise these equations in this section. First, however, we will state the underlying assumptions. + The Earth is represented as a sphere and is rotating with a constant rotational rate «4, + The vehicle has a plane of mass symmetry (X,Y,plane), which means that taupe In Appendix A, the definition of the state variables can be found, The dynamic equations of translational motion: eset reeeenes‘Organization: TUDALR/AZR Date: February 1997, Document code: LR-806 Page: 18° F v > + wf, Acos8(sinycosé ~cosysiné cosy) (2.1.14) F, 2 We 1 + 20,,Veosbsiny + “cosy + DB a (2.1.2a) + wep Ficos8|cosScosy +sinysinécosy) o. #, ; Veosyi = —£ + 2a V{sindcosy-cosdsinycosy) + ™ (2.1.3a) V? nos? 2 + Teostytandsing + wf Acos8sinésiny with Fy = -D ~ mgsiny - mggcosycosy (2.1.1b) -Ssino + Loose - mgcosy + mgysinycosy (2.1.20) =, = ~Scosa - Lsina + mggsing (2.1.36) The corresponding kinematic equations: A= Vsiny (2.1.4) Vsiny.cosy Ysingcosy 2.1.5) Fcosb a Veosxcosy § = Weosxeosy (2.1.6) F (2.1.6) The dynamic equations of rotational motion: (2.1.7) (2.1.8) saps etteabemn‘Organization: TUD/LRVAZR . Date: February 1997_ Document code: LA-806 Page: 197 (2.1.9) with r The corresponding kinematic equations: cos = -pcosasin8 +qcosf-rsinasing + + sing|zcosy-dsinysiny +(t+a,,)(cosBcosysiny -sindcosy)| + (2.1.10) ~ cosofy -dcosy -(i+0,,) cosdsiny] B = psina-reosce + + sino|y-bcosz ~(¢ gq) cosdsiny| + (21.11) + cosozcosy -dsingsiny +(¢+10,4)(cosécosysiny -sindcos?)] & = -pcosacosp -gsinB -rsinacosB + (2.1.12) + asinB -zsiny ~ésinzcosy + (¢+w,,)(cosBcosycosy +sinésiny) In these equations, j, i, 8 and + are given by Eqs. (2.1.2) through (2.1.6). Note that the external moment components are expressed in the body-fixed reference frame. 2.2. Linearisation of the equations of motion. ‘The characteristic motion, or the open-loop behaviour, of a vehicle, is can be described by the eigenvalues and eigenvectors of the equations of motion. However, the eigenvalues and eigen- vectors can only be obtained when the time derivatives of the states are given as a linear com- bination of the states. In other words: the equations of motion have to be linearised. Asa result, the motion given by this linearised system should be regarded as an equilibrium trajectory, so that the characteristic motions are small deviations from this nominal path. To simplify the pro- cess of linearisation, we will make some assumptions: ‘+ We will consider a non-rotating Earth («o,, = 0 rad/s), which is allowed since the rotation of the vehicle is of a much higher frequency than the rotation of the Earth. As a result, the Coriolis and centripetal accelerations are zero as well,(Organization: TUD/LFVAZR, . Date: February 1997_ Document code: LA-805 Page: 20, + The gravity field of the Earth is assumed to be spherical (g5 = 0 m/s). Nota bene: since there is only one component of the gravitational acceleration left, we will omit the subscript '? + The vehicle is assumed to be rotationally symmetric (w.r.t. mass) around the X-axis of the body-fixed reference frame, which means that I,, = 0. With the above assumptions, we can neglect the smaller terms. The resulting equations of motion become: (22.1) (Leoso-Ssin) f coos seine) 2.2 est — (22.2) v, (Lsing +Scoso) a ising, - {Using + Seos) 3) i = Goosrtanssiny maVeasy (2.2.3) R= Vsiny (22.4) Veosysiny, oe ** —Feoss ee) v bes (2.2.6) Foosvoosy, (2.2.6) My B= + Uyntenlar (2.2.7) M, = + (Iee-by)pr (2.2.8) ly M, Et larhy)Pa (2.2.9) Since the kinematic attitude equations are very complex, we will introduce another simplifi- cation. We assume that the vehicle's attitude is only marginally influenced by an asymmetric translational motion, or in other words: we assume that the vehicle's trajectory is parallel to the equator (5 = 0° and x = 90°), and remains that way (5 = rad/s), see also MBB (1988b). L-mgcosycoso mVcosB q ~ (pcosce+ rsina)tanB (2.2.10)Organization: TUDILA/AZR Date: February 1997_ Document code: LA-806 Page: 217 S+mgcosysine = psina - rosa - 2.2.11 B =p: mV ¢ ) poosarsina _ yang L~MGCOSYCOSS , tary Lsino + Scosa (22.12) cos mV “mv ‘As we can see, the equations for x, t and 5 are not coupled to the rest of the equations, so a 9-d.of. linearised model for the state space formed by V, 7, A, p, q, F, a, B and. can be derived. The linearisation is performed as follows. We assume an equilibrium value of each of the state variables (index 0), and we will look at small deviations from this equilibrium state. Note that the vehicle is not powered (no fuel consumption), so the mass properties are constant. So: 2 (22.13) q r a B o The nominal position, velocity and corresponding control history follow from the nominal trajectory, which leaves us with three unknowns Po, do and ry However, since we want no perturbation of the nominal control (a1, and ¢) when we are flying the vehicle, we can compute the equilibrium angular rates from the condition a=6 or, using Eqs. (2.2.10-12), @ = 9. - © cosyycosey (2.2.14) Yo Vo _% Posinay ~ recosdg = Pcosyysinog (22.15) oa pocostty * Mo8inag = —?_tanypsinay (2.2.16) ™V5 Solving for Pp. dg and Fp gives usOrganization: TUDALR/AZR, : Date: February 1997_ Document code: LR806 Page: 22 Po = &Sindg + cpc0s0 (2.2.17) by _ % Taig” Teeomaee8e (2.2.18) Ig = ~C,00809 + cpSinog (2.2.19) with & = Peosygsinay Yo & aagiarosey Note that since By = 0, also Sp ‘Substituting expressions Eq. (2.2.13) in Eqs. (2.2.1-12), neglecting higher-order terms such as AVAy, pga, etc., and subtracting the nominal state (i.e., Vo following 9 first-order differential equations: sy ete.) results in the AD, 2Rsinpan = gocosypAy (2.2.20) ™ 2 2m Yi ve py (2m nag | ey 0 |" Ay} % ° ° (2.2.21) cose; at - Bas ™Ve 7™V5 (2.2.22) (2.2.23) 4 The nominal rotational rates Py qy and rp are small, and can therefore be treated as perturbations themselves.Organization: TUD/LAVAZR Date: February 1997, Document code: LR-806 Page: 25" (2.2.24) (2.2.25) ad = Ag - — at - ®cosigsinayas «| -- cosrgcosay [av + me Vo mye ve o fo (2.2.26) 90. 295 -& Ay - an Prsimieoseg dr ~ =y-COsIpcOsN FPeosvecoscss + Beosipsinopav + ° (2.2.27) 2 . a cosypsinggAR + sPoimgsinesay 6 = -cosagip - sinogar -|!? -% eosycosoy ap + 1 sinayAy + TV Vo TmVo (2.2.28) by Gpsineoav In deriving the above equations, we have used the defini of the gravitational acceleration 1 aad which gives us oe ee (FotARP OR? (14: _ Po 9 * Ag or ag = 2a‘Organization: TUD/LR/AZR Date: February 1997_ Document cade: LR-808 Page: 24 Now that we have linearised the equations of motion, we can write them in matrix form, also called the state-space form. This will be described in the following section. 2.3. The state-space form of the equations of motion. ‘When equations of motion are written in state-space form, they have the following form in matrix notation: Ax + Bu (2.3.1) with x an nx 1 state vector, ua qx 1 control vector, and A and B the n x nstate (or system) and n x qcontrol coefficient matrices, respectively. Eq. (2.3.1) is called the dynamics equation; to complete the description of the state of the vehicle we also need a so-called output equation: y=Cx+ Du (2.3.1b) In the above equation, y is the m x 7 output vector, and C and Dare the m x n output and ‘m xq direct transmission matrices, respectively. For the time being, we will restrict ourselves. to the use of Eq. (2.3.14). To write Eqs. (2.2.20-28) in state-space form, we must distinguish between state variables and control variables. The choice of state variables is obvious, if we look at the original ‘equations of motion with their state variables. We write x = (AV,Ay,4R,4p.4g,4r,4a,48,46)7 The selection of control variables is less clear. The principle of control is obvious: by changing the magnitude and direction of extemal forces (and therefore in principle also the external moments), the trajectory of the vehicle can be changed. As we mentioned in Chapter 1, the guidance system makes sure that the vehicle will follow its nominal trajectory by adjusting the angle of attack and bank angle, the control variables of the guidance system. These control variables determine the size and direction of the aerodynamic force vector, the only controllable extemal force acting on the unpowered vehicle. (The other external force is of gravitational ‘origin; this force is depending on the position of the vehicle and cannot be controlled actively.) The attitude controller has to guarantee that the commanded attitude is obtained (and main- tained) with a certain accuracy in a finite time, which means that eventually there should be moment equilibrium. Note that the actual angle of attack and bank angle are given by the kinematic equations, which means that they have defined values depending on the rotation of the vehicle and thus the size and direction of the external moments. So whereas the forces acting on the centre of mass (c.o.m.) of the vehicle are the guidance control variables, the moments around the c.0.m. are the attitude control variables. Inspecting the equations of motion, we find beside the three force components AD, AS and‘Organization: TUD/LRVAZR Date: February 1997_ Document code: LR-806 Page: 257 AL, three moment components, i.e., AM,, AM, and AM, Part of the moment components is determined by the vehicle, depending on its actual attitude. However, the remaining part is en (and can be changed) by, for instance, aerodynamic control surfaces, reaction contro! thrusters and momentum wheels, depending on the vehicle configuration and sub-systems. So for the selection of the attitude-control effectors, we must have a closer look at the vehicle, the HORUS-2B. MBB (1988a) presents this vehicle as winged, without a major propulsion system, but with 5 aerodynamic control surfaces (two rudders, two wing flaps® and a body flap) and a number of reaction control thrusters. The body flap is only used for trim, and does not have to be considered as a control. The two wing flaps, or elevons, can be deflected symmetrically {elevator function) and asymmetrically (aileron function). The rudders are outward movable only, and only one at a time. It should be noted that by deflecting the control surfaces, an aerodynamic force is generated that gives a moment around the vehicle’s c.o.m. depending on the moment arm, the distance to the c.o.m.. The deflection of a surface is in principle an analogous process, of course with a limited accuracy, which means that the generated moments are a continuous function of the deflection angle, as well as of flight and similarity parameters, such as the Mach number. The reaction control system is only being used when the aerodynamic control surfaces are not sufficiently effective, e.g., in the upper layers of the atmosphere. The aerodynamic control surfaces are activated when their effectiveness is more than 10% of that of the corresponding thrusters and the thrusters are inhibited when their effectiveness is less than 10% of the corresponding aerodynamic surface®, Whereas the aerodynamic control moments are con- tinuous functions, this is not the case with the reaction control moments. The principle of a thruster is that it is either ON or OFF. When it is on, it will generate a constant thrust force that will also result in a constant moment’. By switching the thrusters on and off repeatedly (so- called pulsing), the required moment can be approximated. Furthermore, in case there are more thrusters for generating moments about one particular axis, the magnitude of the moment can be varied by smartly combining the required thrusters. In this study, however, we will assume continuous reaction control moments, for reasons of simplicity. Besides, we will directly use the moments as control variables, instead of the thrust forces. ‘So, summarised we can write for the control vector u: 5 To be more inline with other Iterature, we will not use the term wing flaps, but elevons since this control ‘surface combines the elevator and aileron function ®© For the defined configuration of HORUS-28, the maximal torque of the pitch thrusters is 10,400 Nm, the fone of the rol thrusters is 1,600 Nm and the one of the yaw thrusters is 7,600 Nm. This results in activating the ‘aerodynamic control surfaces at approximately 100 km and switching off the roll thrusters at about 75 km, while ‘the yaw thrusters remain activated until the end of the descent. 7 The thrust force is in principle depending on the atmospheric conditions, which change significantly during the fght. Since we do not have extended thruster models at our disposal, we will assume constant thrust forces. However, it is known from the experience with the Space Shuttle that the operation of the reaction control thrusters can be significantly influenced by atmospheric conditions. Future models should take this effect into. ‘account so thatthe influence on attitude control can be studied. temrteron er vere