NESC ACADEMY WEBCAST

WELCOME TO THE NESC

GN&C TDT WEBCAST SERIES
Today’s Moderator: Neil Dennehy (NESC)
Today’s Speaker: John Burken (DFRC)

Fundamentals of
Aircraft Flight Control

Learning from the Past, Looking to the Future Page: 1

NESC ACADEMY WEBCAST

Learning from the Past, Looking to the Future

NASA Engineering and Safety Center (NESC) Webcast

Fundamentals of Aircraft Flight Control

NESC GN&C Webcast

October 17, 2012

John Burken NASA DFRC

NESC Technical Discipline Team Member

3
 Outline
Objective of Talk
 Why use control
− Some Under Graduate Background Information
− What’s not covered due to time constraints
 Models
− X-29 example
 Design Method Used
− Background on Robust Servomechanism LQR (LQI)
 Robustness Issues
− Add actuators with surface and rate limits
− Sensors (real states or estimated states?)
 Results / Time Histories
 Conclusions
− Remarks
 Appendix
− Call me for models and matlab files

4
 Flight Control Design – a snap shot

Objective of Talk:
 This 50 minute “short course” will hopefully hit the major control design issues and what to
beware of in the “design process”.

 In a linear design world the actuator rates and limits are ignored, along with time delays, and this
can bring down an otherwise “good design method”.

 This class will hopefully show you ways to design a robust control design method with some of the
“real world” issues involved.

 The control method will be the Linear Quadratic Tracker design technique, applied to the X-29
vehicle with matlab scripts included upon request.

 Scripts available for self-study purposes.

− Call or email [email protected] 661-276-3726
 Why the X-29
Models are open literature (no ITAR or SBU issues)
Therefore we can show them, design with them & give them out.

5
 Flight Control Design – Not covered

 Not Covered in this Talk but are important: (maybe covered in later talks?)

− Software and hardware System redundancy {triplex etc…}

− Verification and validation testing {did you design it the way you wanted & does it make sense}

− System Failure Accommodation {one engine out etc…}

− Pilot / Aircraft Handling Qualities (one bad pilot rating can cause a full redesign cycle)

− AeroServoElasticity (ASE)  When flexible structures meets dynamic pressure

− Integrator Windup Protection

− Digital system design (some talk on anti-aliasing)

6
 Basic review of concepts (Under Graduate stuff)

 1st question: Why use a controller?

− Your open loop system does not meet requirements {damping, rise time etc..} (stick to response sucks!)
− Objective of controller: Help damp, make output track your commands, reject disturbances and be
robust.
 How? We need to get a math relationship to model the physics.
− Ordinary Differential Equations ODE
− How do we find a solution? (Find an explicit expression for y(t)!)
− (2nd order ODE time-domain)

 State Space Representation of Dynamic Systems

We define the “so called states variables”: ( states order of or are not unique)
•
•
 Now rewrite the above equations in matrix form (state space format)

• y

7
 Basic review of concepts (continued)

 Now the above equations are in matrix form (state space format)

 This the “state space representation” of the ODE of the dynamic system.

 One very important transform is the Laplace Transform

− The Laplace transform is a method to convert a time-domain function into a frequency-domain

 We will assume you have had root locus design

− Old Time control religion
− The “old timers” told me that a nice rule of thumb for PI controllers:
• Kp = proportional gain , Ki = integral gain
• The Integral gain (Ki) should be {1/10  1/20} of Kp.

8
 Flight Controllers

 Typical Controller types:

Actuator model
− Stability Augmentation Systems (SAS)
• Purpose: Increase damping / very simple
pilot stick
• Pitch, Roll & Yaw Dampers 3
1
• Limited control authority In1 1
0.05
1/s 2 1
act_pos
Rate Position
• Limited number of sensors required Saturation Saturation

.5s
.5 2
.5s+1
Sat Lim
Switch
0
q (d/s) 4
In2
SAS off

− Command Augmentation Systems (CAS)

• Purpose: Track pilots commanded inputs
• _cmd , _cmd or p_cmd etc…
A/C
• Autopilots
response
• Help reduce pilot work load Pilot cmd
• Altitude hold, Mach hold ; etc…
K(s) G(s)

sensors

9
 Feedback in a Nutshell

e = r-y
r u y
k G

 Objective is to make the error e small

− But don’t know G exactly.
− Use feedback and make k large to make e small
− Larger k can give better performance
− Large k is good for disturbance rejection
− Large k can give rate limiting etc…
− Large k is bad for noise rejection

10
 Assuming we have stability!

 Too little feedback won’t stabilize an unstable G

 Too much feedback can destabilize a stable G
 Controller tuning by trial and error can be dangerous: cliffs are
conceivably everywhere? Do you want to chance this??

11 11
 Continuous (s-plane) to discrete (z-plane)

 A word of caution if you have a digital system (and who doesn’t these days?)
− Aliasing
• High frequency signals are “aliased” into lower frequencies.
• Just see a movie with a stagecoach wheel as it starts to turn faster!!!

− Sampling theorem : Critical frequency

• /T  Nyquist frequency

− Don’t let the high frequencies look like low frequencies .

• The fix: Use a Continuous low pass filter, F(s) on all signals used in the digital
controller
• Look up anti-aliasing filters
• Anti-aliasing filter: “Don’t leave home without one”

12
 All models are wrong, but some are very useful

 The model must suit the task

− Balance simplicity and accuracy
− Accommodate uncertainty

 Most models are based on a combination of physics and data

− Physics: Newton, Maxwell, Hooke, Joule
− Data: Measurements and the estimates we derive from them
• Parameters, frequency response, empirical models

 Moral of the story:

Don’t assume your model is right, you need to build in robustness

13 13
 1st Fundamental Question of Control Tuning

What limits feedback performance?

Performance is the ability to

stabilize, follow commands, and reject disturbances

14 14
 What Makes Feedback Control Challenging?

− Unstable and lightly damped dynamics (requires stabilization)

− Noise (sensor, actuator, disturbance)
− Non-minimum-phase zeros (limits achievable performance)
− Time delays
− Loop coupling (MIMO)
• (ex yaw into roll, use a little rudder and see what happens open loop)
− Nonlinearities
• Nonlinear dynamics (geometric, kinematic, chaotic,…)
• Hardware constraints (magnitude/rate saturation, deadzone)
− Constraints on states (hard bounds)
− Model uncertainty
• Exacerbates all of the above

15 15
 2nd Fundamental Question of Feedback Control

What do you need to model and how

accurately do you need to model it?

• Remember all models are wrong.

• You need robustness to uncertainties

• For SISO systems use phase “and/or” gain margins
• For MIMO use  analysis

16 16
 Gain and Phase Margins

Im Im
-GM +GM
-1 +GM Re
Re
-1

L  j1  L  j 2  L  j 3 
+PM
Close to (-1,j0). Small
simultaneous gain and
Gain Margins: GM  L  j1  phase uncertainty will
destabilize closed loop
+PM

GM  L  j3 
system.

L  j 
Phase Margins:   L  j2   
Include These Models In Frequency Domain Analysis
Command Plant Input Actuator Rigid Body
Controller Zero Order +
Inputs Time Delay Hold Dynamics Dynamics

lin ... timedelay zoh actuate lon lat/dir +

ASE
Linear Analysis Model Dynamics

TBD

ASE Plant Output Sensor

Compensation Time Delay Dynamics C FB

asecomp timedelay sensor

17
 Important Transfer Function Loops

Returned Injected
SISO Signal Signal
r + e uo ui y
K(s) G(s)
-
Controller Plant

Inject signal ui , examine returned signal uo

U o  s    K  s  G  s U i  s  K(s),G(s) Transfer Functions
   
L s
U i  s   U o  s   U i  s   K  s  G  s U i  s   1  K  s  G  s U i  s 
1 L s - The Return Difference Dynamics
E s 1 Y s L s  Zeros of the return difference are
  S s   T s the poles of the closed loop
R s 1 L s  R s 1 L s  system

Sensitivity Complementary Sensitivity 18

 Recap

 You need a controller (open loop response is not good)?

− Will SAS work or do you need full CAS
− Will a linear controller work?
 Is the controller Single Input Single Output (SISO)?
− Classical gain or phase margins (anywhere in the loop: same margins)
 Is the controller Multiple Input Multiple Output (MIMO)?
− Robustness
− Use mu (but we still use SISO margins on top of mu – Get pass the FRR).
• Why SISO Gm or Pm: Some axes are very decoupled and SISO is almost OK!
• Why MIMO Mu: It can be shown (Doyle) that infinite margins in SISO format is no
guarantee
 Don’t trust your models
− All models are wrong, but by how much?
 I assumed you have had a root locus design course.
− We’ll start off with the cornerstone to modern control design (LQR)

19
 Design Example

 Design an LQR controller

−Why LQR?
• Guaranteed Stability Margins for LQRs
• Excellent Stability Guarantees { -6, ∞} Gain Margins and {±60 degree} Phase Marin
• Example of LQRs Boeing 777 lateral/directional axes
• Small problem with LQRs
• Regulators like zero commands
• Airplanes don’t fly at zero AOA

−You may need the LQR controller to track a constant command

• Robust servomechanism  RSLQR  LQR Tracker  LQR-PI  LQI
• To track a constant command you need a type 1 controller
• Append the integrator states to Open Loop dynamics
• Examples of LQR trackers
• X-31 (The Germans non-zero set point LQR)
• JDAM (Boeing)

20
 Servomechanism Design Methodology

 Optimize the following cost function.

Optimal linear-quadratic-regulator (LQR) problem.
T
J   ( x ' Qx  u ' Ru )dt
0

 The algebraic Riccati equation

0  A ' P  PA  Q  PBR 1 B ' P

 And the optimal control is given by:

u (t )   R 1 B ' Px(t )  Kx (t )

21
 Servomechanism Design Methodology (cont.)

Consider a MIMO system


X  Ax  Bu  Ew where x  R n , u  R m , y  R p
Y  Cx  Du  Fw
w  the disturbance
The dynamic controller is

x c  Ac x c  Bc (r  y) Note :
 LQR Servo = LQR PI = LQI (matlab)
The open loop augmented system is  Approach is simple to implement.
 x   A 0  x   B 
     x     B DU

 x c   cB C A c  c   c 
Suppose the following condition is satisfied
 λ I  A B
rank  i  np
  C D If this statement is true: there
The system is controllable and there exist a exist a closed-loop system
control law that is stable.
u  kx  k c x c
22
 Servomechanism Design Methodology (cont.)

 Remarks (the good, the bad & the ugly):

 Optimal control with state feedback is the best you can do. “K.A. Wise”
− Some dynamic controllers can out perform RSLQRs.

• Beware: if you need to estimate states there are no guarantees. “J. Doyle”
• Is β estimated or a true calibrated β vane??

 Don’t push the system to hard: are the actuators in your design/simulation,
time delays, etc.???

 With LQR-PI be careful when tracking a rate;

• Check the rank of your augmented system with the integral states

23
 X-29 Lat/Dir Linear Model

Design the controller

to track p & b
Flt condition: Mach .7 / H = 20k ft commands
24
 X-29 Lateral Directional State Space Models

[ ]
− 0.0006 0.0007
𝐵𝑙𝑎𝑡 = 1.3470 0.2365
0.0919 − 0.0706
00 Flt condition: Mach .7 / H = 20k ft

[ ]
1000
𝐶 𝑙𝑎𝑡 = 0 1 0 0 Outputs = x’s = [β p r phi] radians
0010
0 0 01 Inputs = u’s = [Ailerons Rudder] degrees

[ ]
00
𝐷 𝑙𝑎𝑡 = 0 0
00
00

25
 Block Diagram

X-29 La t/Dir Linea r Quadra tic Tra cker

= LQI
for Roll Rate(p) & Bta
11/04/2010

States

bta
X-29 lat/dir
command Dynamics
x
p
bta
b_cmd bta_cmd act_in act_out Demux
LQR_out Control State
.
p r
actuators
p_cmd p_cmd A/C Dynamics

Command
LQR-PI controller phi

25
u2
ts

26
 3 test cases for Mach .7 // H= 20k ft

 Test case 1:
− Design without actuators (not thinking to much about reality)
− Does it meet requirements

 Test case 2:
− Test the first case with real actuators

 Test case 3:
− Redesign with actuators in mind
− Reduce gains to reduce command demand

27
 X-29 Lat/Dir LQI Design

 Test Case 1: First design without actuators:

− Why : If it does meet your requirements here, how can it with real actuators??
X-29 LQI Control without actuators in the Design
p dps 50

-50
0 5 10 15 20 25
10
 deg

-10
0 5 10 15 20 25
2
Ail - deg

-2
0 5 10 15 20 25
10
Rud - deg

-10
0 5 10 15 20 25
Time - sec

Controller gains
Klqi = [-22.0455 5.0649 8.6107 35.5815 -29.5533
96.9614 7.5731 -47.7479 -295.5326 -35.5815] 28
 X-29 Lat/Dir LQI Design

 Test Case 2: Test with actuators {same gains} Has states and real limits {rate limits / position limits}

X-29 LQI Control without actuators in the Design

50
p dps
0

-50
0 5 10 15 20 25
10
 deg

0 Performance??
-10
0 5 10 15 20 25
2
Ail - deg

-2
0 5 10 15 20 25
20
Rud - deg

Rate limiting??
0

-20
0 5 10 15 20 25
Time - sec
Controller gains
Klqi = [-22.0455 5.0649 8.6107 35.5815 -29.5533
96.9614 7.5731 -47.7479 -295.5326 -35.5815] Gains are too BIG!!
29
 X-29 Lat/Dir LQI Design

 Test Case 3: Change the Q & R weightings to penalize the surface activity (Reduce demand on
commands)
X-29 LQI Control without actuators in the Design
50

p dps
0

-50
0 5 10 15 20 25
10
 deg

-10
0 5 10 15 20 25
2
Ail - deg

-2
0 5 10 15 20 25
10
Rud - deg

-10
0 5 10 15 20 25
Time - sec

Controller gains
Klqi = [ -12.4120 5.5202 3.7623 7.4219 -30.7395 Gains are lower!!
3.2892 2.3039 -11.2919 -43.4722 -10.4962] 30
 Final Remarks

− Always design with a reality check in final test (put in real actuators etc..).

− In your simulation: add realistic time delays and non-linear elements.

• Go from simple to more complex (3 dof  6 dof)
• Why not just start in nonlinear high fidelity 6 dof simulation?
• You lose insight to simple dynamics.

− Have the pilot involved early in the 6-dof simulation.

31
 Appreciations / Thanks

 Dennis S. Bernstein
− Department of Aerospace Engineering, University of Michigan

 Kevin A. Wise
− Boeing

32
 Appendix

 If you want the matlab scripts or have questions please call or email me.
 If you improve on the example please share with us (NESC).
− John Burken
• phone 661-276-3726
• [email protected]

33
 NESC ACADEMY WEBCAST
THANK YOU FOR
ATTENDING TODAY’S WEBCAST

ANY QUESTIONS?

Learning from the Past, Looking to the Future Page: 34

 NESC ACADEMY WEBCAST

Upcoming NESC GN&C TDT

Webcasts (as of October 2012)

“Fundamentals of Adaptive Control”, Irene Gregory (LaRC), 28 th

November 2012, 2pm Eastern Time (4th Wednesday of Nov only)

“Fundamentals of Spacecraft Attitude Determination”, John

Crassidis (Univ. of Buffalo), on 16 January 2013

“Fundamentals of Kalman Filtering and Estimation”, Chris

Dsouza (JSC), on 20 March 2013

Learning from the Past, Looking to the Future 20 June 2012 Update