Friction Models and Friction Compensation Introduction

Karl J. Åström • Essential in Motion Control

Department of Automatic Control
• Classics
Lund University
Leonardo da Vinci (1452-1519
1. Introduction Amontons 1699
2. Friction Models Coulomb 1785
3. The LuGre Model • Tribology
4. Effects of Friction on Control Systems • Control
5. Friction Compensation • Physics AFM
6. Summary • Geology

Amontons’s Paradox Application Areas

Observations
• Robotics • Antennas
• Friction is proportional to normal load
• Machine tools • Telescopes
• Friction does not depend on the apparent area of contact
• Valves and actuators • Mechatronics
• Automobiles • Micro-mechanical sys-
Tire-road interaction tems
ABS • Geophysics
Traction control • Surface physics
• Excavators • Physiology

The classical friction law F = µ N

c K. J. Åström TDU, October 2005

F 1
 Very Complex Phenomena Some Friction Phenomena
Reasonably well understood phenomenologically Pre-sliding displacement Dwell time dependence
• Stiction, elastic deformation
Some phenomena
• Steady-state characteristics
• Pre-sliding displacement
• Hysteresis Varying break away force Hysteresis
• Varying break-away force
• Randomness—Repeatability
Much poorly explained
• Friction, surface roughness, and lubrication
• The Mica experiments

Simple Mechanisms The Bristle Model

Metal contact between asperities An abstraction of asperities

Visualization of break away

c K. J. Åström TDU, October 2005

F 2
 A Control Perspective Friction Models and Friction Compensation
• Understand the effects on friction on control systems
1. Introduction
• Typical phenomena
2. Friction Models
– Limit cycles
Static models
– Stick slip motion
Rate and state models
– Hunting
Dahl’s model
– Lack of precision in tracking
The Bliman-Sorine model
• Modeling and Simulation
3. The LuGre Model
• Friction compensation
4. Effects of Friction on Control Systems
• Modeling for control
5. Friction Compensation
6. Summary

Friction Models Static Models

• Classical static models
– Coulomb friction
– Viscous friction
– Stiction
• Mechanics and fluid dynamics
– First principles
– Microscopical contact
– Viscosity
• Empirical phenomenological models
– The Dahl model
– The Bliman-Sorine model a) Coulomb b) Coulomb + viscous c) stiction d) Stribeck effect
– LuGre model
In practice there are often asymmetries!
c K. J. Åström TDU, October 2005
F 3
 Rate and State Models Dahl’s Model
Friction has been studied extensively in the Physics and • P. Dahl Aerospace Corporation 1968
Earth Qake communities. The models are called rat and state • Extensive use in simulations in military projects
models because friction is a function of velocity v and another
variable which is dynamically related to velocity. • Inspired by solid friction

The Ruina-Rice model is a representative models it has the F

form Stress-strain curve c
F
 v  θ/ dF  F α
µ = µ 0 + A log 1 + + B log 1 + =σ 1− sgn v
v0 θ0 dx Fc x˙ > 0
x˙ < 0
dθ θv
=1−
dt d0 • σ stiffness slope σ 0
x

In steady state we have θ = d0 /v and • α shape

- Fc
 v  d0 
µ = µ 0 + A log 1 + + B log 1 +
v0 vθ 0

Ball Bearing Friction is Similar to Solid Friction Dahl’s Model - Continued

The stress-strain curve
dF  F α
=σ 1− sgn v
dx Fc
Differentiate with respect to time
dF dF dx dF  F α
= = v=σ 1− sgn v v.
dt dx dt dx Fc
For α = 1
dF F
= σ v − σ pvp.
dt Fc
Introduce F = σ z then
dz σ pvp
= v− z,
dt Fc
F = σ z.
c K. J. Åström TDU, October 2005
F 4
 Dahl’s Model - Steady State Properties Properties of The Dahl Model
dz
= v−
σ pvp
z, • Simple dynamic model
dt Fc
F = σ z. • Used extensively in simulation studies
In steady state • Captures many phenomena
Fc Zero slip displacement
z= sgn v
σ Hysteresis
F = Fc sgn v
• Friction depends only on displacement
The steady state version of Dahl’s model is thus Coulomb • Does not capture Stribeck effect
friction.
• Does not capture stick-slip

The Bliman-Sorine Model Properties of the Bliman-Sorine Model

Idea: • Dahl models in parallel
• Generalize Dahl to obtain Stribeck effect • At least second order dynamics
• Keep rate independence • Captures many properties
The sliding variable • Captures stiction
• Friction depends on displacement
Z t
s= pv(τ )pdτ .
0 • Friction does not depend on velocity
The model
dxs • Related to hysteresis models
= Axs + Bvs
ds
F = Cxs
−1/(ηε f ) 0 f1 /(ηε f )
   
A= , B= , C = (1 1),
0 −1/ε f − f2 / ε f

c K. J. Åström TDU, October 2005

F 5
 Friction Models and Friction Compensation The LuGre Model
Idea: Generalize Dahl to obtain Stribeck effect and stick-slip
1. Introduction
The Dahl model
2. Friction Models dz σ pvp
= v− z,
3. The LuGre Model dt Fc
F = σ z.
4. Effects of Friction on Control Systems
The LuGre Model
5. Friction Compensation ˆ
6. Summary dz
= v−
pvp
z,
dt ˆ(v)
ˆ(v) = lc + (ls − lc ) e−pvp/vs
dz
F = σ0z + σ1 + σ 3v
dt
Other forms of F and ˆ possible v

Bristle Interpretation Steady State Properties of the LuGre Model

dz pvp
The variable z in the LuGre model can be interpreted as the = v − σ0 z,
dt ˆ(v)
average bristle deflection! dz
F = σ0z + σ1 + σ 2v
dt
In steady state z = 0 and v = v0 con-
F
dz pvp stant
= v− z,
dt ˆ(v) ˆ(v0 )
z0 = sgn v0
ˆ(v) = lc + (ls − lc ) e−pvp/vs σ0
dz F = ˆ(v) sgn v0 + σ 2 v0
F = σ0z + σ1 + σ 2v
dt
The term σ 2 v represents viscous
Variable z has dimension length friction
v

c K. J. Åström TDU, October 2005

F 6
 Functions f and ˆ Give Flexibility Theorem 1 - Boundedness of z
Fss = ˆ(v) sgn(v) + f (v) Assume that 0 < ˆ(v) < a. Then Ω = { z : p zp ≤ a} is an
invariant set for the LuGre model. If p z(0)p < a then p z(t)p ≤ a
for all t ≥ 0.

dz pvp Proof
= v− z,
dt ˆ(v)
v We have
dz dz pvp
F = σ0z + σ1
dt
+ f (v) = v− z
dt ˆ(v)
For z = a we have
dz pvp
= v− a ≤ v − pvp
dt ˆ(a)
• Functions f and ˆ can be chosen so that Fss matches
dz/dt is thus either constant or negative and z cannot be
measured steady state friction
larger than a. Applying the same argument at z = −a gives
• Function Fss is often asymmetrical, easy to deal with the result.

Theorem 2 - Dissipativity Application of Constant Force

Consider the equation Consider a mass m and apply a constant force Fd
dz pvp
= v− z m
dv
= Fd − F
dt ˆ(v) dt
dz pvp
The map ϕ : v → z is dissipative with respect to the energy = v− z,
dt ˆ(v)
function V = z2 /2.
ˆ(v) = lc + (ls − lc ) e−pvp/vs
Proof dz
F = σ0z + σ1 + σ 2v
dt
Along trajetories of the differential eqution we have
Hence
t t
pvp z2 
Z t Z t
pvp z 
Z Z
dz 
V (t) − V (0) = z dt = z v− dt = zv − dt ≤ zvdt dv dz zpvp
0 dt 0 ˆ(v) 0 ˆ(v) 0 m = Fd − σ 0 − σ 1 = −σ 1 v − σ 0 z + σ 1 + Fd
dt dt ˆ(v)
Hence dz zpvp
Z t = v−
dt ˆ(v)
zvdt ≥ V (t) − V (0)
0
c K. J. Åström TDU, October 2005
F 7
 Application of Constant Force Linearization
Linearization gives the dynamics matrix
dv zpvp
m = −σ 1 v − σ 0 z + σ 1 + Fd = −σ 1 v − σ 0 z + z f (v) + Fd  σ (1 − z f ′ ) −σ + σ f
dt ˆ(v)

− 1 0 1
A=

dz zpvp  m m 

= v− = v − z f (v)
 
dt ˆ(v)
′
1 − zf −f

σ 0 ˆ(v) where f (v) = pvp/ˆ(v). We have

Equilibria exists only for p Fd p ≤ σ 0 ls σ1
trace A = −(1 − z f ′ ) − f
Fd m
z= σ0
σ0 det A = (1 − z f ′ )
m
Fd pvp
v= , [ Fd = σ 0 ˆ(v) sgn(v)
σ 0 ˆ(v) At the equilibrium v = 0, z = F D /σ 0 we have f = f ′ = 0 and
v we get trace A = −σ 1 /m and det A = σ 0 /m. The characteristic
Two equilibria v = 0 or Fd = σ 0 ˆ(v0 ) sgn(v0 ), with Fd = σ 0 z0 in polynomial of the A-matrix is
both cases. The second only for σ 0 lc ≤ p Fd p ≤ σ 0 ls
σ1 σ0
s2 + s+
m m

Model Parameters Simulation of Application of Force

For a mass m which moves subject to Coulomb friction we

Position
have
σ 0 lc = µ m ˆ
If we start by specifying lc we thus find that σ 0 = µ mˆ/lc .
To find the parameter σ 1 we make the assumption that the

Velocity
polynomial
σ1 σ0
s2 + s+
m m
has roots with critical damping. Hence σ 12 = 4mσ 2 or
z

√
r
µˆ
σ 1 = 2 mσ 0 = 2m
lc
Parameter ls is typically 50 to 100 % larger than lc . The friction
F

model is characterized by only two parameters µ and lc .

c K. J. Åström TDU, October 2005

F 8
 Properties of LuGre Model Friction Models and Friction Compensation
• Almost as simple as the Dahl model
• Captures many aspects of friction 1. Introduction
– Stiction 2. Friction Models
– Stick slip 3. The LuGre Model
– Stribeck 4. Effects of Friction on Control Systems
– Hysteresis Stick slip motion
– Zero slip displacement Inverted pendulums
• But not all Servo systems
– Some hysteresis related phenomena 5. Friction Compensation
– Item for research! 6. Summary
• Is passive if damping is velocity dependent
– Very important for control design

Effects of Friction in Motion Control Hunting

Friction can has both benefits and drawbacks Friktion
−F
• Essential part in many drive mechanisms Capsubot xd
u 1 v 1 x
PID Σ
• Friction can give rise to oscillations and poor precision
ms s

Stick slip can occure whenever the mechanisms

• Stiction
Position x
1.5

• Instability
1

0.5

are present. Typical examples are 0

Tid
0 20 40 60 80 100

• Traditional stick-slip, spring is the instability mechanism F

1

• Inverted pendulum, gravity is the instability mechanism 0

• Servo systems, integral is instability mechanism. −1

Tid
0 20 40 60 80 100

c K. J. Åström TDU, October 2005

F 9
 Stick-slip Motion Simulation of Stick Slip

A classic phenomena Parameters

k d2 x
m m 2 = k( y − x) − F m=1
dt
F k=2
y = vo t
µ = 0.3
l = y−v
x y lc = 0.003
dl
= v0 − v ls = 2lc
dt
Equations of motion dv vs = 0.1
m = kl − F
dt vd0 = 2
d2 x dz
m = k( y − x) − F F = σ0z + σ1 + σ 2v k=2
dt2 dt µ mˆ
dl dz pvp σ0 =
= vd − v = v− z lc
dt dt ˆ(v) √
dv σ 1 = 2 mσ 0
m = kl − F ˆ(v) = lc + (ls − lc ) e−v/vs
dt σ 2 = 0.0

Simulation of Stick Slip Simulation of Stick Slip - Details

l

l
v

v
z

z
F

c K. J. Åström TDU, October 2005

F 10
 Projection on l, v Plane s2 = 0 Projection on l, v Plane s2 = 0.2
v

v
l l

Effects of k and s2 Effects of Parameter Variations

dl
Equilibrium
Equations of motion dt
= v0 − v
σ 0 lc sgn(v0 ) + σ 2 v0 Fc + σ 2 v0
dv l= =
dl m = kl − σ 0 lc sgn(v) − σ 2 v k k
= v0 − v dt
dt v = v0
m
dv
= kl − F Equilibrium σ2
dt Relative damping ζ = √
dz pvp σ 0 lc sgn(v0 ) + σ 2 v0 2 mk
= v− z l=
dt ˆ(v) k
dz v = v0 σ 2 shifts equilibrium horisontally, influences damping and
F = σ0z + σ1
dt
+ σ 2v existence of limit cycle
Characeristic polynomial
ˆ(v) = lc + (ls − lc ) e−v/vs k shifts equilibrium horisontally, influences damping and
σ2 k existence of limit cycle
z has little influence except s2 + s+
when v is very small
m m v0 shifts equilibrium horisontally and vertically, influences and
Relative damping ζ = σ2
√ existence of limit cycle
2 mk

c K. J. Åström TDU, October 2005

F 11
 Effect of Viscous Friction σ 2 Effect of Pulling Velocity vd
v σ2 = 0 σ 2 = 0.2 vd = 0.2 vd = 0.5

v
l l l l
√
σ 2 = 0.5 σ2 = 2 2 vd = 2 vd = 2.3
v

v
l l l l

Effect of Spring Coefficient k Stability of Sliding Equilibrium

k = 0.5 k=2 Equations of Motion Dynamics matrix
dl 0 −1 0
 
= v0 − v 
σ 1 (1− z f ′ )+σ 2

dt A= k
− σ 0 −mσ 1 f
 
 − 
m m

dv
v

 

m
dt
= kl − F 0 1 − zf′ −f
dz pvp
= v− z Characteristic polynomial
dt ˆ(v)
l l
F = σ0z + σ1
dz
+ σ 2v
s3 + a1 s2 + a2 s + a3
k=5 k=8 dt
ˆ(v) = lc + (ls − lc ) e−v/vs pvp
f (v) =
ˆ(v)
Equilibrium
σ 1 (1 − z f ′ ) + σ 2
v

a1 = + f
v = v0 m
z = ˆ(v0 ) sgn v0 σ 0 (1 − z f ′ ) + σ 2 f + k
a2 =
m
l = (σ 0 z0 + σ 2 v0 )/ k
l l kf
a3 =
c K. J. Åström TDU, October 2005
F m 12
 Specifics Specifics ...
pv0 p sˆn(v0 ) Instability for small v for the simulation example
For large v we have ˆ = lc , f = and f ′ = then
lc Lc
Parameters

a1 a2 − a3 , Fss
′
a1 = σ 2 + f
lc = 0.001
σ 2 pvp
a2 = k + ls = 0.002
lc
kpv0 p µ = 0.3
a3 = a1 a2 − a3 = σ 22 f + σ 2 f 2 + σ 2 k 0.5
lc vs = 0.1 v

The sliding equilibrium is thus stable when the velocity is large.

ˆ = 10 The red curve shows the derivative of the
Stick-slip oscillation requires intial conditions sufficiently far m=1 static friction function Fss′ (v) for v > 0
k=1
from the equilibrium.  
Fss (v) = σ 0 z + σ 2 v = σ 0 lc + (ls − lc ) e−v/vs + σ 2 v
µ mˆ
σ0 =
lc ls − lc −v/vs
√ ′
Fss (v) = −σ 0 e + σ2
σ 1 = 2 ms0 vs
σ 2 = 0.2
Stable for v > 0.501

Analysis Summary
Assuming that z is much faster than the other states the model The stick slip behavior of the LuGre modelis complex.
can be approximated by the singularly perturbed system
• Useful to approximate by neglecting z gives a crude
dl
= v0 − v picture which allows projection on the l , v plane.
dt
dv • The zone around the strip v = 0 and 0 ≤< l ≤ Fs / k acts
m
dt
= kl − Fss (v) like an attractor.
Fss = σ 0 ˆ(v) sgn v + σ 2 v • Solutions can pass through the strip because the problem
The linearization of this system has the dynamics matrix is really three dimensional.

0 1 
 • There is an equilibrium where the velocity and the length
A=
 ′ are constant. The stability of this equilbrium depends on
 k − Fss 
 
the parameters.

m m
This system is stable if • The limit cycle will typically disappear when v or k are
large.
σ 0 ˆ′ (v) sgn(v) + σ 2
Fss′ (v) = >0
m
c K. J. Åström TDU, October 2005
F 13
Friction Models and Friction Compensation Friction Compensation
Methods to reduce effects of friction
1. Introduction
• Dither
2. Friction Models
• Acceleration feedback
3. The LuGre Model
• Model based friction compensation
4. Effects of Friction on Control Systems
• Adaptive friction compensation
5. Friction Compensation
Requirements on system and computations
Exploiting passivity
Servo systems • System structure
The Furuta Pendulum • Velocity measurements and estimates
6. Summary • Computational requirements
Control design methods
• Passivity based designs

Friction Compensation Friction Compensation

System Structure
u = u f b − F̂
d ẑ
F̂ = σ 0 ẑ + σ 1
dt
+ σ 2v G(s)
d ẑ pvp
= v − σ0 ẑ
dt ˆ(v)
2
/v2s
ˆ(v) = Fc + ( Fs − Fc ) e−v

• More general observers F Fric. v

• Velocity measurements Model
• Little dynamics from u to F
Much theory available

c K. J. Åström TDU, October 2005

F 14
 Model Based Friction Compensation Performance Degradation due to Friction
• Laboratory experiments 0.05 y
– Servo systems 0

– Inverted pendulums −0.05

0 20 40 60 80 100
• Industrial experiments F
0.4
– Hydraulic robots
0
– Electric robots
– Positioning systems −0.4
0 20 40 60 80 100
ulin
0.4

−0.4
0 20 40 60 80 100

Probability Distributions With Friction Compensation

y
5⋅10
−3
• Non-gaussian distributions
• Fat tails −5⋅10
−3

• Measures such as variance is questionable 0 20 40 60 80 100

F − F̂
0.05

−0.05
0 20 40 60 80 100
ulin
0.05

−0.05

0 20 40 60 80 100

c K. J. Åström TDU, October 2005

F 15
Laboratory Experiment Effect of Friction Compensation

Parameter Sensitivity Adaptive Friction Compensation

c K. J. Åström TDU, October 2005

F 16
 Experiments with the Furuta Pendulum Effect of Friction

Arm angle ϕ , pendulum angle θ .

Comparison with Simulations Experiment with Friction Compensation

c K. J. Åström TDU, October 2005

F 17
 Friction Compensation Based on Well Tuned Friction Models and Friction Compensation
LuGre Model
1. Introduction
2. Friction Models
3. The LuGre Model
4. Effects of Friction on Control Systems
5. Friction Compensation
Exploiting passivity
Servo systems
The Furuta Pendulum
6. Summary

Summary References
• A classical field J. N. Bhushan, J. N. Israelachvili, and U. Landman. “Nanon-
• Great interest in many disciplines tribology: Friction, wear and lubrication at the atomic scale.”
Nature, 374, pp. 607–616, 1995.
New measurement techniques and new sensors
F.P. Bowden and D. Tabor. The friction and Lubrication of
• Essential in all motion control systems
Solids. Oxford Univ. Press, Part I and II Oxford, 1950 and
• Particularly micro-mechanical systems 1964.
• Static and dynamic models P. Dahl. A solid friction model. Technical Report TOR-
Dahl, Bliman-Sorine, LuGre 0158(3107–18)-1, The Aerospace Corporation, El Segundo,
• Friction compensation CA, 1968.
Model based P.-A. Bliman and M. Sorine. Friction modelling by hysteresis
Accelerometer feedback operators. application to Dahl, sticktion and Stribeck effects. In
Proceedings of the Conference “Models of Hysteresis”, Trento,
The need for adaptation
Italy, 1991.
• The need for adaptation and better models
c K. J. Åström TDU, October 2005
F 18
 References
Nam P. Suh. Tribophysics. Englewood Cliffs, N.J.: Prentice-Hall,
1985.
Ernest Rabinowicz. Friction and wear of materials. New York:
Wiley, second edition, 1995.
C. Canudas de Wit, H. Olsson, K. J. Åström, and P. Lischinsky.
“A new model for control of systems with friction.” IEEE Trans-
actions on Automatic Control, 40:3, 1995.
Israelachvili, J.N. and A. D. Berman. (1999) Surface Forces
and Microrheology of Molecularly Thin Liquid Films. Handbook
of Micro/Nano Tribology. pp. 371–432.
Jianping Gaou, et al. (2004) Frictional Forces and Amontons’
Law: From the Molecular to the Macroscopic Scle. J. Phys.
Chem. B (108) 3410-3425.

c K. J. Åström TDU, October 2005

F 19