This document is Roland Pastorino's 2012 doctoral thesis which experimentally validates a multibody vehicle model and applies it to state observers. It includes field testing of a prototype vehicle to collect experimental data, developing a multibody vehicle model in a simulation environment, validating the model against test data, and using the model for state estimation. The goal is to validate the modeling procedure and provide new insights into automotive research using multibody models and state observers.
A guiding light for the processes in your private cloud
Oral presentation
1. Experimental Validation of a Multibody
Model for a Vehicle Prototype and its
Application to State Observers
Roland Pastorino
A thesis submitted for the degree of Doctor Ingeniero Industrial
June 25, 2012, Ferrol, Spain
Mechanical Engineering Laboratory, University of La Coruña
2. 0 Outline
1 Motivations
2 Vehicle eld testing
3 Vehicle modeling and simulation environment
4 Validation results
5 State observers
6 Conclusions
Mechanical Engineering Laboratory, University of La Coruña 2/48
3. 1 Outline
1 Motivations
2 Vehicle eld testing
3 Vehicle modeling and simulation environment
4 Validation results
5 State observers
6 Conclusions
Mechanical Engineering Laboratory, University of La Coruña 3/48
4. 1 Motivations
Laboratorio de Ingeniería Mecánica (LIM).
Specialized in realtime simulations of rigid and exible multibody systems
Fig. 1 Realtime simulations of Fig. 2 Realtime simulations of
vehicles excavators
Fig. 3 Realtime simulations of Fig. 4 HumanInTheLoop
container cranes simulators of excavators
Mechanical Engineering Laboratory, University of La Coruña 4/48
5. 1 Motivations
Multibody dynamics analysis in the automotive eld.
comfort
ride
accuracy
analysis noise
vibration highly
durability
detailed
analysis
models
crash
worthiness
crash ease
ANALYSIS
analysis of use MB MODELS
TYPES
nonlinear
vehicle
handling dynamics special
analysis formu.
design
realtime eciency
tuning of reduced
app required
controllers models
HIL
HITL
Mechanical Engineering Laboratory, University of La Coruña 5/48
6. 1 Motivations
Multibody dynamics analysis in the automotive eld.
comfort
ride
accuracy
analysis noise
vibration highly
durability
detailed
analysis
models
crash
worthiness
crash ease
ANALYSIS
analysis of use MB MODELS
TYPES
nonlinear
vehicle
handling dynamics special
analysis formu.
design
realtime eciency
tuning of reduced
app required
controllers models
HIL
HITL
Mechanical Engineering Laboratory, University of La Coruña 5/48
7. 1 Motivations
Multibody dynamics analysis in the automotive eld.
comfort
ride
accuracy
analysis noise
vibration highly
durability
detailed
analysis
models
crash
worthiness
crash ease
ANALYSIS
analysis of use MB MODELS
TYPES
nonlinear
vehicle
handling dynamics special
analysis formu.
design
realtime eciency
tuning of reduced
app required
controllers models
HIL
HITL
Mechanical Engineering Laboratory, University of La Coruña 5/48
8. 1 Motivations
Multibody dynamics analysis in the automotive eld.
comfort
ride
accuracy
analysis noise
vibration highly
durability
detailed
analysis
models
crash
worthiness
crash ease
ANALYSIS
analysis of use MB MODELS
TYPES
nonlinear
vehicle
handling dynamics special
analysis formu.
design
realtime eciency
tuning of reduced
app required
controllers models
HIL
HITL
Mechanical Engineering Laboratory, University of La Coruña 5/48
9. 1 Motivations
Multibody dynamics analysis in the automotive eld.
comfort
ride
accuracy
analysis noise
vibration highly
durability
detailed
analysis
models
crash
worthiness
crash ease
ANALYSIS
analysis of use MB MODELS
TYPES
nonlinear
vehicle
handling dynamics special
analysis formu.
design
realtime eciency
tuning of reduced
app required
controllers models
HIL
HITL
Mechanical Engineering Laboratory, University of La Coruña 5/48
10. 1 Motivations
Multibody dynamics analysis in the automotive eld.
comfort
ride
accuracy
analysis noise
vibration highly
durability
detailed
analysis
models
crash
worthiness
crash ease
ANALYSIS
analysis of use MB MODELS
TYPES
nonlinear
vehicle
handling dynamics special
analysis formu.
design
realtime eciency
tuning of reduced
app required
controllers models
HIL
HITL
Mechanical Engineering Laboratory, University of La Coruña 5/48
11. 1 Motivations
Objectives.
Without validation of the vehicle dynamics there is only speculation that a
given model accurately predicts a vehicle response
A.H. Hoskins and M. El-Gindy, Technical report: Literature survey on driving simulator validation studies,
International Journal of Heavy Vehicle Systems, vol. 13 (3), pp. 241252, (2006)
reliability and validity of the
modeling procedure and
MB formulation to simulate
new insigths into comprehensive vehicle models
the automotive re-
search line of the LIM
state estimation using
MB models applied
to vehicle dynamics
Mechanical Engineering Laboratory, University of La Coruña 6/48
12. 2 Outline
1 Motivations
2 Vehicle eld testing
3 Vehicle modeling and simulation environment
4 Validation results
5 State observers
6 Conclusions
Mechanical Engineering Laboratory, University of La Coruña 7/48
13. 2 Vehicle eld testing
Validation methodology.
based on the validation methodology developed by the VRTC (Vehicle Research and
Test Center) for the NADS (National Advanced Driving Simulator)
composed of 3 main phases
1 experimental 2 vehicle parameter 3 simulation predictions
data collection measurement vs experimental data
vehicle eld testing parameter simulations using the same
maneuvers → broad range of measurements for the control inputs that were
vehicle operating conditions MB model and the measured at the 1st phase
subsystem models comparisons between the
repetitive maneuvers to
determine the random simulation experimental
uncertainty results
experimental postprocessing
Fig. 5 NADS bay
Mechanical Engineering Laboratory, University of La Coruña 8/48
14. 2 Vehicle eld testing
Diagram of the iterative validation methodology.
vehicle
reference
maneuver brake model tire model
maneuver MB formu.
repetitions + integrator
sensor outputs
vehicle model
Mechanical Engineering Laboratory, University of La Coruña 9/48
15. 2 Vehicle eld testing
Diagram of the iterative validation methodology.
test track topo. survey
virtual en-
vironment
vehicle
reference
maneuver brake model tire model
maneuver benchmark MB formu.
repetitions input data + integrator
sensor outputs
vehicle model
vehicle parameter measurement
Mechanical Engineering Laboratory, University of La Coruña 9/48
16. 2 Vehicle eld testing
Diagram of the iterative validation methodology.
test track topo. survey
virtual en-
vironment
vehicle
reference
maneuver brake model tire model
maneuver benchmark MB formu.
repetitions input data + integrator
sensor outputs sensor outputs
comparisons
vehicle model
vehicle parameter measurement
Mechanical Engineering Laboratory, University of La Coruña 9/48
17. 2 Vehicle eld testing
Diagram of the iterative validation methodology.
test track topo. survey
virtual en-
vironment
vehicle
reference
maneuver brake model tire model
maneuver benchmark MB formu.
repetitions input data + integrator
sensor outputs sensor outputs
comparisons
vehicle model
vehicle parameter measurement
Mechanical Engineering Laboratory, University of La Coruña 9/48
18. 2 Vehicle eld testing
The XBW vehicle prototype.
selfdeveloped from scratch
onboard digital acquisition system
longitudinal and lateral maneuver repetition
capability
Mechanical characteristics.
Fig. 6 Design and manufacturing
tubular frame
internal combustion engine (4 cylinders, 2barrel
carburator)
automatic gearbox transmission
front suspension → double wishbone
rear suspension → MacPherson
tyres → Michelin E3B1 Energy 155/80 R13 Fig. 7 XBW vehicle prototype
Mechanical Engineering Laboratory, University of La Coruña 10/48
19. 2 Vehicle eld testing
Steering wheel
Encoder B
The bywire systems of the prototype.
Precision gearbox
Encoders A Coreless DC motor
SBW → steerbywire Precision gearbox
Torque sensor
TBW → throttlebywire
Rack and pinion gear system
BBW → brakebywire
Fig. 8 Steerbywire system
Extra sensors.
Measured magnitudes Sensors
vehicle accelerations (X,Y,Z) accelerometers (m/s2 )
vehicle angular rates (X,Y,Z) gyroscopes (rad/s)
vehicle orientation angles inclinometers (rad)
wheel rotation angles halleect sensor (rad) Fig. 9 Throttlebywire
brake line pressure pressure sensor (kP a) system
steering wheel and steer angles encoders (rad)
engine speed halleect sensors (rad/s)
steering torque inline torque sensor (N m)
throttle pedal angle encoder (rad)
rear wheel torque wheel torque sensor (N m)
Fig. 10 Brakebywire system
Mechanical Engineering Laboratory, University of La Coruña 11/48
20. 2 Vehicle eld testing
Driver's force feedback of the SBW.
objective → accurate torque feedback to the
driver
problem → exibility, backlash friction
solution → highly accurate model of the
assembly ampliermotorgearbox
future work → model based torque controller
Fig. 12 Steering wheel system
4 1
3 2
4 Quadrant Linear
PMDC Motor Gearbox Steering Wheel
Servoamplifier
Fig. 11 Scheme of the modeling Fig. 13 CAD model of the steering wheel
system
Mechanical Engineering Laboratory, University of La Coruña 12/48
21. 2 Vehicle eld testing
7 repetitions of a low speed straightline maneuver.
total distance = 63.5 m
max speed = 23 km/h
Mechanical Engineering Laboratory, University of La Coruña 13/48
22. 2 Vehicle eld testing
6 repetitions of a lowspeed Jturn maneuver.
total distance = 59.6 m
max speed = 18 km/h
Mechanical Engineering Laboratory, University of La Coruña 14/48
23. 3 Outline
1 Motivations
2 Vehicle eld testing
3 Vehicle modeling and simulation environment
4 Validation results
5 State observers
6 Conclusions
Mechanical Engineering Laboratory, University of La Coruña 15/48
24. 3 Vehicle modeling and simulation environment
Vehicle modeling.
Type of coordinates : natural + some relative coordinates (angles distances)
MB formulation : index3 augmented Lagrangian formulation with
massdampingstinessorthogonal projections
Bodies / Variables : 18 → all the vehicle bodies / 168 → points and vectors
Steering : kinematically guided
Forces : gravity forces, tire forces, engine torque, brake torques
Degrees of freedom : 14 → suspension systems (4)
chassis (6)
wheels (4)
Fig. 14 CAD model of the prototype Fig. 15 Points vectors of the MB model
Mechanical Engineering Laboratory, University of La Coruña 16/48
25. 3 Vehicle modeling and simulation environment
Ecient MB formulation developed and used at LIM.
index3 augmented Lagrangian formulation with massdampingstinessorthogonal
projections
integration → the trapezoidal rule and the NewtonRaphson method
eq. of motion→
2 rst order trap. rule
predictor → ∆t
f = (Mq +
¨ approx. →
q, q, q
˙ ¨ 4
ΦT αΦ + ΦT λ − Q) fq ∆q = −f λ = λ + αΦ
q q
tangent matrix → eq. of motion→ tangent matrix →
∆t ∆t
fq = M + C+ ∆t 2
fq = M + C+
2 f = (Mq +
¨ 2
∆t2 T 4 ∆t2 T
(Φq αΦq + K) ΦT αΦ + ΦT λ − Q)
q q (Φq αΦq + K)
4 4
no
error
min
→
projections in velocities
t = t + ∆t ∆t ∆t2∆t2 T
fq q =
˙ M+ C+ K q −
˙ ∗
Φq αΦt yes
2 4 4
projections in accelerations →
∆t ∆t2 ∆t2 T
fq q = M +
¨ C+ K q −
¨∗
Φq α(Φq q + Φt )
˙ ˙ ˙
2 4 4
Mechanical Engineering Laboratory, University of La Coruña 17/48
26. 3 Vehicle modeling and simulation environment
Tire model.
part of the empirical and physical
TMeasy model
rstorder dynamics → longitudinal
and lateral deections
transition to standstill → stickslip
behavior
tire curves → linearized model
Fig. 17 Longitudinal deformation of the tire
Fig. 16 Points vectors for the tire modeling Fig. 18 Lateral deformation of the tire
Mechanical Engineering Laboratory, University of La Coruña 18/48
27. 3 Vehicle modeling and simulation environment
Road prole.
topographical survey of the test track with a total station
300 points for a track of 80 meters long
interpolation of the scattered points
Delaunay triangulation → mesh of triangles for the collision detection
1
Height (m)
0.5
0
0
80 −10
60
40 −20
20 −30
Length (m) 0 Width (m)
Fig. 19 Total station used for Fig. 20 3D scattered points Fig. 21 3D model of the test
the topographical survey track
Mechanical Engineering Laboratory, University of La Coruña 19/48
28. 3 Vehicle modeling and simulation environment
Collision detection.
tire normal force → function of the
interpenetration of the stepped
triangles of the ground mesh
4 spheres for the collision geometry of
Fig. 22 3D model of the test track
the tires
Simulation environment.
realistic graphical environment of the
campus
selfdeveloped 3D environment
opensource 3D graphics toolkit (C++)
inclusion of the topographical survey of
the test track
Mechanical Engineering Laboratory, University of La Coruña 20/48
29. 4 Outline
1 Motivations
2 Vehicle eld testing
3 Vehicle modeling and simulation environment
4 Validation results
5 State observers
6 Conclusions
Mechanical Engineering Laboratory, University of La Coruña 21/48
30. 4 Validation results
MB model inputs.
averaging of the experimental data
100 100
Torque (Nm)
0
Torque (Nm)
0
−100 −100
−200 −200
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time (s) Time (s)
Fig. 23 Repetitions (wheel torque) Fig. 24 Mean and CI (wheel torque)
Condence intervals.
assumption → the uncertainty follows a normal distribution
small number of samples → Student's t distribution
S
C.I. bounds: x ± tn−1
¯ (1−α/2) · √
n
n n
1 1
sample mean → x =
¯ xi sample variance → S 2 = 2
(xi − x)
¯
n i=1
(n − 1) i=1
(1 − α/2) critical value for the t
distribution with (n − 1) degrees of freedom
Mechanical Engineering Laboratory, University of La Coruña 22/48
31. 4 Validation results
Simulation of the low speed straightline maneuver.
MB model inputs → averaged experimental data
road prole → topographical survey
Mechanical Engineering Laboratory, University of La Coruña 23/48
32. 4 Validation results
Validation results for the low speed straightline maneuver.
25 3
Angular velocity ( °/s)
20 2
Speed (km/h)
15 1
10 0
5 −1
0 −2
−5 −3
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time (s) Time (s)
Fig. 25 Left front wheel speed Fig. 26 Roll rate of the chassis
1
Acceleration (m/s2)
0
−1
−2
0 2 4 6 8 10 12 14 16 18 20
Time (s)
Fig. 27 Longitudinal acceleration of the chassis
Mechanical Engineering Laboratory, University of La Coruña 24/48
33. 4 Validation results
Simulation of the lowspeed Jturn maneuver.
MB model inputs → averaged experimental data
road prole → topographical survey
Mechanical Engineering Laboratory, University of La Coruña 25/48
34. 4 Validation results
Validation results for the low speed Jturn maneuver.
5 3
Angular velocity ( °/s)
0
Acceleration (m/s2)
2
−5
−10 1
−15
0
−20
−25 −1
0 2 4 6 8 10 12 14 16 18 20 22 0 2 4 6 8 10 12 14 16 18 20 22
Time (s) Time (s)
Fig. 28 Yaw rate of the chassis Fig. 29 Lateral acceleration of the chassis
1
2
Angular velocity ( °/s)
Acceleration (m/s2)
0
0
−1
−2
−2
−4 −3
0 2 4 6 8 10 12 14 16 18 20 22 0 2 4 6 8 10 12 14 16 18 20 22
Time (s) Time (s)
Fig. 30 Roll rate of the chassis Fig. 31 Longitudinal acceleration of the
chassis
Mechanical Engineering Laboratory, University of La Coruña 26/48
35. 5 Outline
1 Motivations
2 Vehicle eld testing
3 Vehicle modeling and simulation environment
4 Validation results
5 State observers
6 Conclusions
Mechanical Engineering Laboratory, University of La Coruña 27/48
36. 5 State observers
Model running in realtime with the same inputs
DISTURBANCES
unknown
forces, etc
OUTPUTS
sensors set of
sensors
INPUTS
forces, Real Mechanism
torques, etc
CONTROLLERS
MODEL OUTPUTS
variables of model variables =
the model numerous avail-
Realtime model able magnitudes
Mechanical Engineering Laboratory, University of La Coruña 28/48
37. 5 State observers
Model running in realtime with the same inputs
DISTURBANCES
unknown
forces, etc
OUTPUTS
sensors set of
sensors
INPUTS
forces, Real Mechanism
torques, etc
CONTROLLERS
DRIFT OF THE MODEL
disturbances,unmodeled
dynamics, param-
eters of the model
MODEL OUTPUTS
variables of model variables =
the model numerous avail-
Realtime model able magnitudes
Mechanical Engineering Laboratory, University of La Coruña 28/48
38. 5 State observers
State observers for mechanical systems based on Kalman lters
DISTURBANCES
unknown
forces, etc
OUTPUTS
sensors set of
sensors
INPUTS
forces, Real Mechanism
torques, etc
CONTROLLERS
+
Kalman
corrections -
sensors of
the model
MODEL OUTPUTS
variables of
model variables
the model
= virtual sensors
Realtime model
State observer based on Kalman lters
Mechanical Engineering Laboratory, University of La Coruña 28/48
39. 5 State observers
State observers for mechanical systems based on Kalman lters
DISTURBANCES
unknown
forces, etc
OUTPUTS
sensors minimum set
of sensors
INPUTS
forces, Real Mechanism
torques, etc
CONTROLLERS
+
Kalman
the model follows the
corrections -
motion of the mechanism
with sensors of delay
minimum
the model
MODEL OUTPUTS
variables of
model variables
the model
= virtual sensors
Realtime model
State observer based on Kalman lters
Mechanical Engineering Laboratory, University of La Coruña 28/48
40. 5 State observers
State estimation in mechanical systems.
Advantages
minimum set
new sensors virtual sensors
of sensors
ubiquitous reduced
reduced costs
magnitude failure rate
The more detailed the model is,
the more it provides information about the motion of the mechanism
Past researches
Kalman lter + linear mechanical systems lack of generality
linearized Kalman lter + nonlinear mechanical linear models
systems simple nonlinear models
Mechanical Engineering Laboratory, University of La Coruña 29/48
41. 5 State observers
First implementations using a 4bar linkage and a VW Passat.
Recent researches at the LIM
extended Kalman lter + realtime MB general approach
models complex nonlinear models
A
D
B
C
s
Fig. 32 4bar linkage Fig. 33 VW passat model state observer w/
MB model
Mechanical Engineering Laboratory, University of La Coruña 30/48
42. 5 State observers
Description.
simple mechanism: 5bar linkage → 2 DOFs
mechanism parameters → experimental
measurements
parameters of the sensors → characteristics from
otheshelf sensors
MB Modeling.
Fig. 34 5bar linkage image
natural coordinates (8 variables, 6 constraints, 2
DOFs)
MB formulations
independent coordinates → projection matrixR
method
dependent coordinates → penalty formulation
simulation of the real mechanism for
Fig. 35 5bar linkage modeling scheme
comprehensive comparisons
known errors between the real mechanism and
its MB model → lengths, masses, inertias. . .
Mechanical Engineering Laboratory, University of La Coruña 31/48
43. 5 State observers
Free motion simulation.
Drift of the model due to errors in the parameters of the model
real mechanism
(matrixR, trap. rule)
model
(matrixR, trap. rule)
Mechanical Engineering Laboratory, University of La Coruña 32/48
44. 5 State observers
Comparison of NL Kalman lters. SPKF
EKF ,UKF ,SSUKF
What is it? de facto NL Kalman lter recent NL Kalman lters
Why to use it ? eciency accuracy and easy implementation
Which form ? continuous discrete
How does it state estimates → propaga- state estimates → propagation
work ? tion through NL system. through NL system.
mean and state estimation mean and state estimation uncer-
uncertainty → propagation tainty → propagation of sigma
through linearization points through the NL system
Assumptions additive white Gaussian noises
rst order dierential equations
x(t) = f(x(t)) + w (t)
˙
with independent states
System
dynamics
nonlinear dierence equations
xk+1 = φk (xk ) + w k
with independent states
Mechanical Engineering Laboratory, University of La Coruña 33/48
45. 5 State observers
The extended Kalman lter EKF.
rst order ODEs rst order DAEs
Mq + ΦT λ = Q
¨ q
x(t) = f(x(t))
˙ =
Φ=0
Mechanical Engineering Laboratory, University of La Coruña 34/48
46. 5 State observers
The extended Kalman lter EKF.
rst order ODEs rst order DAEs
Mq + ΦT λ = Q
¨ q
x(t) = f(x(t))
˙ =
Φ=0 independent coordinates
duplication of variables state space reduction
(positions velocities) method matrix R
z(t)
x= ¨ = (RT MR)−1 [RT (Q − MRz)]
z ˙˙
w(t)
z(t)
˙ w
= =
w(t)
˙ (RT MR)−1 [RT (Q − MRz)]
˙˙
integration
implicit integrator
trapezoidal rule
Mechanical Engineering Laboratory, University of La Coruña 34/48
47. 5 State observers
The extended Kalman lter EKF.
state estimates ˆ(t) = f(ˆ(t), t) + K[y(t) − ˆ(t)]
x
˙ x ¯ y
Kalman gain K(t) = P(t)H[1]T (t)R(t)
¯
Riccati equation P(t) = F[1] P(t) + P(T)F[1]T + GQ(t)GT − KR(t)K
˙ ¯ ¯
∂ f(x, t) ∂ h(x, t)
linear approximations F[1] (t) H[1] (t)
∂x ∂x
0 I
0 I
F[1] (t) = ∂(M−1 Q)
¯ ¯ ¯ −1 ¯
∂(M Q) −1 −1
−M
¯ RT (KR + 2MRq Rw)
˙ −M
¯ RT (CR + MR)
˙
∂z ∂w
Mechanical Engineering Laboratory, University of La Coruña 35/48
48. 5 State observers
The unscented Kalman lter UKF.
nonlinear recurrences
rst order DAEs
with independent states
Mq + Φ T λ = Q
¨ q
xk+1 = φk (xk ) =
Φ=0
Mechanical Engineering Laboratory, University of La Coruña 36/48
49. 5 State observers
The unscented Kalman lter UKF.
nonlinear recurrences
rst order DAEs
with independent states
Mq + Φ T λ = Q
¨ q
xk+1 = φk (xk ) =
Φ=0
independent coordinates
state space reduction
method matrix R
¨ = (RT MR)−1 [RT (Q − MRz)]
z ˙˙
= integration
Mechanical Engineering Laboratory, University of La Coruña 36/48
50. 5 State observers
The unscented Kalman lter UKF.
nonlinear recurrences
rst order DAEs
with independent states
Mq + Φ T λ = Q
¨ q
xk+1 = φk (xk ) =
Φ=0
independent coordinates
dependent coordinates state space reduction
penalty formulation method matrix R
q = (M + ΦT αΦ)−1 [Q − ΦT α(Φq q + 2ζω Φ + ω 2 Φ)]
¨ q q
˙ ˙ ˙
¨ = (RT MR)−1 [RT (Q − MRz)]
z ˙˙
= integration
= integration
Mechanical Engineering Laboratory, University of La Coruña 36/48
51. 5 State observers
The unscented Kalman lter UKF.measurementupdate equations (executed if
timeupdate equations (always executed) information from the sensors is available)
2L
x xk
ˆk = ˆ− + Kk (yk − ˆ− )
¯ yk
xk
ˆ− = wim χi,k|k−1
i=0
χk|k−1 = φk−1 (χk−1 ) Kk = Pxk yk P˜k
¯ −1
y
2L
x
ˆk−1
yk
ˆ− = ωim Y i,k|k−1
x Pk−1
χi,k−1 = ˆk−1 + γ i=0
i
x Pk−1
ˆk−1 − γ
i
Y k|k−1 = hk (χk|k−1 )
2L
Pxk =
−
ωic (χk|k−1 − xk
ˆ− )(χk|k−1 − xk
ˆ− )T
T
i=0 Pxk = P− − Kk P˜k Kk
xk
¯ y ¯
Mechanical Engineering Laboratory, University of La Coruña 37/48
52. 5 State observers
The unscented Kalman lter UKF.
timeupdate equations (always executed)
zk+1 , zk+1
˙
2L
xk
ˆ− = wim χi,k|k−1
integ.
i=0
χk|k−1 = φk−1 (χk−1 ) z
¨k+1
x
ˆk−1 zk+1
¨1 zk+1
¨2 zk+1
¨3 zk+1
¨4 zk+1
¨5
x Pk−1
χi,k−1 = ˆk−1 + γ
i
x Pk−1
ˆk−1 − γ
φk φk φk φk φk
i
z
¨k z
¨k z
¨k z
¨k z
¨k
zk
˙ zk
˙ zk
˙ zk
˙ zk
˙
z1
k z2
k z3
k z4
k z5
k
Fig. 36 Set of sigmapoints (2 dim. GR variable) ¨k , zk , zk
z ˙
Mechanical Engineering Laboratory, University of La Coruña 37/48
53. 5 State observers
The spherical simplex UKF.
same equations as for the UKF
reduced set of sigmapoints
UKF → (2L+1) sigmapoints
SSUKF → (L+2) sigmapoints
equations of the reduced set of sigmapoints
χj−1
0 for i = 0
0
χj−1
i
1 for i = 1, . . . , j
χj = −
i
j(j + 1)w1
0j−1
Fig. 37 Reduced set of sigmapoints (2
1 for i = j + 1
dim. GR variable
j(j + 1)w1
Mechanical Engineering Laboratory, University of La Coruña 38/48
54. 5 State observers
Free motion simulation.
Drift of the model due to errors in the parameters of the model
The observers estimate correctly the motion of the real mechanism
real mechanism
(matrixR, trap. rule)
model
(matrixR, trap. rule)
EKF
(matrixR, trap. rule)
UKF
(matrixR, trap. rule)
SSUKF
(matrixR, trap. rule)
SSUKF
(matrixR, RK2)
SSUKF
(penal, RK2)
Mechanical Engineering Laboratory, University of La Coruña 39/48
55. 5 State observers
Performance comparisons of the lters → eciency vs RMSE.
non multirate
10
∆tinteg = 2ms
9 ∆tsensors = 2ms
multirate
8
∆tinteg = 2ms
7 ∆tsensors = 6ms
multirate
case
6
EKF (matrixR, trap.
rule)
RMSE
5
UKF (matrixR, trap.
4 rule)
3 SSUKF (matrixR, trap.
rule)
non multi
2 SSUKF (matrixR, RK2)
rate case
1 Simulation SSUKF (penal, RK2)
100 125 150 175 200 225 250 time (%)
Mechanical Engineering Laboratory, University of La Coruña 40/48
56. 5 State observers
Data
acquisition EKF
system
param-
eters UKF
NL
mechanism Kalman SPKF SSUKF
STATE OB- lters
SERVERS
sensors NL Kalman lters
with MB models
indepen-
dent
coord.
MB
matrix
implicit integrators R
formulation
method
depen-
TR dent
coord.
explicit
penalty-
formu.
RK2
Mechanical Engineering Laboratory, University of La Coruña 41/48
57. 5 State observers
Pros Cons.
EKF
Pros → most ecient lter with independent coordinates
Cons → involved and errorprone calculation of the Jacobian, not suitable to employ
with dependent coordinates, not multirate
SPKFs
Pros → easiest implementation, possible use of dependent coordinates, highest
accuracy
Cons → high computational cost due to the sigmapoints
Mechanical Engineering Laboratory, University of La Coruña 42/48
58. 6 Outline
1 Motivations
2 Vehicle eld testing
3 Vehicle modeling and simulation environment
4 Validation results
5 State observers
6 Conclusions
Mechanical Engineering Laboratory, University of La Coruña 43/48
59. 6 Conclusions
Vehicle eld testing.
1 X-by-wire vehicle prototype from scratch
2 highly detailed model of the driver's force feedback system
3 repetitions of 2 reference manoeuvres in the campus
Vehicle modeling and simulation environment.
1 real-time 14 DOFs MB model
2 modelling of subsystems → tire, brake
3 topographical survey of the test track
4 realistic 3D simulation environment
Mechanical Engineering Laboratory, University of La Coruña 44/48
60. 6 Conclusions
Validation results.
1 condence intervals for the experimental data
2 simulation of the test manoeuvres using the experimental data
3 evaluation of the accuracy of the MB model
State observers.
1 use of MB models with the extended Kalman lter
2 application to a 4bar linkage and a VW Passat
3 use of MB models with SPKFs lters
4 implementation using a 5bar linkage
Mechanical Engineering Laboratory, University of La Coruña 45/48
61. 6 Conclusions
Future research.
1 eld testing
manoeuvres at higher speeds → new test track
GPS RTK for realtime positioning of the vehicle
2 vehicle modelling and simulation environment
better characterization of the tire curves
HumanInTheLoop simulation using experimental inputs
3 state observers
EKF in discrete form
research on the observability of MB models
tests of the UKF/SSUKF with more complex mechanisms
implementation using the MB model of the XBW prototype
Mechanical Engineering Laboratory, University of La Coruña 46/48
62. 6 Conclusions
Congress papers.
[1] J. Cuadrado, D. Dopico, J. A. Pérez, and R. Pastorino. Inuence of the sensored magnitude in the performance of
observers based on multibody models and the extended Kalman lter. In Proceedings of the Multibody Dynamics
ECCOMAS Thematic Conference, Warsaw, Poland, June 2009.
[2] R. Pastorino, M. A. Naya, J. A. Pérez, and J. Cuadrado. Xbywire vehicle prototype: a steerbywire system with geared
PM coreless motors. In Proceedings of the 7th EUROMECH Solid Mechanics Conference, Lisbon, Portugal, September
2009.
[3] J. Cuadrado, D. Dopico, M. A. Naya, and R. Pastorino. Automotive observers based on multibody models and the extended
Kalman lter. In The 1st Joint International Conference on Multibody System Dynamics, Lappeenranta, Finland, May 2010.
[4] R. Pastorino, M. A. Naya, A. Luaces, and J. Cuadrado. Xbywire vehicle prototype: automatic driving maneuver
implementation for realtime MBS model validation. In Proceedings of the 515th EUROMECH Colloquium, Blagoevgrad,
Bulgaria, 2010.
[5] R. Pastorino. La simulation des systèmes multicorps dans l'automobile. Interface, revue des ingénieurs des INSA de Lyon,
Rennes, Rouen, Toulouse, (111):22, 3ème et 4ème trimestre 2011.
[6] R. Pastorino, D. Dopico, E. Sanjurjo, and M. A. Naya. Validation of a multibody model for an xbywire vehicle prototype
through eld testing. In Proceedings of the ECCOMAS Thematic Conference Multibody Dynamics 2011, Brussels,
Belgium, 2011.
[7] R. Pastorino, M. A. Naya, A. Luaces, and J. Cuadrado. Xbywire vehicle prototype : a tool for research on realtime
vehicle multibody models. In Proceedings of the 13th EAEC European Automotive Congress, Valencia, Spain, June 2011.
[8] R. Pastorino, D. Richiedei, J. Cuadrado, and A. Trevisani. State estimation using multibody models and nonlinear Kalman
lters. In The 2nd Joint International Conference on Multibody Systems Dynamics, Stuttgart, Germany, May 2012.
[9] R. Pastorino, D. Richiedei, J. Cuadrado, and A. Trevisani. State estimation using multibody models and unscented Kalman
lters. In Proceedings of the 524th EUROMECH Colloquium, Enschede, Netherlands, 2012.
[10] E. Sanjurjo, R. Pastorino, D. Dopico, and M. A. Naya. Validación experimental de un modelo multicuerpo de un prototipo
de vehículo automatizado. In XIX Congreso Nacional de Ingeniería Mecánica (to be presented), Castellón, Spain, Nov. 2012.
Mechanical Engineering Laboratory, University of La Coruña 47/48
63. 6 Conclusions
Journal papers.
[1] J. Cuadrado, D. Dopico, J. A. Pérez, and R. Pastorino. Automotive observers based on multibody models and
the Extended Kalman Filter. Multibody System Dynamics, 27(1):319, 2011.
[2] R. Pastorino, M. A. Naya, J. A . Pérez, and J. Cuadrado. Geared PM coreless motor modelling for driver's force
feedback in steerbywire systems. Mechatronics, 21(6):10431054, 2011.
Journal papers in preparation.
[1] R. Pastorino, M.A. Naya, and J. Cuadrado. Experimental validation of a multibody model for a vehicle prototype.
Vehicle System Dynamics, 2012.
[2] R. Pastorino, D. Richiedei, J. Cuadrado, and A. Trevisani. State estimation using multibody models and nonlinear
kalman lters. Proceedings of the Institution of Mechanical Engineers, Part K: Journal of Multibody Dynamics,
2012.
Acknowledgments.
Spanish Ministry of Science and Innovation Grant TRA200909314 (ERDF funds)
Mechanical Engineering Laboratory, University of La Coruña 48/48