probabilistic analysis: applications to biomechanics students: saikat pal, jason halloran, mark...
TRANSCRIPT
Probabilistic Analysis: Applications to Biomechanics
Students: Saikat Pal, Jason Halloran, Mark Baldwin, Josh Stowe, Aaron Fields, Shounak Mitra
Collaborators: Paul Rullkoetter, Anthony Petrella, Joe Langenderfer, Ben Hillberry
The QuestionWhat do I learn from probabilistic modeling that I don’t
already know from deterministic modeling?• Distribution of performance
• Assessment includes variable interaction effects• Understanding of the probabilities associated with component
performance– Probability of failure for a specific performance level– Minimum performance level for a specific POF
• Sensitivity information
Two common applications• Evaluation of existing components
• Guidance for tightening/loosening the tolerances of specific dimensions
• Design of future components• Predict performance and identify potential issues prior to prototyping
and testing
Bounding predictions of TKR performance in a knee simulator
Stanmore Wear Simulator
Explicit FE Model
Research Question
What impact does variability in component placement and experimental setup have on the kinematic and contact mechanics results? Wear?
Approach• Experimental setup has inherent variability • To more rigorously validate the model
• Scatter to setup parameters ( and ) is introduced• Distributions of results evaluated
Computational Model
• Explicit FE model of Stanmore simulator (Halloran, Petrella, Rullkoetter)
• Rigid body analysis with optimized pressure-overclosure relationship
• Non-linear UHMWPE material• Simulated gait cycle
• Profiles: AP load, IE torque, flexion angle, axial force
• Computation time• Rigid-rigid 6-8
minutes/run• Rigid-deformable 6-8 hours/run
0
5
10
15
20
25
30
35
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35Tr
ue S
tres
s (M
Pa)
True Strain
Model Variables
Insert_Tilt
Init_Fem_FE
FEax_AP Fem_IEFEax_IS
IEax_ML
IE Axis
IEax_AP
Insert_VV
FE Axis
Coefficient of Friction
ML LoadSplit
(with medial offset)
ML
Spring Constant (K)
Probabilistic Approach
ProbabilisticInputs
Performance Measures
SensitivityFactors
Probabilistic Inputs• 4 translational alignments• 4 angular alignments• 4 experimental/setup variables
Output Distributions• Kinematics
• AP and IE position• Contact pressure• Wear
Probabilistic Model
Deterministic Inputs
Deterministic Inputs• Component geometry• Gait profile (ISO)• Material behavior
Variable Description Mean Value Std.Dev. (Level A)
Std.Dev. (Level B)
FEax_AP AP position of femoral FE axis 0 mm 0.25 mm 0.5 mm
FEax_IS IS position of femoral FE axis 25.4 mm 0.25 mm 0.5 mm
IEax_AP AP position of tibial IE axis 7.62 mm 0.25 mm 0.5 mm
IEax_ML ML position of tibial IE axis 0 mm 0.25 mm 0.5 mm
Init_Fem_FE Initial FE position of femoral 0° 0.5° 1°
Insert_Tilt Tilt of the insert 0° 0.5° 1°
Fem_IE Initial IE rotation of femoral 0° 0.5° 1°
Insert_VV Initial VV position of insert 0° 0.5° 1°
ML ML position of spring fixation 28.7 mm 0.25 mm 0.5 mm
ML_Load ML load split (60%-40%) 60% 1.0% 1.0%
K Spring constant 5.21 N/mm 0.09 N/mm 0.09 N/mm
Coefficient of friction 0.04 0.01 0.01
Model VariablesAll variables assumed as normal distributions
AP TranslationModel-predicted envelopes (1% to 99% confidence intervals) as a function of gait cycle
Max. Range: 1.79 mm (Level A ), 3.44 mm (Level B)
-6
-4
-2
0
2
0 20 40 60 80 100% Gait Cycle
AP
Tra
nsla
tion
(mm
)
Experimental
Level A
Level B
IE RotationModel-predicted envelopes (1% to 99% confidence intervals) as a function of gait cycle
Max. Range: 2.17° (Level A ), 4.30° (Level B)
-10
-8
-6
-4
-2
0
2
4
0 20 40 60 80 100% Gait Cycle
IE R
otat
ion
(Deg
)
ExperimentalLevel ALevel B
IE
4
8
12
16
20
0 20 40 60 80 100% Gait Cycle
Con
tact
Pre
ssur
e (M
Pa)
Level A
Level B
Peak Contact PressureModel-predicted envelopes (1% to 99% confidence intervals) as a function of gait cycle
Max. Range: 1.3 MPa (Level A ), 1.6 MPa (Level B) @ 40% Gait
Sensitivity FactorsNormalized absolute average of sensitivity over the entire gait cycle
0.0
0.2
0.4
0.6
0.8
1.0F
Eax
_AP
FE
ax_I
S
IEax
_AP
IEax
_ML
Init_
Fem
_FE
Inse
rt_T
ilt
Fem
_IE
Inse
rt_V
V
ML_
Load
ΔM
L K μ
Sen
sitiv
ity
AP
IE
CP
Parameter sensitivities varied significantly throughout the gait cycle
Evaluating Measurement Uncertainty in Predicted Tibiofemoral Contact Positions using Fluoro-driven FEA
Evaluating Measurement Uncertainty in Predicted Tibiofemoral Contact Positions using Fluoro-driven FEA
• Video fluoroscopy is widely used to obtainimplant kinematics in vivo• Evaluate performance measures
(e.g. range of motion, cam-post interaction)
• Uncertainty exists in spatial positioning of theimplants during the model-fitting process(Dennis et al., 1998)
• Due to image clarity, operator experience, and differences in CAD and as-manufactured geometries
• Errors up to 0.5 mm and 0.5° for in-plane translations and rotations (Dennis et al., 2003)
• Objectives: • Develop an efficient method to account for measurement uncertainty in
the model-fitting process
• Evaluate the potential bounds of implant center-of-pressure contact estimates
Methods• Probabilistic analysis based on previous
fluoro-driven FE model (Pal et al., 2004)
• Fixed-bearing, semi-constrained, Sigma PS implant• Weight-bearing knee bend from 0° to 90°
• Inputs: Six DOFs describing pose of each componentat each flexion angle (0° to 90°, at 10° intervals)• Gaussian distributions with mean based on model-fitting process• In-plane DOFs: SD = 0.17 mm and 0.17°• Out-of-plane DOFs: SD = 0.34 mm and 0.34°
• To allow both condyles to contact throughout flexion, model loading conditions were:• Compressive force and in vivo kinematics (AP, IE and FE)• Unconstrained in ML and VV
• Output: Distribution of contact location throughout flexion
Results• Substantial variability in AP
contact position observed• Average ranges:
• Medial: 10.9 mm (0°-30°) 5.4 mm (30°-90°)
• Lateral: 9.3 mm (0°-30°) 6.3 mm (30°-90°)
• Maximum ranges:• 12.2 mm (M) and 10.7 mm (L)
• Uncertainty in implant position affected cam-post interaction• Underscores the need for careful
procedures when extracting kinematics using fluoroscopy
0 20 40 60 80 100Flexion Angle (deg.)
Lateral
-20
-10
0
10
0 20 40 60 80 100Flexion Angle (deg.)
Mean1% Bound99% Bound
Medial
A(+
)/P
(-)
Po
sitio
n (
mm
)
Contact patches at 90° flexion
Predicted tibiofemoral contact positions
+
_
Medial Lateral
Medial Lateral
1% 99%
Effects of Bone Mechanical Properties on Fracture Risk
Assessment
Effects of Bone Mechanical Properties on Fracture Risk Assessment
• CT scans are often used to create geometry and material properties of bone • Assess bone stresses• Predict fracture risk• Evaluate implant load transfer
• Significant variability present in relationships between HU and Modulus and Strength
• What effect does this variability have on predicted stress and risk assessment?
Keller, 1994
Methods
Proximal femur under stance loading (Keyak et al., 2001)
20° Relationship Coefficient Exponent
E(GPa) = ab a ( = 1.99, = 0.30) b ( = 3.46, = 0.12)
S(MPa) = cd c ( = 26.9, = 2.69) d ( = 3.05, = 0.09)
CT dataNessus
Material Relations BoneMat
Abaqus StressRisk
Material relation variability (Keller, 1994)
-10
0
10
20
30
40
0 0.5 1 1.5 2Density [g/cm3]
Mod
ulus
[G
Pa]
0
100
200
300
400
500
Str
engt
h [M
Pa]
Modulus
Strength
Results• Average bounds (1-99%)
• Stress: 13.9 MPa• Risk: 0.25• Potential to impact findings of bone
studies
• Computation time < 2 hours• Variability should be considered
when applying lab-developed material relations to patient- specific bone models
Maximum Stress
Femur Mean Min Max
1R 111.0 104.2 117.7
1L 100.2 94.5 106.1
2L 217.1 203.2 233.5
Fracture Risk
Femur Mean Min Max
1R 0.396 0.314 0.523
1L 0.375 0.298 0.494
2L 0.658 0.521 0.870
1R
1L 2L
Summary
• Probabilistic analysis has been demonstrated as a useful computational tool in materials and biomechanics• Efficient MPP-based methods make probabilistic FE
analysis quite feasible• Knowledge of bounds (distributions) of performance
and important parameters useful in design decisions• Developed framework can be “easily” applied to most
computational models