Transcript 3D Muscle Models

Meeting, Sept 11
Muscle Fibers and Volumetric Models
1
Muscle Fibers
• Muscle fiber orientation (pennation) has direct impact on skeletal
forces
• Not all fibers are activated at once, they are excited by groups of
motor neurons
• Point-to-point models incorporate a single pennation angle along
action line (Zajac 89, Delp 90)
2
3D Models
• Allow for contact, volume preservation and non-penetration
constraints
• Scheepers 97 / Wilhelms 97 / Kahler 02
• Implicit surface techniques
• Potential field defined by blending ellipsoid primitives
• Chen 92 / Zhu 98:
• Simplified linear FEM model, muscle surface embedded in an FEM lattice
• No internal muscle architecture
• Hirota 01:
• Isotropic FEM with contact forces, but no force generation
3
3D Models
• Johansson 00:
• FEM including active component in constitutive, toy examples with single fiber
direction
• Lemos 01:
• FEM, multi-pennate toy example, for volumetric deformation
4
3D Models
• Teran 05:
• Finite Volume Method, applied to tetrahedral elements
• Body-Centered Cubic (BCC)Tetrahedral lattice
• 10x speed-up, compared to FEM, but results not validated
• Blemker 05:
•
•
•
•
FEM with active component/fiber directions in constitutive model
Use hand-crafted template fiber arrangement
Geometric/moment-arm validation for hip flexion
Currently used in ArtiSynth
5
Pseudo-3D Models
• “Graphical” 3D models
• Dynamics driven by p2p model along line of action
• Volumetric deformation based on length of action line
• B-Spline solid (Ng-Thow-Hing 00), radial forces (Porcher-Nedel 98, Aubel 02),
medial representation (Gilles 07)
• A type of “skinning”, contact forces can be transmitted back to medial lines
6
Key Points
• Fiber pennation is important, should be captured in model
• Most muscle simulations still use point-to-point representations
• 3D methods generally lack validation, and include only a basic
description of fiber patterns
• FEM meshes are either hand-crafted or tetrahedral
• Tetrahedral meshes exhibit locking artifacts, hex meshes are preferred
• Automatic hex mesh generation is still an open problem
• Much of current 3D research focuses on graphics applications, sacrificing fidelity
for speed
7
Mesh-Free Methods
• Relatively new field, mostly developed in last 20 years
• Eliminate much of the hassle involved in mesh generation
• Traditional FEM methods will require you to re-mesh an entire volume
to change scale, which is a non-trivial problem
• Rely on the “Weakened weak formulation” (W2)
• Can be point-based, edge-based, or cell-based
• Smoothed Point-Interpolation Methods (S-PIM)
• Can produce upper-bound solutions with no volumetric locking
(FEM methods typically produce lower-bound solutions)
• Offers possibility of "soft" models that work well with tetrahedra
• Smoothed Finite Element Method (S-FEM), linear version of S-PIM
• S-PIM and S-FEM are currently used in solid mechanics and
computational fluid dynamics problems
8
Frame-based Approach
• Francois Faure et al. 2011
• A type of “mesh-free” method
• Introduce material properties directly into the shape functions, as
opposed to simple radial-basis functions used in S-PIM
• Allow very coarse discretization for non-uniform stiffness
• Currently only linear, isotropic material
• Focused on applications in graphics
9
Questions to be answered:
• For what situations is a point-to-point muscle not sufficient (if any)
• For kinematic studies?
• For dynamic studies?
• What level of detail is required to show significant differences?
• Resolution of FEM mesh
• Resolution of muscle fiber description
• Can we develop a mesh-free method that incorporates the nonlinear/anisotropic behaviour of skeletal muscle tissue?
• Evaluation of models
• Compare to state-of-the-art point-to-point
• Various resolutions of FEM
10
Next Steps
• Construct bone-joint model of forearm/wrist/hand
• Implement point-to-point muscle/tendon model
• Align all arm fibers to FEM muscle meshes (or muscle meshes
to fibers)
• Implement FEM muscle model
• Note: still significant work to create valid meshes, attachment
areas, tendons
• Investigate existing mesh-free methods, with the goal of
creating one to handle muscle actuation
11
Issues with current model
• Fiber alignment:
• If geometries very different, alters
pennation angle significantly
• Take a look at Mayo. Rav.’s thesis,
attempted to compensate
• Incomplete muscles / missing tendon components
• Use tendons from fiber scans
• May need to go back to visible
human data
• Un-natural shapes
• Bicep/Tricep too large, tricep has
wrong number of heads
12