OpenSim Virtual Workshop Fall 2016

Josh Roth and Colin Smith

University of Wisconsin-Madison

How Do Changes in the Joint Lines Affect Passive Biomechanics of the Knee?

Big Picture Motivation

One key surgical factor associated with the outcome of total knee arthroplasty (TKA) is the component alignment [1, 2], and while modern technology allows for precise placement of the components, the alignment goal remains uncertain. Because multiple component alignment goals cannot be studied either in vitro or in vivo, computational analysis has potential to provide important insight. Accordingly, the overall goal of our research is to use OpenSim to investigate how different component alignment goals for TKA change the passive joint mechanics in the knee.

Workshop Goals

• Convert SIMM TKA model to OpenSim 3.3 format
• Perform forward simulations of clinical assessments of varus-valgus laxity at 0°, 45°, and 90° of flexion
• Determine how ±1°, ±2°, and ±3° changes in the distal femoral and proximal tibial joint lines change the kinematics, ligament forces, and compartmental contact forces during clinical assessments of varus-valgus laxity and passive flexion
• Develop framework to perform stochastic simulations of changes in the joint lines on the Open Science Grid

​

Accomplishments

​

• Built TKA model in OpenSim 3.3
• 6 DOF TF and PF joints with contact
• Ran Forward simulation of passive knee flexion
• Compiled OpenSim 4.0 on Linux
• Ran 10 Static Optimization jobs on High Throughput Computing Cluster (HTC)
• Wrote io and basic functionality of MATLAB toolbox for setting up and analyzing .mot files from HTC simulations
• Created github repo to share HTC toolbox

Josh Roth¹, Colin Smith²

How Do Changes in the Joint Lines Affect Passive Biomechanics of the Knee?

​

Check out our webinar on Nov 17 on using OpenSim with High Throughput Computing to perform probabilistic simulations of knee mechanics. An announcement will be on the OpenSim website in a few days.

https://github.com/clnsmith/opensim_htc

Model: 6 DOF Tibiofemoral and Patellofemoral

Elastic Foundaton Contact

Subject 2015 In vivo knee load Grand Challenge

​

Simulation of Passive Knee Flexion:

Forward simulation, with TF flexion prescribed 0-90^o and all other 5 TF and 6 PF DOFs unconstrained

Dario Cazzola¹, Ezio Preatoni¹ and Grant Trewartha¹

Application of synergies optimisation to neck functional movements

Big Picture Motivation

Our broader research goal is to better understand mechanisms of acute and chronic injuries during functional activities involving neck motions, such as office-based tasks, sport movements, and portable device use.

​

Workshop Goals

• Implement method to evaluate/validate synergies generated via NNMF and understanding ‘muscle activation space’.
• Compare simulated muscle activation (OpenSim Output driven by synergies) with experimental EMG for different number of synergies.
• Identify groups of muscles that match EMG and group of muscles that do NOT match EMG. Compare number of groups and minimum synergies needed (from VAF).

¹ University of Bath, UK

[W] Synergy matrix

[C] Synergy Act

Muscle Act vs EMG

No Synergy

Synergy

Workshop Accomplishments

• NNMF and VAF analysis implemented in Matlab;
• Run SO using simplified model (only 20 muscles);
• Analysis ‘muscle activation space’ via VAF for different number of synergies;
• Comparison simulated muscle act with EMG.
• Initial comparison between muscle grouping and VAF analysis

​

Dario Cazzola¹, Ezio Preatoni¹ and Grant Trewartha¹

​

​

Individual muscles VAF and average VAF (black line). Individual muscle VAF requires to be > 85%, whilst overall VAF >95%

Application of synergies optimisation to neck functional movements

Min num Synergies from VAF analysis

> 85%

Individual muscles

> 95%

-) Creation simplified version of neck model (20 muscles)

SO

Manual Synergies Vectors ‘W’

SynOpt

NNMF

VAF

NNMF

(8 synergies)

SynOpt

Flexion driven by only 8 synergies !!

Output: Simulated Muscle Activation

VAF analysis: SIMULATED MUSCLE ACT vs EMG

Flexors:CLEIDO OCC do not match with EMG → discard!

Extensors: Very bad match (except SCALENUS)

RESULTS

-) Number of muscle groups that match EMG is lower than min number of synergies (see VAF analysis on the left);

​

-) Initial muscle grouping is useful for designing future exp protocol;

​

-) Need more exp EMGs for running reliable initial NNMF, and allowing better muscle grouping;

Motivation Summary

​

Low Back Pain (LBP) is a common and costly health conditions. Manual lifting is one of the leading risk factors associated with its development. A full-body model with a detailed anatomy of the lumbar spine and trunk musculature is required to further explore (spinal joint loads, spinal stability, muscles forces) lifting methods such as the squat and stoop. The current full-body model (Raabe et al. 2016) will need to be validated for lifting activities specifically; it has not been validated in this particular task that involves large trunk angles.

​

Workshop Goals

​

• Remove the coordinate coupler constraints after IK analysis to see their influence on JR spinal loads
• Integrate bushings in the lumbar spine to represent the passive contributions of intervertebral discs and ligaments
• Compare the JR spinal loads obtained with the model to published data.

Accomplishments

​

• Added actuators to the coordinates where the coupler had been removed with different optimal values to solve SO
• Learned how to obtain the forces applied by the constraints using Force Reporter
• Started to evaluate the effect of the actuators on the JR forces
• Started implementing bushings into the model

Full body model with detailed back to evaluate one-handed & two-handed lifts

¹ University of Adelaide, Adelaide, Australia

Erica Beaucage-Gauvreau¹

Raabe et al. 2016

The coupler coordinates have an effect on the resulting spinal loads. However, their influence is still being evaluated. In addition, the presence of actuators to solve SO and their effects on the results are still being evaluated.

Full body model with detailed back to evaluate one-handed & two-handed lifts

Erica Beaucage-Gauvreau

Bryce Killen

Subject-specific MTU paths informed from MRI

Motivation Summary

​

Subject specific neuromusculoskeletal models have been shown to yield better estimations of tibiofemoral joint contact loads than generic or generic scaled models. While boney geometries have previously been incorporated into OpenSim models, subject specific MTU paths are the final step in creating subject specific models

Workshop Goals

​

• Use MRI based objective measures to place wrapping surfaces to prevent VM penetration into the femur
• Optimize wrapping surface placement to match normalized muscle moment arms across a ROM
• Replace wrapping surfaces with conditional paths points to mimic wrapping surface actions.

​

Accomplishments

​

• Subject-specific muscle path points
• Customized wrap cylinder position and size
• Started developing code to replace wrap surfaces with conditional path points on simplified model

Griffith University, Gold Coast, Australia. Menzies Health Institute Queensland. Innovations in Health Technologies

​

​

Bryce Killen

Subject-specific MTU paths informed from MRI

Blooper video caused by incorrect surface orientation

Subject-specific VM wrapping

Differences in MA of VM without optimization of wrapping surfaces

Challenges

​

• Origin is subjective point
• Correct orientation of objects
• MedPat point should be stored in Patella body which doesn’t exist in Gait2392
• Determining where the muscle wraps on the cylinder using the API - all functions not exposed

What’s next.. ?

​

• Include multiple muscles and wrap objects
• Write API functions to automate wrapping surface placement
• Include the patella in model for more accurate quadricep paths
• Write optimization for origin and insertion point

William K. Thompson¹, NASA Digital Astronaut Project (DAP)

Optimizing the Application of External Loading to Minimize Residual Forces

in an OpenSim Models of a Squat and Split-squat Exercise with a Harness

​

Big Picture Motivation

​

DAP uses modeling to predict and assess spaceflight health and performance risks, and to enhance countermeasure development. Our research goal is to use modeling to provide quantitative answers to questions regarding astronaut health and performance on NASA missions and countermeasure efficacy. Our current modeling efforts focus on mitigating the risk that, given the small size of candidate exercise devices on Exploration missions to deep space, they may not be sufficient to provide the localized loading stimulus required to maintain musculoskeletal performance in microgravity. We are investigating long bar vs. T-bar vs. harness vs. free weight (gold standard) loading to determine the best method for long duration missions, given the operational constraints of spaceflight. Modeling of the external force applied by the harness has proved to be challenging. Our current methods (e.g., fixed 4-point and fixed single-point loading) have resulted in unacceptably high residuals.

.

Workshop Goals

​

• Model the highly distributed loading of the torso during harness exercise by a single force with a known magnitude but a direction and COP that varies during the movement.
• Use the Matlab API to write optimization code that minimizes the output residuals (from ID) by varying the direction and COP of applied force.
• Use the known constraints
• Force magnitude = cable tension
• COP must lie within the torso segment, likely near the torso COM
• Find ways to speed up execution time

¹ NASA Glenn Research Center, Cleveland, Ohio

​

^{William K. Thompson}

Optimizing the Application of External Loading to Minimize Residual Forces

in an OpenSim Models of a Squat and Split-squat Exercise with a Harness

Workshop Accomplishments

​

• Wrote the optimization routine using the Matlab API to run multiple Inverse Dynamics analyses
• Used fmincon function to optimize the direction and placement of the external loading force vector that represents the harness loading so that the sum of squared and normalized residuals is minimized
• The routine is working and the optimizer converges.
• Force residual improvements are substantial, moments remain too high
• Execution times are <1 hour for a full repetition of the exercises (500-600 frames)

Harness loading during SLS exercise is modeled as a single force vector applied to the torso object

Forward Work

​

• The optimizer is still not finding a global minimum. Reducing the size of the input vector may help:
• Use only two force components and solve for the other
• Do not optimize fore-aft placement of force vector within the torso
• Experiment with weighting of the residuals--the code already handles this
• Execution time can likely still be reduced after identifying opportunities for coding efficiency

Kathleen Lewicki

Changes in Deltoid Activation Following Reverse Shoulder Arthroplasty

Big Picture Motivation

Reverse shoulder arthroplasty (RSA) is a procedure used to restore function and relieve pain for specific patient populations. Most often, RSA is recommended to patients with rotator cuff deficiencies and is accomplished by reversing the articulation and the center of rotation. Through the use of OpenSim, we hope to model the biomechanical changes that occur following RSA and examine how muscle function changes with these biomechanical changes.

​

​

Workshop Goals

• Compare and validate deltoid muscle activation for coronal abduction and/or scapular plane abduction in a native, healthy model
• Accurately model the biomechanical changes associated with neutral implantation and two lateralized implantation configurations (+3mm, +6mm)
• Quantify the maximum deltoid force and corresponding abduction angle for each configuration

Accomplishments

​

• Created models that move the lateral deltoid fiber as the humerus is moved with implantation configurations
• Learned the importance of reserve actuators and constraints
• Identified potential problems with the use of the model for this problem

Kathleen Lewicki

Changes in Deltoid Activation Following Reverse Shoulder Arthroplasty

Created OpenSim models and corresponding motion files for changes in center of rotation and humerus location

​

Obtained results for muscle activation that reasonable replicates EMG data for abduction, but reserve actuators at the glenohumeral joint are large (~⅓ of total moment for elevation) → this still needs to be resolved!

​

When reserve actuators are decreased, all aspects of the deltoid activation change and no longer resemble EMG data, or max out.

​

Additional work to be done: Further examine muscle parameters to ensure that muscle fiber lengths/slack lengths are appropriate

​

Andrew LaPre¹, Ericber Francisco¹, Mark Price¹, Vinh Nguyen¹

Robotic Linear Actuator

Motivation Summary

Currently our methods allow us to generate predictive gait simulations of persons with lower limb loss, walking on passive prosthetic devices. To perform a comparable simulation with a robotic device, the dynamics of the robotic actuator must be modeled accurately instead of using simplistic linear force models. This will narrow the gap in the research community between roboticists seeking realistic simulation techniques for human-robot interactions and the OpenSim platform.

Workshop Goals

The goal is to create a new class of actuator, containing a parameterized model of a robotic linear actuator. The actuator model will include effects from motor rotational inertia, however bodies will be modeled to represent the mass and inertial properties of the actuator components.

• Complete the source code for the new actuator
• Simulate a tug-of-war example responding to a step command
• Compare to a similar simulation in MATLAB Simulink for validation

Accomplishments

• DC motor actuator class implemented in OpemSim 3.3
• DC motor coupled with ball-screw mechanics to simulate linear actuator behavior in OpenSim
• Tug-of-war scenario between actuator and passive spring successfully modeled
• Actuator behavior validated against MATLAB model of motor characteristics.

¹University of Massachusetts Amherst

​

Robotic Actuator

Spring Element

Linear actuator “tug of war”

• Step input voltage
• Motor, 2:1 reduction transmission, 2.5mm lead ballscrew
• Actuator pulls against passive spring (not shown)
• Reaches equilibrium position as motor stalls

Andrew LaPre¹, Ericber Francisco¹, Mark Price¹, Vinh Nguyen¹

Robotic Linear Actuator

Challenges:

• Implementing more complex mechanical model (friction effects, mechanical efficiency)
• Practical nonlinearities - electrical current limit
• Working between v3.3 and v4.0 - eventually decided to implement in 3.3

<- Left: OpenSim Actuator Model (blocks indicate rotor and ballscrew rotation)

​

DC motor actuator: Responds to input voltage, outputs motor shaft position and electrical current (output torque).

​

Output matches MATLAB model of a motor with the same characteristics (Maxon EC-4pole 30)

Andrew LaPre¹, Ericber Francisco¹, Mark Price¹, Vinh Nguyen¹

Robotic Linear Actuator

<- Left: OpenSim model outputs

​

Right: MATLAB model outputs ->

Antoine Falisse¹, Gil Serrancolí^1,2

Automatic differentiation in Simbody

Motivation Summary

​

The use of automatic differentiation to compute derivatives (for Gradient, Jacobian, Hessian) is assumed to speed up the CPU time and increase the accuracy. In this project, we explore the feasibility of integrating this feature in Simbody using the open-source package ADOL-C.

Workshop Goals

​

• Build the 3 Simbody libraries using the ADOL-C data type adouble
• Obtain the equations of motion of a double pendulum in an automatically differentiable form
• Calculate the Jacobian of the equations of motion using automatic differentiation
• Compare and validate our results

¹ KULeuven, Belgium

² Polytechnic University of Catalonia

​

Accomplishments

​

• Simbody libraries built using adouble
• Equations of motion obtained in an automatically differentiable form (via function calcResidualForceIgnoringConstraints)
• Jacobian obtained via automatic differentiation
• Jacobian validated against derived Jacobian and Jacobian computed via finite differences

=> PROOF OF CONCEPT

Antoine Falisse¹, Gil Serrancolí^1,2

​

Automatic differentiation in Simbody

​

The datatype adouble was assigned to SimTK::Real (previously double), allowing ADOL-C to recognize active variables and save their expressions for differentiation.

The three Simbody libraries, SimTKcommon, SimTKmath and SimTKsimbody, were built using the ADOL-C datatype adouble

Mohammadhossein Saadatzi

A Simplified Musculoskeletal Model for Predictive Simulation of Human Gait

Big Picture Motivation

For muscle driven simulation of human gait, a new and a thoroughly validated musculoskeletal model has been released by Rajagopal et al. [1]. But, the model has 80 muscles for the lower limbs, which makes its use for predictive simulation impractical.

Major Muscles

During gait, only 21 muscles in each leg generate peak moments that reach above 10% of the peak joint moments at the hip, knee, or ankle. Focusing only on these muscles may be a reasonable approach.

Muscle Groups

Ideally, the number of muscles in the model should be reduced further. As shown in Figure 1 some muscles have similar moment profiles and they can be combined for further simplification. Based on my inspections so far, it seems feasible to have a model with 11 muscle groups (Table1).

Workshop Goals

The parameters of the muscles should be modified in the simplified model (the model with 11 muscles groups) in a way that

• the final model is capable of generating the necessary torques for locomotion, and
• the final model results in reasonably accurate metabolic energetics.

​

​

Fig 1: Ankle joint moment (ID), and muscles’ moment (CMC). ‘edl’ and ‘tibant’ muscles (and also ‘gasmed’ and ‘gaslat’) have similar moment profiles.

Mechanical Engineering Department

Colorado School of Mines

http://brl.mines.edu/

​

​

Mohammadhossein Saadatzi

​

A Simplified Musculoskeletal Model for Predictive Simulation of Human Gait

​

Table 1: Ratio of peak moment of 21 muscles on each leg to the maximum flexion/extension moments around hip, knee and ankle joints is more than 10% (LN: large number).

I implemented my project using OpenSim API in C++. I used CMC tool to compute the muscle forces and muscle moments around different joints, and to do the optimization, I used the IPOPT algorithm available in OpenSim API.

In my first step, I tried to combine iliacus and psoas to one muscle (iliopsoas). Based on the Table 1, as iliacus has a larger effect than psoas, I removed the psoas and modified the parameters of iliacus. Path points of iliacus and maximum isometric force of it constitute the optimization parameters. The following steps shows the pseudocode my programs:

1) Run CMC for main model for the experimental data (walking)

a. Load the moment around hip by iliacus and psoas (two arrays of doubles)

b. Add the iliacus and psoas moments (as iliopsoas moment) as the reference moment during

optimization

2) Optimize (minimize) the joint moment error for the modified model

a. Run CMC for model without psoas for the same experimental data

b. Load moment around hip by iliacus (an array) and subtract it from the reference iliopsoas moment

(an array). Use the Integral of the final array as the cost function of our optimization.

c. Using IPOPT algorithm, change the path point location and strength of the muscle to minimize the

above cost function.

The codes for this project are implemented successfully. However, CMC has been used to compute the objective function of the IPOPT algorithm. Hence, the written code needs more time to run. To improve the performance of my codes, I’m going to use Static Optimization instead of CMC.

Tishya Wren¹

Joint Reaction Analysis of Tibia Loading in Children with Myelomeningocele

​

​

Motivation Summary

​

Bone adapts to mechanical loading which is induced by joint loads and muscle forces. The purpose of this study is to evaluate tibia loading using joint reaction analysis for later correlation with bone properties in children with myelomeningocele and controls.

Workshop Goals

​

• Develop complete workflow to perform joint reaction analysis
• Perform analysis and validate results on a couple of subjects

Accomplishments

​

• Developed template setup files for all steps of analysis
• Successfully ran analysis on two control subjects and one patient
• Results similar to knee contact force and muscle activations/forces in the literature

¹ Children’s Hospital Los Angeles &

University of Southern California

​

Knee joint axial load for sample patient and control

Tishya Wren¹

Joint Reaction Analysis of Tibia Loading in Children with Myelomeningocele

Muscle activation from static optimization compared with normal data from Liu et al. 2008

Future Plans

​

• Add more scripting to simplifypreparation of setup files
• Add graphs to check key outputs
• Analyze more subjects and relate joint loads/muscle forces to bone properties

Knee axial load in control subject (left) compared with grand challenge in vivo and previous simulation data from Kinney et al. 2013 (below)

Model Validation

Maxime Bourgain, Samuel Hybois

Shoulder modeling for sports wheelchair movement

Motivation Summary

Musculoskeletal modeling is a powerful tool in order to get a better insight on the apparition of musculoskeletal disorders of the upper limbs, and to help prevent them. It is particularly the case for wheelchair locomotion and golf, during which upper limbs are highly stressed.

To improve basic models, our first step was to enable the motions of the clavicle and scapula by adding unconstrained degrees of freedom.

However, the lack of accurate physiological description of the shoulder kinematics with respect to the scapula limited the impact of the kinematic results that we obtained, and a fortiori of the muscular forces computed through static optimization.

​

Workshop Goals

We choose to focus on fullbody kinematic model development

• Symmetrize the ellipsoid joint made by Seth, in order to study non-symmetric movements such as golf swing
• Adding the lower limb
• Adding head and cervical movement
• Tackle muscle implementation problem

Accomplishments (IBHGC_fullbody_MBSH_CSRT.osim)

• Fullbody Ik development
• Shoulder x2 (symmetrization of Seth 2016 model)
• Lower limb
• Head and cervicals
• Scaling new model
• IK computing for golf and wheel chair movement
• Adding a golf club !

Institut de Biomécanique Humaine Georges Charpak, Arts et Métiers ParisTech, Paris (France)

​

​

Raabe model duplication

Seth shoulder model duplication

And Seth shoulder model symmetrization

Combinaison of cervical movement (to improve with Cazzola model)

Bilateral Scapulo-thoracic joint available on simtk

Maxime Bourgain, Samuel Hybois

​

Shoulder modeling for sports wheelchair movement

​

Adapt lumbar movement function with subject specific rotation measurements (Bourgain et al 2016)

Mass distribution and ellipsoid scaling to adapt from BodyScan acquisitions (Nerot et al 2016)

Subject specific lower limb geometries for golf swing movement (Bourgain et al 2016).

What’s next?

• Subject specific kinematics
• Lower limb

​

• Spine (geometries)

​

• Subject specific dynamics
• Mass distribution

​

• Spine behavior)

What was performed ?

​

• Muscle distribution

Seth 20??

Raabe 2016

GOLF

Wheelchair

Sauret et al in progress

Next step: continue patient specific modelling with EOS stereoradiographies

Error reduction of acromion kinematics compared to Holzbaur et al model (modified with mobile clavicles)

Use of our new model

Maxime Bourgain, Samuel Hybois

​

Shoulder modeling for sports wheelchair movement

​

Let’s now understand those skills

Marjolein van der Krogt¹, Lynn Bar-On²

Simulation of passive gastrocnemius muscle-tendon behavior in CP

Big Picture Motivation

The overall goal of this project is to develop a model that simulates ankle joint hyper-resistance in children with spastic cerebral palsy (SCP) and compare it to data from typically-developing children (TDC):

​

→ simulate contracture (slow movements, passive muscle properties)

→ simulate spasticity (fast movement, include enhanced reflexes)

​

Workshop Goals

1. Improve slow ankle movement optimizations
2. Update our reflex controller to latest version of OpenSim.
3. Add reflex controller to simulate fast ankle movements.

​

Workshop achievements

1. Added a ligament to our model, fitted its properties�Optimized slow ankle movement simulations
2. Built a working reflex controller with delay in OpenSim 3.3
3. Created fast stretch simulations with increased reflex activity

​

→ Achieved all three goals! :)

​

​

¹ VU University Medical Center, Amsterdam, The Netherlands.

² KU Leuven, Leuven, Belgium.

​

​

Marjolein van der Krogt¹, Lynn Bar-On²

Simulation of passive gastrocnemius muscle-tendon behavior in CP

GOAL 1: Optimized slow ankle movements

​

Poor fit at start of motion.

Incorrect passive muscle forces

​

Improved fit with addition of anterior ligament

​

GOAL 2:

Built reflex

controller plugin�for OpenSim 3.3

GOAL 3: Added reflex controller to simulate fast ankle movements

​

← Forward dynamic simulation on top of measured ankle motion.

Optimized gastroc and tib ant properties and an anterior ligament

​

​

Martina Barzan¹

Subject-specific kinematic model of the knee with deformable ligaments

Motivation Summary

Subject-specific models of the knee can help assess complex and subtle phenomena involved in the tibiofemoral joint (TFJ) and patellofemoral joint motion, such as patellar maltracking. We previously implemented a subject-specific passive kinematic model of the TFJ and patellofemoral joint, based on MRI, and we implicitly included it in a full lower limb OpenSim model via splines (model A). The next step is to explicitly implement a 6-link parallel mechanism of the TFJ in a subject specific full lower limb OpenSim model and allow for minimal deformation of four ligaments.

​

​

Workshop Goals

• Explicitly implement a 5-rigid-link TFJ mechanism in a full lower limb skeletal model (model B);
• Verify if model B solves the kinematics in the same way as model A;
• Add an extra link (ligament) to the TFJ mechanism (model C);
• Minimize ligament elongations for model C. Validate results against experimental TFJ kinematics.

​

​

¹ School of Allied Health Sciences and Menzies Health Institute Queensland, Griffith University, Australia

​

​

Accomplishments

• Implemented a 5-rigid-link TFJ mechanism in a full lower limb subject-specific skeletal model (model B);
• Compared the TFJ kinematics during gait between model A and model B;
• Added an extra link (ligament) to the TFJ mechanism (model C);
• Refined knowledge on constraints.

​

Model A

Model B

TFJ mechanism

Martina Barzan¹

Subject-specific kinematic model of the knee with deformable ligaments

​

Model A

TFJ angles and displacements during gait solved by Inverse Kinematics for Model A and Model B

​

• Ab/adduction and int/ext rotation exhibit the major differences between the two models;
• Ab/adduction excursion for model B is higher than in vivo measurements (Benoit et al., 2007).

Model B

Michail Mamalakis, Dimitar Stanev, Kostantinos Moustakas

Multiscale simulation of the knee complex with OpenSim and FEBio

​

Motivation Summary

Investigate a complex movement (such as walking, running, e.t.c.) through OpenSim pipeline and provide the initial conditions for a more detailed model that isolates the knee.

Workshop Goals

• Extract the tibia and femur body kinematics from OpenSim and

properly scale and translate them to FEBio

• Extract the forces on the tibia and femur from OpenSim and translate to FEBio
• Perform an open loop simulation, for given initial conditions from OpenSim, and estimate the knee ligament and cartilage stress

Accomplishments

In our project one of the mainly problems which we deal with is how will someone apply the forces and kinematics which applied in the bodies from the OpenSim in FEBio.

• So for the prescribed motion which we have to determine in the FEBio, we will apply the results from a body-kinematics analysis from the OpenSim.
• For the loads which act in the two bodies of FEBio we will used the vectors from the getMultibodySystem().getRigidBodyForces of SimBody Dynamic state after translate them in another point, corresponding to the point in FEBio.

Michail Mamalakis, Dimitar Stanev, Konstantinos Moustakas

Multiscale simulation of the knee complex

with OpenSim and FEBio

We estimate the position and orientation of the bodies from OpenSim for a given movement. Then we will update the OpenKnee(s) accordingly, by first aligning the two models (find the corresponding transformations for a reference pose). We are considering porting the analytical geometric representation from FEBio to OpenSim.

In addition we will estimate the body wrenches, with respect to a point of interest. These wrenches will be applied to the analytical model.

We introduced a fixed rigid-body in the FEBio as a reference frame. The reason of this is that, now we can apply the motion transformations extracted from OpenSim on the femur and the tibia with respect of the fixed frame.

Motivation Summary

​

As for the shoulder-arm-area, I tried to set up and run muscle simulations for the lower-body as well, in order to correlate surface deformation and muscle data. However with a few obstacles...

​

​

​

Key Workshop Goals

​

• 1. Correct IK transfer to OpenSIM based on 3d data
• 2. Approximate Ground Reaction Forces from vertical forces and COP
• The pressure plate used during the recordings, only gives us a grid of vertical forces, Fx and Fz components of GRF needed!

Accomplishments

​

• Prerequirements settled: Both the full detail right leg mesh as well as the low detail full body mesh were prepared for OpenSIM
• Markers placed in 3D and exported
• Model Gait2354 scaled
• Inverse Kinematics

BioSurface modelling for the Lower Limb

Stefanie Gassel¹

¹ University of Applied Sciences (HTW) Dresden, Germany

​

Stefanie Gassel¹

BioSurface modelling for the Lower Limb

​

1. Key Goal: Correct Kinematics

• Simple landmark based Kinematics transfer based on landmarks accomplished
• Further ideas identified of making use of full 3d data
• Additional markers on functional joint centers
• Bone pose estimation methods like PCT (point cluster technique)
• Exact scaling and kinematic estimation based on 3d mesh and transfer of scaled model and coordinate curves to OpenSIM

2. Key Goal: GRF

• Lot more time and work needed! Pressure data separated to both feet and Fy and COP exported, however ID and SO are still failing
• Several Techniques identified for approximating Fx & Fz:
• Constraint-based Induced Acceleration Analysis
• Contact force models to simulate GRFs purely based on kinematics and contact geometry
• Deduce Forces from acceleration of COM and body mass

Future Plans involve improving the GRF approximation and the kinematics to receive ID and SO results matching the measured EMG data for validation, before surface and muscle data can be matched.

Expansion of a Neuromusculoskeletal Gait Model to OpenSim

Motivation Summary

We currently have a robust neuromusculoskeletal model with which we can study human walking. By expanding the control of this model to OpenSim, we hope to be able to use OpenSim models and tools to study gait more effectively.

Workshop Goals

• Assess whether modifying source code, creating a plugin, updating Simulink code, using OpenSim/Matlab scripting, or using any currently unknown method will allow for the best implementation of the control
• Use the identified method to develop a simple model with either the linked control or with an elementary redevelopment of the control that can run a forward dynamic simulation

Accomplishments

• Properly prepared a development environment with which an expansion of an OpenSim model can be properly created and executed
• Evaluated potential methods and identified two different ways through which the CPG control could be implemented

​

William Barker¹, Mukul Talaty^1,2

¹ The Pennsylvania State University, Abington College

² Gait and Motion Analysis Laboratory, MossRehab

​

William Barker, Mukul Talaty

​

Expansion of a Neuromusculoskeletal Gait Model to OpenSim

Two methods have been identified as good candidates to explore for the implementation of the CPG based control. The plugin method (left) uses a plugin to create a class of neurons with neural state variables, which comprise the neural oscillators in our control. This method uses C++ based code and allows for full use of OpenSim, but requires more development to implement. The Matlab scripting method uses the OS/Matlab link to work primarily from Matlab. This method would likely allow us to use our existing code for the control.

Plugin

Matlab Scripting

OpenSim

OpenSim

Plugin

Matlab

Neuron class w/ neural state variables

-Skeleton

-Muscles models

-Contact dynamics

-Physiological states

*Neuron class referenced from plugin

*Control and integration carried out in OpenSim

​

​

-Skeleton

-Muscles models

-Contact dynamics

-Physiological states

*State variables sent to Matlab

​

​

​

​

​

​

​

​

-Control equations

​

-Neural state variables

​

-Base CPG code

​

*Control and integration carried out in Matlab

​

​

Co-Simulation using OpenSim and ABAQUS for the Biomechanics of Reverse Shoulder Total Arthroplasty

​

Motivation Summary

​

The goal of the specific research project is to study the effects of acromion sizes on bone stresses in reverse shoulder total arthroplasty (RSTA). We use OpenSim/DSEM to examine the muscle and joint forces of RSTA shoulder. An FEA software ABAQUS will be applied to solve for the bone stresses.

​

Workshop Goals

• Use the DSEM shoulder as the baseline shoulder
• Learn the locations of the access points in OpenSim for information on the musculoskeletal structures
• Learn the methods available to export these data
• Convert to formats that are compliant with those of the ABAQUS input decks

​

Western Michigan University

​

William W. Liou and Yang Yang

Co-Simulation using OpenSim and ABAQUS for the Biomechanics of RSTA (Liou and Yang)

​

Accomplishments

• Learned the locations of the access points in OpenSim for information on the musculoskeletal structures
• Learned the methods available to export these data
• Converted to formats that are compliant with those of the ABAQUS input decks
• Applied Delt-scap_9 muscle force to ABAQUS
• Accomplished coupling of OpenSim and ABAQUS

​

OpenSim

ABAQUS

Stress

Displacement

Delt_scap_9 Muscle Force

Nithin Babu Rajendra Kurup¹

Dynamic Path Optimization for Handle Based Wheelchair Propulsion

Motivation Summary

​

The main issue faced for the dynamic optimization problem was to design a resistance for the propulsion mechanism and then to derive the necessary power generated from the path.

Workshop Goals

​

• Main goals was to find an alternative method to calculate the power due to propulsion.
• Implement the objective functions and the necessary penalty functions.
• Method to couple 2 coordinates to form the necessary shape of propulsion.

¹ Vienna University of Technology

​

Accomplishments

​

• Most of the works performed during the week involve coding for the dynamic optimization routine. Hence no graphs to display.
• Performing torque and muscle optimization simultaneously.
• Was able to implement resistive torque using torque actuator The aim is to use muscles to counteract this torque.
• The power produced from above can be calculated using the calcPower() function.
• Learned how to implement penalty functions as shown in figure1. to constrain the motion of the crank, such as final time constraint, angle constraint, angular velocity constraint.
• Learned how to implement the event handler to record continuous stream of data needed for my objective function.
• Used the coordinate coupler constraint to make one coordinate the function of the other via coding in c++.

Dynamic Path Optimization for Handle Based Wheelchair Propulsion

Nithin Babu Rajendra Kurup¹

Fig1. A section of the objective function

Megan Pottinger¹, Katherine Mavrommati¹, Greg Orekhov¹

EMG-Driven Elliptical Trainer Modeling

Motivation Summary

​

The goal of our lab is to find knee joint contact loads using EMG-driven OpenSim analysis on various types of exercise. We focus on experiments involving gait, cycling, and elliptical to recommend exercises to subjects prone to knee osteoarthritis.

Workshop Goals

​

• Develop a process for analyzing elliptical exercise and simplified this model by improving our cycling OpenSim analysis.
• Improve our workflow in OpenSim by establishing a better understanding of the tools available.

¹ California Polytechnic State University

​

Accomplishments

​

• We managed to run inverse kinematics with experimental cycling data. With this, we ran multiple RRA computations using different pelvic models: normal, ball and socket, and welded to the seat. We then calculated inverse dynamics with these results to compare the pelvic loading.
• Refined knowledge on actuator and tasks files required to run RRA and CMC tools. Successfully incorporated EMG data to run CMC for gait model.

Megan Pottinger¹, Katherine Mavrommati¹, Greg Orekhov¹

EMG-Driven Elliptical Trainer Modeling

​

These figures are the results from inverse dynamics when the pelvis is modeled in three different ways: allowed to move freely, not translate, and not translate or rotate. These results will allow us to compare the pelvic loads to seat and handlebar loads to determine the best method for modeling the pelvis during cycling and then validate our loading results using published data¹.

¹Stone, C., & Hull, M. (1995). The effect of rider weight on rider-induced loads during common cycling situations. Journal of Biomechanics, 28(4), 365-375.

Megan Pottinger¹, Katherine Mavrommati¹, Greg Orekhov¹

EMG-Driven Elliptical Trainer Modeling