HPCC Group / Project Overview
Scalable and Dependable Applications and Infrastructure for
High-Performance Computing and Networking
A Computational Framework for Simulating
National Institutes of Health
This research project is an interdisciplinary work involving biomechanics and
high-performance computing. The HPC group focuses on developing efficient,
scalable and reliable computational framework for simulating human joint
mechanics, and Computational
Biomechanics lab at the Mechanical and
Aerospace Engineering Department, University of Florida, emphasizes on
developing computer human joint model for simulating joint mechanics, especially
Specific objectives of the project include:
(1) Create a dynamic musculoskeletal model with deformable knee joint contact.
Deformable contact in the knee will be studied initially since the knee is the
most commonly injured joint.
(2) Incorporate this model into a parallel-processing optimization framework.
Parallel processing will used to reduce the computational time for predictive
optimizations from weeks to a matter of hours.
(3) Evaluate the model's ability to predict experimental movement data.
Pre-existing experimental movement data will be used to evaluate the model's
ability to predict motion and ultimately joint contact stresses.
The resulting functional virtual human model can then be used for basic research
and clinical applications.
Simulation of joint mechanics (see Computational
Biomechanics lab for more details)
Mechanical loading (contact stresses) is believed to play a major in
degenerative joint diseases, so the knowledge of in vivo joint motion and
loading during functional activities could help address this clinically
significant issue. Since non-invasive experimental approaches do not exist
for measuring in vivo joint loading, computer simulations have been used to
develop predictions given estimates of the muscle forces acting on the joint.
However, current rigid body and deformable modeling approaches are not able to
calculate accurate contact stress results during movement in critical joints
such as the knee. A logical solution to this problem is to incorporate
deformable joint models into a larger rigid body dynamic model.
Since the dynamic optimization problem solved via simulation has very high
computational complexity, high-performance computing is key to resolve the
limitations. Parallel optimization algorithms, which parallelize each function
evaluation to processors in parallel computers are potential solutions to
simulation problems. So far, parallel optimization algorithm research have
been done using gradient or non-gradient based optimization algorithms. Owing to
their fast convergence characteristics, gradient based methods have been used
typically in these kinds of simulation even though the methods require
time-consuming parameter sensitivity studies as the presence of numerical noise
will often cause premature convergence. Non-gradient based methods converge to a
global minimum but at the cost of slow convergence.
However, when the simulation itself has a small number of design variables,
gradient based parallel optimization algorithm is not efficient to use.
For example, parallelization of gradient calculation in Quasi-Newton method can
only use the same number of processors as the number of design variables.
This limitation affects scalability and the method is not efficient if resources
are large. Moreover, non-parallelized line searches in Quasi-Newton method
decrease the speedup and parallel efficiency. To overcome such limitation,
another layer of parallelization at the analysis function level itself can be
Current work focuses on the developing parallel gradient based optimization (BFGS)
algorithm using commercial optimization tool, and developing finer parallel
algorithms in single function evaluation. The hybrid approach can increase the
scalability and overcome the limitations in speedup. Additionally,
adapting fault tolerance concepts in this simulation can provide efficient,
scalable, and reliable computational framework so that the result of simulating
human joint mechanics can be valuable for clinical issues on joint diseases.
OTHER HPCC Group PROJECTS
Reconfigurable Computing (RC) Hardware Empowered Grid Computing
Parallel and Distributed Computing for Fault-tolerant Sonar Arrays
High-Performance Networking Related Projects