Related Experiment Videos
Application of high-performance computing to numerical simulation of human movement
F C Anderson1, J M Ziegler, M G Pandy
1Department of Kinesiology, University of Texas at Austin 78712, USA.
Journal of Biomechanical Engineering
|February 1, 1995
Summary
Supercomputers like the Cray and Intel significantly reduce computation time for human movement optimization problems. A hybrid architecture combining parallel and vector processing offers the most efficient solution for complex biomechanical simulations.
Area of Science:
- Biomechanics
- Computational Science
- Supercomputing
Background:
- Solving large-scale optimization problems in human movement is computationally intensive.
- Previous methods on conventional serial machines required extensive processing time.
Purpose of the Study:
- To evaluate the feasibility of using massively-parallel and vector-processing supercomputers for human movement optimization.
- To compare the computational expense of different supercomputer architectures for gait analysis.
Main Methods:
- Modeled the human body as a 14 degree-of-freedom linkage with 46 musculotendinous units.
- Compared computational time on a serial machine (SGI Iris), a MIMD parallel machine (Intel iPSC/860), and a parallel-vector machine (Cray Y-MP).
- Calculated optimal controls for the single support phase of gait.
Main Results:
- Gait optimization on the SGI Iris could take up to 3 months of CPU time.
- The Cray Y-MP reduced computation to approximately 77 CPU hours.
- The Intel iPSC/860 reduced computation to approximately 88 CPU hours.
- Intel excelled at derivative computations; Cray excelled at parameter optimization.
Conclusions:
- Massively-parallel and parallel-vector processing supercomputers can solve large-scale human movement optimization problems efficiently.
- A hybrid system integrating vector processing with MIMD parallel architecture is ideal for complex optimal control problems.