Date of Award

Spring 2011

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Electrical and Computer Engineering

First Advisor

Riedel, Susan A.

Second Advisor

Harris, Gerald F.

Third Advisor

Brown, Ronal H.

Abstract

The Orthopaedic and Rehabilitation Engineering Center (O.R.E.C) has been studying two conditions which can wreak havoc on the human body's balance control system. The conditions being studied are cerebral palsy (CP) and adolescent idiopathic scoliosis (AIS)[1, 2, 3, 4, 5, 6]. A greater understanding of postural deficits in children with these conditions and the causes for these conditions may help researchers achieve better interventions and treatments [3]. A tool used to characterize postural stability is a bi-planar postural stability model. This model is designed where 10 parameter values have to be adjusted until the simulated data is statistically the same as the actual patient data. This process becomes a minimization problem in a 10-dimensional space.A previous method for input parameter value adjustment involves an operator manually adjusting each parameter value. This process is very time consuming and requires approximately 4 uninterrupted days for computation and analysis. An improved automatic method brought the computation and analysis time down to a reasonable 12 hours [1]. To further improve the input process for parameter values new search methods were explored, including evolutionary algorithms. A evolutionary algorithm was integrated with the existing bi-planar postural stability model [3, 4, 5, 6, 1, 2] to produce more accurate results more efficiently by exploring the full ten-dimensional search space using a previously-designed cost function [1, 2]. Simulation time was drastically improved after streamlining the existing code provided by Andrew Sovol [1, 2]. After making improvements, a single iteration ran 13.21 times faster on average. The average time for a simulation was reduced to 6005.74 seconds (approximately 1 hour and 40 minutes) from 12 hours. Statistical mismatches were also computed with a smaller p-value reducing the possible error from 10% to 0.5%.

COinS