[abstract] AN OBJECT-ORIENTED SYSTEM FOR OPTIMIZING PROBABALISTIC DECOMPRESSION MODELS TO EMPIRICAL DECOMPRESSION DATA.

Rubicon Research Repository/Manakin Repository

[abstract] AN OBJECT-ORIENTED SYSTEM FOR OPTIMIZING PROBABALISTIC DECOMPRESSION MODELS TO EMPIRICAL DECOMPRESSION DATA.

Show full item record


Title: [abstract] AN OBJECT-ORIENTED SYSTEM FOR OPTIMIZING PROBABALISTIC DECOMPRESSION MODELS TO EMPIRICAL DECOMPRESSION DATA.
Author: Howle, L; Lebental, S; Vann, RD
Abstract: BACKGROUND: The hypotheses of theoretical decompression models can be tested by fitting them to empirical diving data according to the method of maximum likelihood introduced by the U.S. Navy. Maximum likelihoods of competing models may be compared (formally or informally according to the nature of the models) to determine which provides better conformance between theory and observation (1-3). Few such systems are currently in operation. We present a new system and discuss its applications to previous work MATERIALS AND METHODS: The system code was written in C#.NET (2003) with class libraries handling the hierarchical data structure found in decompression data (4). The system is extensible, general-purpose, and applicable to a range of probabilistic models. The optimization engine for likelihood maximization contains elements of gradient descent, genetic algorithms, and continual starting point injection. The multivariate non-normality of likelihood maximization of common probabilistic models is specifically addressed. We are reproducing previous work (2) to validate system performance RESULTS: We were able to find exact solutions of singular supersaturation-based risk functions with high gradients, and this substantially improved computational speed and accuracy of parameter estimation over numerical solutions. For delayed risk functions, we were also able to derive some exact risk function solutions but must use numerical estimation for certain roots. Agreement of our system with earlier decompression models was excellent CONCLUSIONS: We have developed a general-purpose, extensible set of object-oriented class libraries for manipulation of experimental decompression data. Our optimization methods were designed to address the non-normality of multivariate probabilistic DCS models. Our goal is a system for developing decompression models by testing them against experimental data. References: 1. Weathersby. JAP 72:1541. 1992 2. Thalmann UHM 24:255. 1997 3. Gerth. UHM 24:275. 1997 4. Temple. NMRC 99-02. 1999
Description: Undersea and Hyperbaric Medical Society, Inc. (http://www.uhms.org )
URI: http://archive.rubicon-foundation.org/3756
Date: 2006

Files in this item

Files Size Format View

There are no files associated with this item.

This item appears in the following Collection(s)

  • UHMS Meeting Abstracts
    This is a collection of the published abstracts from the Undersea and Hyperbaric Medical Society (UHMS) annual meetings.

Show full item record

Browse

My Account