Sequential Design of Experiments with Physically Based Models 23

Abstract

The application of physically based models to sequential optimization was explored and the benefits measured by comparison to optimizations performed using control variable polynomial models. The optimization algorithm developed is based on the sequential use of local (weighted) linear regression models. A new operating point or design is recommended at the optimum of the model within the region where the model is considered valid. This region, within which extrapolations based on the model are believed to be sufficiently accurate, is defined by constraints based on the estimated predictive error of the model and the distance from the data. A new model is created after each data point is collected. As a test case, the sequential optimizer was applied to the design of a paper helicopter for maximum time of flight. The physically based model of the paper helicopter was developed through the use of dimensional analysis, a technique which groups variables according to their dimensions. It was found that the physically based model improved the design of the helicopter more rapidly than the polynomial models. For example, in one comparison of the physical model and the linear model, the physical model reached the optimum design after four sequential designs, while the linear model hadn't reached the optimum after ten designs.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
May 01, 1989
Accession Number
ADA211918

Entities

People

  • Michele E. Storm

Organizations

  • Massachusetts Institute of Technology

Tags

Communities of Interest

  • Air Platforms
  • Materials and Manufacturing Processes

DTIC Thesaurus Topics

  • Algorithms
  • Computer Programming
  • Computer Programs
  • Computers
  • Confidence Limits
  • Covariance
  • Data Science
  • Data Sets
  • Engineering
  • Errors
  • Experimental Design
  • Information Science
  • Manufacturing
  • Mechanical Engineering
  • Random Variables
  • Standards
  • Statistical Analysis

Readers

  • Adaptive Control and Estimation with Uncertainty in Dynamic Systems.
  • Computational Modeling and Simulation