TRAINING-ADJUSTED DECISION SYSTEMS FOR PROCESS CONTROL.

Abstract

Methods for the control of discrete product manufacturing processes are developed. Emphasis is placed on the end requirement for economical physical implementation of an adaptive controller with analog circuits. The proposed process controller is capable of making pass/reject decisions (classifications) at each of the manufacturing stations comprising the process for each discrete unit of manufacture. The process controller is capable of estimating control parameters at future manufacturing stations, or at final test. The minimum average risk classification operator is explored where the required conditional probability densities are multivariate symmetric. A search procedure is developed for determining the quadratic classification operator, which does not require exact prior knowledge of the (possibly different) functional forms of the conditional probability densities. The individual search algorithms are shown to converge for finite training sets. The derived value will converge in probability with increasing size of the sample set to the required value, which is shown to be median-unbiased. Implementation of the translation, rotation, and dilation algorithms with analog circuits is considered as an approach to realizing the slave classification operator.

Document Details

Document Type
Technical Report
Publication Date
Aug 16, 1965
Accession Number
AD0489330

Entities

People

  • Baxter F. Womack
  • Charles Loren Kettler

Organizations

  • University of Texas at Austin

Tags

Communities of Interest

  • Materials and Manufacturing Processes

DTIC Thesaurus Topics

  • Algorithms
  • Classification
  • Manufacturing
  • Probability
  • Rotation
  • Training
  • Translations

Readers

  • Regression Analysis.
  • Robotics and Automation.
  • Software Engineering