Prediction of Equipment Failure by Acoustical Signature Analysis. Phase II.

Abstract

Vibration data were gathered from accelerometers mounted on 8V71T diesel engines at Letterkenny Army Depot and US Army Aberdeen Proving Ground. A pattern recognition computer program was tested to see if it could distinguish between the vibration signature of good engines and the vibration signature of faulty engines. Vibration signatures from 24 engines (12 good, 12 faulty) were used to 'train' the computer program. The classifier routine produced by the computer program was then tested on the 24 'training' signatures and 14 'mystery' signatures (9 good, 5 faulty) that it had not seen before. Classification was accomplished with 100% accuracy on the 'training' data and 79% accuracy on the 'mystery' data. Further study indicated, however, that the accuracy of the classifier was not significant when using a training sample this small.

Document Details

Document Type
Technical Report
Publication Date
Jul 01, 1978
Accession Number
ADB030310

Entities

People

  • W. Scott Walton

Tags

DTIC Thesaurus Topics

  • Accuracy
  • Computer Programs
  • Computers
  • Diesel Engines
  • Engines
  • Machine Learning
  • Pattern Recognition
  • Recognition
  • Training
  • Vibration

Readers

  • Acoustics.
  • Applied Combinatorial Optimization and Logic Circuit Design.
  • Neural Network Machine Learning.

Technology Areas

  • AI & ML
  • AI & ML - Neural Networks