Unsupervised learning-based hierarchical diagnosis of analog circuits

Piotr Bilski

Abstract

The paper presents the automated diagnostic method of analog circuits combining supervised and unsupervised learning. The purpose of this sophisticated approach is to effectively distinguish easily identified states of the analyzed system from the more difficult ones. The latter are often related with existence of ambiguity groups, which create problems for the distinction of specific states. The proposed approach uses subsequently two algorithms. The first one extracts and learns to classify such “simple” training examples. The second one aims at the classification of more difficult ones. For both stages, Self-Organizing Maps and Random Forest were used, respectively. The scheme was tested on the model of the 3rd order Bessel highpass filter, confirming effectiveness of the approach.

Download

IMEKO-TC10-2017-016.pdf

DOI

-

IMEKO TC

TC10 - Measurement for Diagnostics, Optimization and Control

Back to the proceedings