A NEW METHOD FOR DIAGNOSIS OF ANALOG PARTS IN ELECTRONIC EMBEDDED SYSTEMS WITH TWO-CENTER RADIAL BASIS FUNCTION NEURAL NETWORKS

Zbigniew Czaja, Michał Kowalewski

Abstract

An approach of self-testing of analog parts in mixed-signal embedded systems controlled by microcontrollers is presented. It bases on a dictionary method of detection and localization of single soft faults of passive components in analog electronic circuits. At the Measurement Stage the tested analog part is stimulated by a square impulse generated by the microcontroller, and its response is sampled by microcontroller’s ADC in moments exactly established by an internal timer. Fault detection is made by the two-layer feed-forward perceptron. If the analog part is classified as faulty, the fault localization is performed by the specialized neural network based on Two-Center Radial Basis Functions (TCRBF). Both networks are created and trained on the Before Test Stage on the basis of family of localization belts, which are created by dispersion of identification curves under assumption of components tolerances. Identification curves placed in measurement space illustrate behaviour of the tested analog circuit incoming from changes of values of its components. The proposed method with TCRBF classifier is dedicated for testing and localization of single soft faults in analog electronic circuits. It can also be simply adopted for rough identification of analog circuits’ components.

Download

IMEKO-TC4-2008-043.pdf

DOI

-

IMEKO TC

TC4 - Measurement of Electrical Quantities

Back to the proceedings