|
|
Computers and AI techniques such as artificial neural networks and expert systems are being applied to the fault diagnosis of electronic circuits in collaboration with the Universities of Hull and Bordeaux. Interest so far has been in the diagnosis of analogue circuits. The supply current and output voltage responses to test signals have been recorded and simulated. These responses depend upon the faults present. The pattern recognition and regression properties of artificial neural networks have been used to identify and quantify the faults from these response data. Work is being extended to mixed signal (analogue and digital) circuits, starting with a delta-sigma modulator and a digital-to-analogue converter. I am a member of the NORMATE group of researchers in mixed signal test. Artificial neural networks used include the multilayer perceptron, Kohonen Self-Organising Map, and the Simplified Fuzzy ARTMAP.
Results from the research include:
(i) 100% accurate fault diagnosis for single gate oxide short faults in a CMOS operational amplifier, and in a sigma-delta modulator using only the available pins,
(ii) the diagnosis of multi-faults in a bipolar integrated differential amplifier using only the available pins.
Dr. B.W. Jervis, b.w.jervis@shu.ac.uk
Back To: