The Diagnosis and Prognosis of Analogue and Digital Electronics using Biologically Inspired Neural Networks and Artificial Intelligence

摘要

This paper presents a study into the effectiveness of a Neural Network (NN) in learning analogue and digital electronic systems to enable diagnosis and prognosis of health metrics, throughout a product lifecycle. Given the inherent difficulty in the repeatability of manufacturing variables such as component resistance or solder barrel fill, every manufactured electronic system may be deemed individual, despite being of a common architecture; much like humans. Throughout their lifecycle these systems will be prone to unique fatigue and functional degradation and thus prone to bespoke failure modes, which cannot be accounted for in a generic Failure Mode, Effects and Criticality Analysis (FMECA) during system development and qualification. In this concept study, generic system architecture, nominal functioning, training data was obtained from a sub-assembly functional test by recording the electronic characteristics from pre-determined internal interfaces. Then the NN was provided with partial function training data i.e. a single failed signal, obtained from forced failures during a second functional test. This subsequently created an, infant NN of limited system knowledge. One may postulate that this infant NN could then be injected into subsequent sub-assemblies of common design so as to learn their idiosyncrasies and become a mature NN, now capable of component level diagnostics and prognostics, without the expense of bespoke test equipment and routine test schedules.