On-the-fly neural network construction for repairing F-16 flight control panel using thermal imaging
Abstract
When the card-level tester for the F-16 flight control panel (FLCP) had been dysfunctional for over 18 months, infrared thermography was investigated as an alternative for diagnosing and repairing the 7 cards in the FLCP box. Using thermal imaging alone, over 20 FLCP boxes were made serviceable, effectively bringing the FLCP out of awaiting parts (AWP) status. Through the incorporation of a novel on-the-fly neural network paradigm, the neural radiant energy detection system (NREDS) now has the capability to make correct fault classification from a large history of repair data. By surveying the historical data, the network makes assessments about relevant repair actions and probable component malfunctions. On one of the circuit cards, a repair accuracy of 11 out of 12 was achieved during the first repair attempt. By operating on the raw repair data and doing the network calculations on the fly, the network becomes virtual, thus eliminating the need to retain intermediate calculations in trained network files. Erroneous classifications are correctable via a text editor. Erroneous training of neural networks has been a chronic problem with prior implementations. In view of the current environment of downsizing, the likelihood of obtaining functionality at the card-level tester is remote. Success of the imager points to corresponding inadequacies of the automatic test equipment (ATE) to detect certain kinds of failure. In particular, we were informed that one particular relay had never been ordered in the life of the F-16 system, whereas some cards became functional when the relay was the sole component replaced.
- Publication:
-
Thermosense XVIII: An International Conference on Thermal Sensing and Imaging Diagnostic Applications
- Pub Date:
- March 1996
- DOI:
- 10.1117/12.235386
- Bibcode:
- 1996SPIE.2766..284A