Research Sophie ten Zeldam - XAI methodology

In the aircraft industry defects and irregularities of components or systems are reported on a repair card. These cards are often filled in incomplete or inconsistent. This can result in additional maintenance cost or an increase in downtime. This research tried to solve this problem by combining maintenance and usage data with AI (Artificial Intelligence). The model was able to determine the right failure mode for 81% of the cases.
Since most technicians won’t accept only an outcome, the model also had to explain its result. Therefore an XAI (eXplainable Artificial Intelligence) methodology is invented which explains a decision based on the measure of similarity to a certain class. This is expressed with a numerical number and with visual representation.
With this model the inconsistency and incompleteness can be decreased which results in better data analysis and more efficient troubleshooting. Besides, it will also be useful for a repair of a NFF (No Failure Found) after premature uninstallation after just-in-time maintenance.