How fraud detection technologies can help to detect damages in aircraft
structures
Abstract
A software architecture based on Machine Learning (ML) and Finite
Element Method (FEM) and aimed at improving the detection of damages in
aircraft structure subjected to complex variable loadings is presented
here. Firstly, the software relies on statistical tools used among
others in fraud detection (One-Class Support Vector Machine, Local
Outlier Factors, Isolation Forest, DBSCAN) to identify anomalies in a
vast amount of data recorded over time by multiple strain gauges located
on the structure of the aircraft. Once an anomaly is detected at a given
time and for a specific set of strain gauges, it can be classified as
insignificant or critical by the user. If the anomaly is critical, the
data of the associated strain gauges can be used as input data for a FEM
optimization. This static optimization allows to visually assess the
position and geometry of possible cracks in the structure.