[1B4] Identification and characterisation of defects from strain gauge measurements using simulation-driven machine learning techniques
A Ballisat, W Alden and M Freeman
Centre For Modelling and Simulation (CFMS), UK
Strain gauges are a common structural health monitoring tool used across many industries to monitor the status of an asset. Within aerospace structural testing, strain gauges are a crucial part of test protection, aiding operators with early identification and characterisation of defects and other problems before critical damage occurs. A key issue is a lack of knowledge of the implications of strain gauge measurements and whether they are significant. In this work, we apply a combination of neural networks and statistical methods to train a model to identify and characterise defects in a representative static load test, providing operators with earlier warning and more detail of defects. Due to the lack of experimental data, approximately 50,000 realisations of the test case were simulated, accounting for expected real-world variability in geometry, material properties, loads, defects and other process parameters. This was achieved through parallel evaluation of a transient finite element structural model of the test scenario. The model achieved over 90% accuracy for identifying and classifying defects. This approach significantly increases the information that can be derived from strain gauge monitoring and as it is agnostic of this specific test case can easily be applied to other trials.
