Fault classification SOM and PCA for inertial sensor drift

FDI is an active research field in several areas. In fact, there are still many challenges in on-line detection and identification. Several approaches have been pursued such as model-based or knowledge-based techniques, however, these present several drawbacks like time consumption or the lack of adaptability. Here a proposal to classify faults for both known and unknown scenarios is presented. This is based upon a statistical approach, principal component analysis (PCA), and non-supervised neural networks such as self organizing maps (SOM). Experimental results are presented based upon an aircraft flight dynamics model.