Fault Detection for Re-initialization of Online Gaussian Process Regression Using Kernel Linear Independence Test

Abstract

This study addresses methods for detection of faults in dynamic systems that can be represented as rigid bodies. We propose an online Gaussian process regression (GPR) re-initialization method for fault conditions, accomplished by detecting faults using a kernel linear independence test. The KLI test evaluates whether new input data shares the nominal dynamics represented by previous data points. Re-initialization of GPR is triggered by the KLI test results, enabling online GPR for real-time applications. We validated our method by simulating the generic transport model (GTM) of a fixed-wing aircraft, developed by NASA, focusing on scenarios with severed left-wing configurations.