Fault Diagnosis Using Kernel Linear Independence Test in Reproducing Kernel Hilbert Space

Fault detection and identification (FDI) is essential for ensuring the safety of mechanical systems. Typically, FDI algorithms begin by establishing fault detection methods and subsequently identifying the faults. The fault detection process aims to detect abnormalities within the system. To initiate the fault detection process, a mathematical model is required for comparison. However, this mathematical model may not be fully known in advance. In this study, we propose the implementation of Gaussian process regression (GPR) to learn the current nominal model using sensor datasets. The learned model is then compared with newly acquired data to detect any occurrence of faults. GPR learns the model using a nonparametric Bayesian kernel defined in the reproducing kernel Hilbert space. Consequently, the kernel linear independence (KLI) test is employed to examine the new data set for fault detection. The simulation environment incorporates the NASA-developed generic fixed-wing aircraft transport model (GTM), specifically considering cases involving severed left-wing configurations for simulation purposes.