Kernel Deep Principal Component Analysis Based Fault Detection of Autonomous Underwater Vehicle Actuators

This paper addresses the fault detection issue of Autonomous Underwater Vehicle (AUV) actuators based on the kernel deep principal component analysis (KDPCA) method, which effectively overcomes the limitations of traditional Principal Component Analysis (PCA) in handling both nonlinear data and the extraction of longitudinal data features. Firstly, the data are transformed into a linear form by introducing a predesigned Gaussian kernel function. Secondly, Deep Principal Component Analysis (DPCA) is employed for hierarchical decomposition to extract potential fault information, thereby significantly enhancing the sensitivity and reliability of fault detection. Subsequently, a multi-level fault detection framework has been constructed by integrating the Hotelling $T^{2}$ statistics and the squared prediction error (SPE) statistics. Finally, the simulation results highlight that the proposed method not only accurately detects faults in AUV actuators, but also exhibits superior fault detection capabilities when compared to conventional multivariate statistical analysis-based methods.