A "Learning from Models" Cognitive Fault Diagnosis System

We present an unsupervised cognitive fault diagnosis framework for nonlinear dynamic systems working in the space of approximating models. The diagnosis system detects and classifies faults by relying on a fault dictionary that is empty at the beginning of the system's life and is automatically populated as faults occur. Outliers are treated as separate instances until enough confidence is built and either are integrated in existing classes or promoted to a new faults class. Simulation results show the effectiveness of the proposed approach.