This paper demonstrates fault diagnosis on unmanned underwater vehicles (UUV) based on analysis of structure of the nonlinear dynamics. Residuals are generated using different approaches in structural analysis followed by statistical change detection. Hypothesis testing thresholds are made signal based to cope with non-ideal properties seen in real data. Detection of both sensor and thruster failures are demonstrated. Isolation is performed using the residual signature of detected faults and the change detection algorithm is used to assess severity of faults by estimating their magnitude. Numerical simulations and sea trial data show results with very favorable balance between detection and false alarm probabilities. ; (c) IFAC 2014. This is the authors' accepted and refereed manuscript to the article.