Global sensitivity analysis plays an important role in robustness optimization design of aero-engine fuel gear pumps to investigate how input parameters uncertainties contribute to performances uncertainties. In this paper, inspired by the Leave-One-Out method, an advanced Polynomial Chaos Expansion (PCE) method equipped with Adaptive design of both training Points and polynomial Order, denoted as APO-PCE, is proposed to efficiently estimate the variance-based sensitivity indices. A novel active learning strategy is developed for identifying the optimal candidate sample point to adaptively design training points and simultaneously update the polynomial order. Input parameters, including rotational speed, inlet pressure, outlet pressure, and gear tip clearance, are reasonably characterized as nine convenient sampling Gaussian variables by probabilistic modeling and Karhuben–Loève expansion. The results show that the proposed APO-PCE method is superior to the classical PCE method, and the uncertainties of rotational speed, inlet pressure, and outlet pressure are the main factors for the uncertainties of supply performances and flow field characteristics, while the gear tip clearance uncertainty only affects the flow loss at lower rotational speed.