As a rigorous and efficient uncertainty propagation method,the polynomial chaos (PC) approach has been widely employed in robust design optimization.For the gradient-based robust optimization,the local sensitivity or gradient of the objective function consisted of the first two statistical moments of the system performance should be calculated during each optimization iteration.Generally,the classic finite difference technique is employed to obtain the local sensitivity by repeatedly calling the system performance function,which may be very time-consuming especially for some engineering problems involving computationally expensive analysis model,such as the computational fluid dynamics.