One concern in the optimization of realistic complex systems is to ensure that the optimal model response is reliable and robust with respect to the inherent uncertainties associated with the design variables, constraints and the objective function. Traditional optimization techniques together with uncertainty analysis are computationally expensive and time consuming when it comes to identify what drives the response variability. A stochastic optimization framework combining stochastic surrogate model representation and optimization algorithm is proposed. A generalized Polynomial Chaos stochastic representation is used as the surrogate model. This representation is obtained from a high-order stochastic collocation method. The surrogate model can then be put to use within an appropriate optimization algorithm and provides fast approximations at new design points. This approach allows both sensitivity and optimization analysis. The stochastic optimization method is applied to a multi-layer reacting flow device. The geometric configuration is assumed to be uncertain. The structure design is optimized to maximize the energy transfer between the reacting flow and the device moving parts. Gradient descent and simulated annealing optimization techniques are successfully tested on the kinetic energy surrogate model of the device inner parts.