This article describes the application of neural networks for the design optimization of a curved wind turbine blade using an aero-elastic simulator with synthetic inflow turbulence. A vortex particle method where the wind turbine blades are represented by lifting-line theory is used, while the wind turbine structural dynamics are modeled using a finite-element multi-body based approach. A neural network together with a gradient-based optimizer allows to quickly design a new curved wind turbine blade in a complex aero-elastic wind-turbine simulation scenario. The blade design found from the neural network has increased pre-bend and sweep compared to the straight blade design. It produces approximately 1% more power on average with a slight increase of mean thrust on the rotor of 0.02% compared to the straight one. This study demonstrates that neural networks can be effective for designing wind turbine rotor blades involving complex aero-elastic simulation scenarios with turbulent inflow conditions. Further work may improve the performance of the neural network's predictive capabilities as well as the optimized design. (C) 2019 Elsevier Ltd. All rights reserved.