Methodologies for tracking of load extremes and error estimatin using probabilistic techniques

This work, conducted at CIMNE under ALEF project task 1.2.3, presents an investigation about the potential capabilities of neural networks to assist simulation campaigns. The discrete gust response of an aircraft has been chosen as a typical problem in which the determination of the critical loads requires exploring a large parameter space. A very simple model has been used to compute the aerodynamic loads. This allows creating a large database while at the same time retaining some of the fundamental properties of the problem. Using this comprehensive dataset the effects of network structure, training method and sampling strategy on the level of approximation over the complete domain have been investigated. The capabilities of the neural network to predict the peak load as well as the critical values of the design parameters have also been assessed. The applicability of neural networks to the combination of multi-fidelity results is also explored.