Repetitively enhanced neural networks method for complex engineering design optimisation problems

Repetitively Enhanced Neural Networks (RENN) method is developed and presented for complex and implicit engineering design problems. Enhance neural networks module constructs an accurate surrogate models and ensures for avoiding over-fitting during neural networks training from supervised learning data. The optimizer is executed by the enhanced neural networks models to seek for a tentative optimum point. It is repetitively added into the supervised learning data set to refine surfaces till the RENN tolerance reaches. The RENN method demonstrates the effectiveness and feasibility for 2D highly non-linear numerical example and the structure design of two-member frame reaching convergent solution at 10 and 14 iterations respectively at the maximum error of 1% when compared with the exact solution. Then, the RENN method is applied for a long endurance unmanned aerial vehicle (UAV) airfoil design optimization. Class/Shape function transformation (CST) geometry parameterization method represents an accurate UAV airfoil with 10 geometry design variables. The high-fidelity analysis solvers with structured mesh for airfoil is used for UAV airfoil design problem. The total 88 experiment points are required to obtain an optimal UAV airfoil configuration after 13 RENN iterations and 75 initial experiments by Latin Hypercube method in reasonable turnaround time. The optimal UAV airfoil shows 10.8% in drag reduction in cruise condition and improvement in the maximum lift coefficient and stall angle of attack.