Survey of Recent Developments in Rotorcraft Design Optimization

摘要

From the literature reviewed in this paper, the following conclusions can be drawn about the nature of the helicopter optimization problem. These problems involve 1) computationally expensive and unwieldy aeroelastic or finite element analysis, 2) presence of local minima, and 3) presence of numerical noise and/or convergence difficulties. The increasing power of computers decreases the problem of expensive aeroelastic and finite element analysis. However, the search for better modeling is resulting in the use of CFD for aerodynamic predictions and full finite element method formulations for the composite rotor blade structural analysis. The use of computer power to advance modeling and analysis of systems always appears to precede the use in optimization. Therefore, it is unlikely that the rotor aeroelastic problem will be computationally cheap for quite some time. Whenever gradient-based methods are used for optimization, a large portion of the computer time is taken up by gradient calculation. Finite difference-based derivatives are prohibitively expensive. In addition, finite difference derivatives need a good estimation of step size. In some case, forward difference derivatives may not be sufficiently accurate leading to the requirement of central difference derivatives and the resultant increase in computer time.83 Whereas the use of analytical or semi-analytical sensitivity derivatives can alleviate the computer time requirements to a great extent, they need major and time-consuming changes to be made to computer programs. Whereas this may be possible in an academic setting, it is almost impossible in an industrial setting, where proprietary computer codes are guarded carefully. Furthermore, any changes to the computer code or inclusion of new design variables or objective functions and constraints require a rederivation of the analytical derivatives and subsequent recoding into the computer program. The analysis and sensitivity calculations will deviate with the passage of time unless proper code management is done. Finally, the time saved in terms of computer expense by analytical derivatives should be weighed with the cost of labor involved on the part of skilled domain experts.