| description abstract | A hardware-in-the-loop system was created for the purpose of aerodynamic optimization. The system was used for fully independent optimization of an automotive diffuser model in a wind tunnel, with linear actuators used to modify the geometry. The model had three degrees of freedom with 7×105 possible configurations. High reliability of the system was demonstrated and its time-efficiency was quantified, with up to 400 configurations tested per hour. The system was tested with two implementations of a genetic algorithm, which demonstrated high consistency and resistance to the noise and hysteresis inherent in experimental data. Next, an attempt was made to minimize the main overheads of the optimization process, that is, the wind tunnel settling time and the measurement sampling time. In the original experiments, these overheads accounted for 35% of the total optimization time. A series of simulated optimization runs using a presampled experimental database were carried out, and it was found that the settling time could be eliminated, and the sampling time reduced to as little as 0.01 s, without a detrimental effect on convergence efficiency. Thus, the average duration of a function evaluation was reduced, and the overall efficiency of the optimization system was improved. The results were validated experimentally and the performance was quantified through hardware-in-the-loop, real-time optimization of a bluff body equipped with a diffuser. As many as 700 configurations per hour could be tested with the reduced overheads, leading to much faster convergence to the global optimum. | |
