Surrogate Modeling: Artificial Neural Networks
THE COOPER UNION FOR THE ADVANCEMENT OF
SCIENCE AND ART
THE ALBERT NERKEN SCHOOL OF ENGINEERING
DEPARTMENT OF MECHANICAL ENGINEERING
AN OPTIMIZATION APPROACH TO THE DESIGN OF A FORMULA-STYLE RACECAR THROUGH ARTIFICIAL NEURAL NETWORK SURROGATE MODELS
BY
VITO DIIENNA
A THESIS SUBMITTED IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF ENGINEERING
APRIL 21, 2006
PROFESSOR GEORGE J. DELAGRAMMATIKAS
THESIS ADVISOR
Simulation-based optimization is the direction being taken by modern automotive engineering design. Simulations reduce the need to build physical prototypes in large numbers, which had traditionally been the source of great time, material, and cost expense for large corporations. This thesis presents a design methodology that links a high-fidelity vehicle simulation package, called CarSim, within a numerical optimization routine in order to automate the vehicle design process. The particular problem addressed was that of minimizing the track time of a Formula-style racecar, subject to fuel consumption, acceleratory, and tire slip angle constraints. The variables examined were transmission upshift schedule, suspension spring rate, and final gear ratio.
CarSim utilizes the fundamental equations of vehicle dynamics to calculate vehicle performance. However, a built-in and systematic optimization scheme does not currently exist within the CarSim framework. To address this deficiency, an extensive set of data was produced with CarSim. These data were then used to train Artificial Neural Network models for the vehicle's track time, fuel consumption per lap, minimum and maximum lateral acceleration, and minimum and maximum tire slip angles.