Estimation of Dynamic Parameters of Ground Vehicles based on Genetic Algorithms 2013-36-0636
In cars, optimizing their performance and improving their stability are of vital importance to manufacturers because of the highly competitive automotive market. These improvements could be achieved by developing analytical techniques for modeling the dynamic behavior of the car under certain conditions. They allow the vehicle to undergo various situations, which if were made with actual vehicles, would be expensive and require extreme or even impossible maneuvers. Certain dynamic parameters of the vehicle are required for them to model the real behavior of this system with an accurate precision. These parameters must be obtained from the real system by a data acquisition system or specialized tests. Some parameters, such as moments of inertia, center of gravity location and the rigidity and damping coefficients of the suspension system and tires cannot be measured directly by the sensors, and also change its properties during the time and with different settings in which the vehicle works. Given this scenario, it is evident the importance of developing a method for the estimation of dynamic parameters of the vehicle. In this paper, we develop an estimation method based on Genetic Algorithms (GA). The estimation algorithm will minimize a cost function defined as the difference between a particular mathematical model output and the reference obtained from a commercial dynamic simulator. This simulator, CarSim, is referenced by our actual behavior, in other words, the measurements obtained in this simulator will be considered as obtained from an actual system data acquisition mounted in the vehicle. The outputs of the model evolved with each generation, making the cost function approximate to zero, according to the estimation of the parameters of the vehicle made by the algorithm. The modeling of the dynamics of 4 degrees of freedom (DOF) of the vehicle has been developed in MATLAB and the estimated parameters were compared with the reference parameters of the simulator CarSim. The estimation method shown in this paper has advantages with respect to other classical estimation methods.