基于RBF神经网络的FSC赛车转向梯形断开点优化

Steering trapezium splitting points optimization of FSC racing car based on RBF neural network

为了增强FSC(Formula Student China)赛车过弯时的响应和操纵稳定性,提出一种由RBF(radius basis function)神经网络进行转向梯形断开点优化的方法。首先在Adams/car虚拟样机模型中用insight模块获取64组转向梯形断开点的原始数据,然后应用原始数据对RBF神经网络进行训练,用训练好的网络优化断开点。将优化断开点代入Adams/car模型中与Car Sim赛车模型进行不足转向梯度试验和方向盘角阶跃试验对比仿真。仿真结果显示,优化后的横摆角速度峰值、稳态横摆角速度和调整时间分别降低2.13%,2.07%,16.44%,不足转向梯度值最大减少0.1 deg/g。与此同时对实车前轮反向跳动工况进行K&C台架试验,对前束角变化值进行测定。在相同轮心垂直位移为±15 mm工况下,实车试验、优化后、优化前前束角变化值分别为0.02°,0.017°,0.114°,优化后前束角变化量降低85.1%。仿真和试验结果均表明,该方法有较高的可信度,提升了赛车过弯时的响应和操纵稳定性,为断开点量化设计提供一种方法。

In order to increase FSC（ Formula Student China） car cornering response and handling stability,a steering trapezium splitting points RBF（ radius basis function） neural network（ NN） optimization algorithm is established. Firstly,64 groups of raw data of steering trapezium splitting points are generated by insight module in Adams / car virtual model. Raw data are used for training RBF NN.Splitting points are optimized by trained RBF NN. Understeer gradient simulation and steer wheel step response simulation are contrasted between Adams / car model with optimized splitting points and CarSim model. The simulation results showed that optimized peak yaw rate,steady yaw rate and setting time decreased respectively 2. 13%,2. 07%,16. 44%,and the maximum of under-steer gradient was reduced to 0. 1 deg / g. Furthermore,wheel travel condition was tested on a real car to measure toe an-gle. Under condition that wheel center travel distance is ± 15 mm at vertical direction,toe angle in real car test,with optimized model and un-optimized model changed to 0. 02°,0. 017° and 0. 114° respectively,and the toe angle of optimized model decreased alteration is 85. 1%. Results of simulation and test showed that RBF NN algorithm has high reliability,provides an excellent cornering response and handling stability,and is a optimization method for quantization splitting points.