基于神经网络的受流器滑块材料载流磨损预测

Prediction on Electrical Sliding Wear Behavior of Collector Shoe Material Based on Neural Network

针对我国电气化铁路和磁浮交通对受流器滑块材料的性能要求,采用冷压烧结粉末冶金法制备了铜石墨材料,考察了该材料的载流磨损行为.结果表明:试样的磨损率随着试验载荷、速度、电流密度的增加而增大.载流条件下电流产生的电弧热是磨损率增加的主要因素.建立了摩擦磨损试验参数与磨损率之间的人工神经网络模型.以载荷、速度、电流密度作为网络的3个输入,以运行100km后试样的磨损率作为网络的1个输出,调试设计了一个3×3×1的反向传播(BP)神经网络.对神经网络的训练和检验表明该BP神经网络能够较好地预测影响因素对材料载流滑动磨损的作用,预测值与试验值的误差在10%以内,其仿真精度能够满足实际的磨损预测要求.

This paper presents the study of sliding wear behavior of a copper-graphite composite material for use in maglev vehicles and high-speed railway trains, prepared by means of the powder metallurgy method. Wear tests were conducted under laboratory conditions with a specially designed sliding wear apparatus, which simulated the tribological conditions of sliding current collectors in a maglev system. The results indicate that the wearing capacity is increased with the increasing of loads, speed and the electrical current intensity. Electrical sliding wear behavior of copper matrix materials by powder metallurgy was simulated and estimated based on neural network. The back propagation（BP） neural network with a 3× 3 × 1 structure was designed, taking loads, speed and the electrical current intensity as the input nodes, and taking wearing capacity as the output nodes. The estimated values of the wearing capacity were close to the corresponding experimental data and the errors between the estimated values and the corresponding experimental data were less than 10 %.