基于BP神经网络的前轴锻压工艺优化

Forging Process Optimization of Front Shaft Based on BP Neural Network

以始锻温度、终锻温度、锻压比和前轴材质4个参数作为输入层函数,以耐磨损性能和疲劳性能作为输出层函数,采用4×16×8×2四层拓扑结构构建了前轴锻压工艺的神经网络优化模型,并进行了训练、预测和验证。结果表明,神经网络的耐磨损性能相对训练误差在3.2%~5.7%、疲劳性能相对训练误差在3.2%~5.5%;耐磨损性能的相对预测误差在2.6%~4.2%、平均相对预测误差为3.15%,疲劳性能的相对预测误差在2.6%~4.1%、平均相对预测误差为3.17%。

Taking the four parameters of initial forging temperature, final forging temperature, forging ratio and front shaft material as input layer functions and the wear resistance and fatigue properties as output layer functions, the neural network optimization model of the front shaft forging process was constructed by using 4 ×16 ×8 ×2 four-layer topological structure. And training, prediction and verification were carried out. The results show that the relative training error of the wear resistance of the neural network is 3.2%-5.7%, and the relative training error of fatigue performance is 3.2%-5.5%. The relative prediction error of wear resistance is 2.6%-4.2%, the average relative prediction error is 3.15%, the relative prediction error of fatigue performance is 2.6%-4.1%, and the average relative prediction error is 3.17%.