基于Matlab的SCM822H齿轮钢性能预测

PROPERTY PREDICTIONS OF SCM822H GEAR STEEL BASED ON NEURAL NETWORK TOOLBOX IN MATLAB

采用Matlab的人工神经网络工具箱建立BP人工神经元网络,预测SCM822H齿轮钢的性能。选择10×12×3网络结构及基于Levenberg-Marquardt优化算法和改进的误差函数的训练函数trainbr,BP网络对SCM822H齿轮钢的性能进行快速训练的同时,使网络的泛性得到提高。最后对网络性能进行回归分析,证明了网络设计的合理性。使用训练好的网络对SCM822H齿轮钢力学性能及淬透性进行预测,预测结果表明,网络具有较高的预测精度,可在实际生产和科学研究中进行应用。

A feed-forward back propagation （BP） artificial neural network （ANN） is introduced for the prediction of properties of SCM822H gear steel based on the artificial neural network toolbox in MATLAB. With proper selection of neural network architecture of 10 × 10 × 3 and training function trainbr based on Levenberg-Marquardt optimization algorithm and improved error function, the BP network performs fast training on the properties of SCM822H gear steel with the raise of network generalization ability at the same time. Regressive analysis is made on network performance, and the analysis result demonstrates the rationality of this network design. The trained network is used to predict the mechanical properties and penetration hardenabilities of SCM822H gear steel. The results show that the ANN used can predict the properties with low error, and it can be applied in scientific research and practical production.