Authors

Abstract

In this paper, the ultimate crushing displacement Y1 and load Y2 of the coated fertilizer granules were obtained by uniaxial compression test as 0.450 mm and 58.668 N, respectively. The Plackett-Burman and Steepest ascent tests were taken to determine factors that had significant effects on the results and their ranges of values, respectively. Finally, the Particle Swarm Optimization - Back Propagation (PSO-BP) neural network was trained, and the correlation coefficients of training, validation, testing and overall performance were obtained as 0.98057, 0.95781, 0.96724 and 0.97459, respectively. The Y1 and Y2 are 0.450 mm and 58.703N, with a relative error of 0.06% from the actual value.

Abstract in Chinese

采用PSO_BP神经网络模型为代理模型对Bonding模型参数进行标定，首先通过单轴压缩试验得到包膜肥料颗粒的极限破碎位移和极限破碎载荷分别为0.450 mm和58.668 N。建立包膜肥料的DEM模型，分别采取Plackett-burman和Steepest ascent test确定对结果影响显著的因素及其取值范围。采用全因素试验数据训练PSO_BP神经网络，得到训练过程、验证过程、测试过程和整体性能的相关系数分别0.98057、0.95781、0.96724和0.97459，表明训练后的PSO_BP神经网络拟合效果良好，可以预测极限破碎位移和极限破碎载荷。PSO_BP神经网络预测结果显示，当法向刚度X1、切向刚度X2、切向极限应力X4和粘结半径X5分别为1.006E+10 N/m2、1.021E+10 N/m2、1200000Pa和0.20 mm时，压缩位移Y1和压缩载荷Y2分别为0.450 mm和58.703 N，与实际值相对误差最小为0.06%。本研究可以为离散元仿真参数的标定提供新方法和思路。