基于机器学习预测环氧树脂复合材料抗冲击性能  

Predicting the Impact Resistance of Epoxy Resin Composite Materials by Machine

作　　者：伍宝华[1] 关留祥 方秀苇[3] WU Baohua;GUAN Liuxiang;FANG Xiuwei(Quality Science Research Center,Henan Quality Polytechnic,Pingdingshan 467000,China;Textile College,Zhongyuan Institute of Technology,Zhengzhou 451191,China;School of Food and Chemical Engineering,Henan Quality Polytechnic,Pingdingshan 467000,China)

机构地区：[1]河南质量工程职业学院质量科学研究中心,河南平顶山467000 [2]中原工学院纺织学院,河南郑州451191 [3]河南质量工程职业学院食品与化工学院,河南平顶山467000

出　　处：《塑料工业》2024年第10期119-125,143,共8页China Plastics Industry

基　　金：中国纺织工业联合会科技指导性计划项目(2021043)。

摘　　要：剩余压缩强度(RCS)是评价复合材料受到冲击损伤后力学性能的重要指标。采用声发射技术(AE)对玻璃纤维增强环氧树脂复合材料冲击载荷进行了在线监测,分析了振铃计数、峰值计数、信号强度和信号均方根值4种冲击载荷参数,采用人工神经元网络(ANN)和径向基网络(RBF)基于冲击载荷参数预测了试件RCS。结果表明,高冲击能量造成了试件分层、玻璃纤维断裂、环氧树脂基体开裂、纤维脱黏,当冲击能量为10、15、20和30 J时,冲击3 ms后冲击能量达到最大值,分别为10.53、16.67、21.77和27.13 J,随后冲击能量不断下降。随着冲击能量的增加,试件冲击深度从0.18 mm增加到3.35 mm,RCS从56.87 MPa降低到20.45 MPa。最优ANN模型结构为4-48-1,预测和实验RCS的均方误差(MSE)最低为0.03 MPa,最优RBF模型结构为4-21-1,MSE最低为0.01。RBF模型的局部响应特性使得其对输入数据中的噪声具有较好的鲁棒性,预测与实验RCS数据的相关系数(R2)为0.9863,而ANN模型预测结果为0.9514。Residual compressive strength(RCS)is an important indicator for evaluating the mechanical properties of composite materials after impact damages.The impact load of glass fiber reinforced epoxy resin composite materials was monitored online using acoustic emission(AE)technology.Four impact load parameters,including counts,counts to peak,signal strength,and root means square value were analyzed.The RCS of the specimen was predicted based on the impact load parameters using artificial neural network(ANN)and radial basis function network(RBF).The results show that the high impact energy causes the delamination of the specimen,glass fiber fracture,epoxy resin matrix cracking,and fiber debonding.When the impact energy is respectively 10,15,20 and 30 J,the impact energy reaches its maximum after 3 ms of impact,which is respectively 10.53,16.67,21.77 and 27.13 J.Then the impact energy continues to decrease.As the impact energy increases,the impact depth of the specimen increases from 0.18 mm to 3.35 mm,and RCS decreases from 56.87 MPa to 20.45 MPa.The optimal ANN model structure is 4-48-1,with a minimum mean square error(MSE)of 0.03 MPa for predicted and experimental RCS.The optimal RBF model structure is 4-21-1,with a minimum MSE of 0.01.The local response characteristics of the RBF model make it more robust to noise in the input data.The correlation coefficient(R 2)between the predicted and experimental RCS data is 0.9863,while the predicted result of ANN model is 0.9514.

关 键 词：径向基网络 人工神经元网络 环氧树脂复合材料 声发射 剩余压缩强度 

分 类 号：TQ323.5[化学工程—合成树脂塑料工业]