针对卷积神经网络(Convolutional neural network,CNN)模型在对工业数值型数据分类方面存在特征使用不充分、模型分类性能不佳等问题,提出了一种基于自适应卷积核的改进CNN(Improved CNN based on adaptive convolution kernel, ACK-IC...针对卷积神经网络(Convolutional neural network,CNN)模型在对工业数值型数据分类方面存在特征使用不充分、模型分类性能不佳等问题,提出了一种基于自适应卷积核的改进CNN(Improved CNN based on adaptive convolution kernel, ACK-ICNN)算法。该算法为了增加特征的重复使用率,构建了一种多尺度卷积核的模型结构,通过融合处理卷积核提取的不同特征来实现,增强了模型的适应能力;为了进一步提升该算法的性能,利用网格搜索算法自适应选取CNN中最优的卷积核大小,使得模型能够提取出最优的特征。采用TE过程的故障数据对其进行测试,并与支持向量机、极限学习机、最近邻等典型的数据驱动方法进行对比,测试结果表明,该算法能有效提升各类故障的分类精度。展开更多
In the present analysis, several parameters used in a numerical simulation are investigated in an integrated study to obtain their influence on the process and results of this simulation. The parameters studied are el...In the present analysis, several parameters used in a numerical simulation are investigated in an integrated study to obtain their influence on the process and results of this simulation. The parameters studied are element formulation, friction coefficient, and material model. Numerical simulations using the non-linear finite element method are conducted to produce virtual experimental data for several collision scenarios. Pattern and size damages caused by collision in a real accident case are assumed as real experimental data, and these are used to validate the method. The element model study performed indicates that the Belytschko-Tsay element formulation should be recommended for use in virtual experiments. It is recommended that the real value of the friction coefficient for materials involved is applied in simulations. For the study of the material model, the application of materials with high yield strength is recommended for use in the side hull structure.展开更多
文摘针对卷积神经网络(Convolutional neural network,CNN)模型在对工业数值型数据分类方面存在特征使用不充分、模型分类性能不佳等问题,提出了一种基于自适应卷积核的改进CNN(Improved CNN based on adaptive convolution kernel, ACK-ICNN)算法。该算法为了增加特征的重复使用率,构建了一种多尺度卷积核的模型结构,通过融合处理卷积核提取的不同特征来实现,增强了模型的适应能力;为了进一步提升该算法的性能,利用网格搜索算法自适应选取CNN中最优的卷积核大小,使得模型能够提取出最优的特征。采用TE过程的故障数据对其进行测试,并与支持向量机、极限学习机、最近邻等典型的数据驱动方法进行对比,测试结果表明,该算法能有效提升各类故障的分类精度。
文摘In the present analysis, several parameters used in a numerical simulation are investigated in an integrated study to obtain their influence on the process and results of this simulation. The parameters studied are element formulation, friction coefficient, and material model. Numerical simulations using the non-linear finite element method are conducted to produce virtual experimental data for several collision scenarios. Pattern and size damages caused by collision in a real accident case are assumed as real experimental data, and these are used to validate the method. The element model study performed indicates that the Belytschko-Tsay element formulation should be recommended for use in virtual experiments. It is recommended that the real value of the friction coefficient for materials involved is applied in simulations. For the study of the material model, the application of materials with high yield strength is recommended for use in the side hull structure.
