Permeable roads generally exhibit inferior mechanical properties and shorter service life than traditional dense-graded/impermeable roads.Furthermore,the incorporation of recycled aggregates in their construction may ...Permeable roads generally exhibit inferior mechanical properties and shorter service life than traditional dense-graded/impermeable roads.Furthermore,the incorporation of recycled aggregates in their construction may exacerbate these limitations.To address these issues,this study introduced a novel cement-stabilized permeable recycled aggregate material.A total of 162 beam specimens prepared with nine different levels of cement-aggregate ratio were tested to evaluate their permeability,bending load,and bending fatigue life.The experimental results indicate that increasing the content of recycled aggregates led to a reduction in both permeability and bending load.Additionally,the inclusion of recycled aggregates diminished the energy dissipation capacity of the specimens.These findings were used to establish a robust relationship between the initial damage in cement-stabilized permeable recycled aggregate material specimens and their fatigue life,and to propose a predictive model for their fatigue performance.Further,a method for assessing fatigue damage based on the evolution of fatigue-induced strain and energy dissipation was developed.The findings of this study provide valuable insights into the mechanical behavior and fatigue performance of cement-stabilized permeable recycled aggregate materials,offering guidance for the design of low-carbon-emission,permeable,and durable roadways incorporating recycled aggregates.展开更多
文本分类是自然语言处理中一项基本且重要的任务。基于深度学习的文本分类方法大多只针对单一的模型结构进行深入研究,这种单一的结构缺乏同时捕获并利用全局语义特征与局部语义特征的能力,且网络的加深会损失更多的语义信息。对此,提...文本分类是自然语言处理中一项基本且重要的任务。基于深度学习的文本分类方法大多只针对单一的模型结构进行深入研究,这种单一的结构缺乏同时捕获并利用全局语义特征与局部语义特征的能力,且网络的加深会损失更多的语义信息。对此,提出了一种融合多神经网络的文本分类模型FMNN(A Text Classification Model Fused with Multiple Neural Network),FMNN在最大限度减小网络深度的同时,融合了BERT,RNN,CNN和Attention等神经网络模型的特性。用BERT作为嵌入层获得文本的矩阵表示,用BiLSTM和Attention联合提取文本的全局语义特征,用CNN提取文本多个粒度下的局部语义特征,将全局语义特征和局部语义特征分别作用于softmax分类器,最后采用算术平均的方式对结果进行融合。在3个公开数据集和1个司法数据集上的实验结果表明,FMNN模型实现了更高的文本分类准确率,其中在司法数据集上的准确率达到了90.31%,证明了该模型具有较好的实用价值。展开更多
基金Project(2024JJ2073)supported by the Science Fund for Distinguished Young Scholars of Hunan Province,ChinaProjects(2023YFC3807205,2019YFC1904704)+4 种基金supported by the National Key R&D Program of ChinaProject(52178443)supported by the National Natural Science Foundation of ChinaProject(2024ZZTS0109)supported by Fundamental Research Funds for the Central Universities of Central South University,China。
文摘Permeable roads generally exhibit inferior mechanical properties and shorter service life than traditional dense-graded/impermeable roads.Furthermore,the incorporation of recycled aggregates in their construction may exacerbate these limitations.To address these issues,this study introduced a novel cement-stabilized permeable recycled aggregate material.A total of 162 beam specimens prepared with nine different levels of cement-aggregate ratio were tested to evaluate their permeability,bending load,and bending fatigue life.The experimental results indicate that increasing the content of recycled aggregates led to a reduction in both permeability and bending load.Additionally,the inclusion of recycled aggregates diminished the energy dissipation capacity of the specimens.These findings were used to establish a robust relationship between the initial damage in cement-stabilized permeable recycled aggregate material specimens and their fatigue life,and to propose a predictive model for their fatigue performance.Further,a method for assessing fatigue damage based on the evolution of fatigue-induced strain and energy dissipation was developed.The findings of this study provide valuable insights into the mechanical behavior and fatigue performance of cement-stabilized permeable recycled aggregate materials,offering guidance for the design of low-carbon-emission,permeable,and durable roadways incorporating recycled aggregates.
文摘文本分类是自然语言处理中一项基本且重要的任务。基于深度学习的文本分类方法大多只针对单一的模型结构进行深入研究,这种单一的结构缺乏同时捕获并利用全局语义特征与局部语义特征的能力,且网络的加深会损失更多的语义信息。对此,提出了一种融合多神经网络的文本分类模型FMNN(A Text Classification Model Fused with Multiple Neural Network),FMNN在最大限度减小网络深度的同时,融合了BERT,RNN,CNN和Attention等神经网络模型的特性。用BERT作为嵌入层获得文本的矩阵表示,用BiLSTM和Attention联合提取文本的全局语义特征,用CNN提取文本多个粒度下的局部语义特征,将全局语义特征和局部语义特征分别作用于softmax分类器,最后采用算术平均的方式对结果进行融合。在3个公开数据集和1个司法数据集上的实验结果表明,FMNN模型实现了更高的文本分类准确率,其中在司法数据集上的准确率达到了90.31%,证明了该模型具有较好的实用价值。
基金Project(52178443)supported by the National Natural Science Foundation of ChinaProject(2019YFC1904704)supported by the National Key R&D Program of China+1 种基金Project(2022YJ120)supported by the Key R&D Program of Chinese Academy of Railway SciencesProjects(2022JZZ03,2021JZZ01,2021JZZ02)supported by the Open Foundation of Key Laboratory of Engineering Structures of Heavy Haul Railway of Ministry of Education,China。
