结合预训练模型和语言知识库的文本匹配方法

针对文本匹配任务,该文提出一种大规模预训练模型融合外部语言知识库的方法。该方法在大规模预训练模型的基础上,通过生成基于WordNet的同义—反义词汇知识学习任务和词组—搭配知识学习任务引入外部语言学知识。进而,与MT-DNN多任务学习模型进行联合训练,以进一步提高模型性能。最后利用文本匹配标注数据进行微调。在MRPC和QQP两个公开数据集的实验结果显示,该方法可以在大规模预训练模型和微调的框架基础上,通过引入外部语言知识进行联合训练有效提升文本匹配性能。     

液滴撞击干燥倾斜壁面铺展实验研究

通过高速摄影技术及像素分析方法对液滴撞击干燥倾斜壁面进行实验研究，分析了不同倾斜角及韦伯数（We）对液滴飞溅及铺展的影响，结果表明，倾斜角的增加有利于抑制飞溅的产生；We一定时，液滴的前铺展因子随着倾斜角的增加而增加，后铺展因子随着倾斜角的增加而减小；液滴的前后初始铺展速度均大于液滴的撞击速度，且随着倾斜角的增加而减小。      

基于DoG检测图像特征点的快速二进制描述子

针对SIFT描述子实时性差和传统二进制描述子对尺度、旋转和视角变化鲁棒性差的问题,本文通过优化采样模式和添加灰度差分不变量比较测试进行改进,提出了一种鲁棒性更高的二进制描述子。首先,设计了一种尺度关联、编号标记的采样模式;然后,旋转采样模式中各采样点到特定位置,确保描述子尺度、旋转不变性;接着,分析了采样点点对模式对描述子的影响,选择使用机器学习训练后的128对采样点对;最后,选择灰度值比较测试及梯度绝对值和比较测试构建二进制描述子。实验中采用DoG检测图像关键点,结果表明:本文提出的描述子在描述子构建和描述子匹配上比SIFT描述子分别快84%和67%;在有视角变化的图像匹配上,准确率比传统的二进制描述子高3%~5%,召回率平均要高30%以上。本文提出的特征点描述方法适用于时间要求高的图像匹配领域。     

Synthesis of a novel DOPO‐based polyphosphoramide with high char yield and its application in flame‐retardant epoxy resins

A novel and highly effective flame retardant (FR), DOPO‐TPMP oligomer, was synthesized by a simple condensation of 4‐(hydroxymethyl)‐2,6,7‐trioxa‐1‐phosphabicyclo[2.2.2]octane‐1‐oxide and phosphorus oxychloride followed by a polycondensation reaction with 6‐(2,5‐dihydroxyphenyl)‐6H‐dibenzo[c,e][1,2]oxaphosphinine‐6‐oxide. The chemical structure of DOPO‐TPMP was well characterized using Fourier transform infrared and NMR spectra. DOPO‐TPMP was used as an additive‐type FR for epoxy resin (EP). The FR properties of the resultant EP composites were investigated by limiting oxygen index (LOI) test, UL‐94 vertical burning test and cone calorimeter measurements. Specifically, the EP composite containing 10.0% DOPO‐TPMP achieved a LOI value of 36.1%, V‐0 rating in the UL‐94 test and a 58% reduction in peak heat release rate. Further mechanism analysis attributed the enhanced flame retardancy to the increased char yield on the addition of DOPO‐TPMP. © 2019 Society of Chemical Industry     

全过程集成管理服务在西安地铁5号线二期工程的应用

近年来,我国不断推动房地产和市政工程建设实施全过程工程咨询管理,即全过程集成管理服务.西咸轨道投资建设公司成立时间短、人员少,却仅用了一年零三个月的时间,完成了西安地铁5号线二期车站设备系统的工程建设任务,这得益于全过程集成管理模式的应用.     

Effect of Al2O3/CaO on the melting and mineral transformation of Ningdong coal ash

Coal ash melting characteristics has a direct impact on the smooth operation of entrained gasifier. Mineral conversion of coal ash is very significant to be investigated, because the mineral can affect the melting temperature and viscosity under high temperature conditions. In this paper, the effects of different Al₂O₃/CaO on the mineral conversion, melting temperature and viscosity of Ningdong coal ash are studied by the combination of experiment and simulation. The trend of melting temperature decreases firstly and rises with increasing Al₂O₃/CaO. The ash melting point reached to the lowest when the ratio is 1.23. XRD and Factsage software are used to analyze crystallization behavior of samples. The results show that the content of anorthite, albite and corundum increases and subsequently decreases, while the content of mullite decreases firstly and then rises with increasing Al₂O₃/CaO. High content with CaO can contribute to form albite and anorthite of low-melting. Besides, high content with Al₂O₃ can tend to produce mullite of high-melting. The results of experimental and simulation are basically the same, which lays a foundation for the melting characteristics of Ningdong coal ash and can provide technical support for the smooth operation of the entrained-gasifier.