卷积神经网络用于近红外光谱古筝面板木材分级

目前,我国乐器制作行业在古筝面板用木材等级的筛选上主要依赖于技师主观评判,但此法缺少科学理论的依据,效率低,客观性及出材率的提高等方面受到限制,无法满足乐器市场的大量需求。实现古筝面板用木材快速、智能化的分级工作是一个急需解决的课题。近红外光谱非常适用于测量含氢的有机物质。古筝面板木材主要化学成分的化学键均由含氢基团组成,不同等级板材的化学成分存在差异,这些差异反映在近红外光谱中,为判断木材等级提供了可能。同时卷积神经网络对非线性数据具有较强的特征提取能力,所以提出一种应用卷积神经网络模型对光谱数据进行分析的方法,进而判别木材的等级。应用了Savitzky Golay一阶、二阶微分两种预处理方法和核主成分分析、连续投影算法两种数据压缩方法,通过所设计的卷积神经网络模型以样本识别准确率和模型构建过程中的损失值作为判定指标选出最佳预处理和数据压缩方法。为了提高模型提取分析光谱数据的能力和避免过拟合现象,应用了多通道卷积核、批量归一化和early stopping策略,将通过两层卷积层提取的特征信息送入全连接层,从而充分提取剩余信息,通过Softmax函数获得板材的最终预测等级,从而确定了最终模型。最终Savitzky Golay一阶微分和核主成分分析为最佳数据处理方法,同时得出用于区分不同等级的古筝面板用木材的主要关键谱带,分别为1 163~1 243, 1 346~1 375和1 525~1 584 nm。将该模型应用于测试集样本,古筝面板用木材的等级识别准确率为95.5%。实验结果表明所提出的方法可以高效地处理光谱数据,有效识别区分不同等级的古筝面板用木材的关键特征,从而为广阔的乐器市场提供一定的技术支持。     

Path to achieving molecular dispersion in a dense reactive mixture

A uniform dispersion of reactants is necessary to achieve a complete reaction involving multicomponents. In this study, we have examined the role of plasticizer in the reaction of two seemingly unlikely reactants: a highly crystalline hexamethylenetetramine (HMTA) and a strongly hydrogen bonded phenol formaldehyde resin. By combining information from NMR, infrared spectroscopy and differential scanning calorimetry, we were able to determine the role of specific intermolecular interactions necessary for the plasticizer to dissolve the highly crystalline HMTA and to plasticize the phenol formaldehyde resin in this crosslinking reaction. The presence of the plasticizer increased the segmental mobility, disrupted the hydrogen bonded matrix, and freed the hydroxyl units, which further increased the solubility of the HMTA. Both the endothermic and exothermic transitions are accounted for in the calorimetric data obtained. For the first time, it is possible to obtain the effective molar ratio of each component needed to complete the crosslinking reaction efficiently. © 2015 Wiley Periodicals, Inc. J. Polym. Sci., Part B: Polym. Phys. 2015 , 53, 1519–1526     

小孔流速实验中的排水时间和流量系数的研究

本文以水作为理想流体,考虑到水头损失和孔口缩流效应,对小孔流速实验涉及的容器排水问题进行了系统的理论和实验研究.从理论上推导得出了圆柱形容器排水时间的解析解,分析了排水时间和自由液面速度、流量系数之间的规律,提出了排水时间的等效性.实验上,加工了底部开有不同小孔的大型圆柱形容器,测量了容器排水时间随液面高度的变化关系,借助实验结果计算了流量系数值,验证了容器排水时间的等效性关系,实验结果与理论计算符合较好.     

DFT mechanistic study of the H2‐assisted chain transfer copolymerization of propylene and p‐methylstyrene catalyzed by zirconocene complex

DFT computations have been performed to investigate the mechanism of H₂‐assisted chain transfer strategy to functionalize polypropylene via Zr‐catalyzed copolymerization of propylene and p‐methylstyrene (pMS). The study unveils the following: (i) propylene prefers 1,2‐insertion over 2,1‐insertion both kinetically and thermodynamically, explaining the observed 1,2‐insertion regioselectivity for propylene insertion. (ii) The 2,1‐inserion of pMS is kinetically less favorable but thermodynamically more favorable than 1,2‐insertion. The observation of 2,1‐insertion pMS at the end of polymer chain is due to thermodynamic control and that the barrier difference between the two insertion modes become smaller as the chain length becomes longer. (iii) The pMS insertion results in much higher barriers for subsequent either propylene or pMS insertion, which causes deactivation of the catalytic system. (iv) Small H₂ can react with the deactivated [Zr]?pMS?PP_n facilely, which displace functionalized pMS?PP_n chain and regenerate [Zr]? H active catalyst to continue copolymerization. The effects of counterions are also discussed. © 2014 Wiley Periodicals, Inc. J. Polym. Sci., Part A: Polym. Chem. 2015 , 53, 576–585     

基于深度学习的ELM实时识别研究

基于深度学习的方法，在HL-2A装置上开发出了一套边缘局域模(ELM)实时识别算法。算法使用5200次放电数据(约24.19万数据切片)进行学习，得到一个深度为22层的卷积神经网络。为衡量算法的识别能力，识别了HL-2A装置自2009年实现稳定ELMy H模放电以来所有历史数据(约26000次放电数据)，共识别出1665次H模放电，其中误识别35次，误报率为2.10%。在实际的1634次H模放电中，漏识别4次，漏识别率为0.24%。该误报率和漏报率可以满足ELM实时识别的精度要求。识别算法在实时控制环境下，对单个时间点的平均计算时间为0.46ms，可以满足实时控制的计算速度要求。     

CiADS束流负载效应前馈补偿的仿真评估

在高功率超导质子直线加速器中，束流负载效应是影响超导腔幅相稳定性的一个重要因素。本工作基于谐振腔建场模型，开发了超导腔系统束流负载效应的时域仿真程序，分析了束流负载效应对超导腔幅相稳定性的影响，并在C-ADS注入器II上通过相关实验测量对仿真结果进行了验证。利用该程序，评估了CiADS超导直线加速器脉冲束流的脉冲长度，以及前馈补偿的时序抖动和束流纹波等因素对腔中电磁场幅相稳定度的影响。仿真结果表明:在当前CiADS直线加速器设计参数下，为满足超导腔中电磁场0.1%与在高功率超导质子直线加速器中，束流负载效应是影响超导腔幅相稳定性的一个重要因素。本工作基于谐振腔建场模型，开发了超导腔系统束流负载效应的时域仿真程序，分析了束流负载效应对超导腔幅相稳定性的影响，并在C-ADS注入器II上通过相关实验测量对仿真结果进行了验证。利用该程序，评估了CiADS超导直线加速器脉冲束流的脉冲长度，以及前馈补偿的时序抖动和束流纹波等因素对腔中电磁场幅相稳定度的影响。仿真结果表明:在当前CiADS直线加速器设计参数下，为满足超导腔中电磁场0.1%与$0.1^{\circ}$的幅相稳定度指标，前馈时序抖动的偏差不能超过0.79 μs，束流流强的直流偏差不能超过0.9%，并且给出了束流纹波的最大抖动幅值与纹波频率之间的关系。这些结果将为CiADS超导直线加速器相关子系统技术指标的确定提供依据。     

A new molecular rectifier device and some research in its processing

Organic materials of D-π-A type MR-X (MR-1: p-dimethylaminophenylethenetrica-rbonitrile and MR-2: p-diphenylaminophenylethene tricarbonitrile) were designed and synthesized. The device with a sandwich structure shows good rectificative phenomena. The highest rectification ratio 10000 was achieved in device Cu/MR-1/Ag, and about 100 in other device M/MR-X/M (M: Cu, Ag). It has been found that rectificative phenomena exist only in the atmosphere-liquid interface region by means of liquid adsorption, and electric field could help form the oriented molecular film. __________ Translated from Journal of Fudan University (Natural Science), 2005, 44(4) (in Chinese)     

边剖分、点扩张与图的最大亏格的可约性

设γM（G）是连通图G=（V，E）的最大亏格，记EM^-（G）={e∈E(G)|G\e连通，且γM（G\e）=γM（G）}。若EM^-(G)≠0，则称G是γ（G）-可约的；否则称G是γM（G）-不可约的。本文证明了边的剖分不改变图的最大亏格可约性，点的扩张不改变上可嵌入图的最大亏格可约性；并给出了两类满足EM^-（G）=E（G）的非4-边连通图。