螺旋桨静平衡检测算法改进及实验研究

针对大型船用螺旋桨制造完成后存在的静不平衡量的问题,基于静压球面轴承悬浮结构开发了螺旋桨立式静平衡检测装置。在对其结构和工作原理进行分析的基础上,利用力矩平衡原理,提出改进的求解螺旋桨静不平衡量的算法,能快速地计算出不平衡质量和其位置。通过一次性打磨或配重,将螺旋桨静不平衡量控制在允许范围内,简化了传统的反复多次打磨配重的过程;同时数据采集处理系统以单片机为核心对称重传感器数据进行采集、处理和显示。试验结果表明,该检测系统在试验过程中运行稳定,测量数据精确。     

关于卷积码编码方案的研究

文章详述了纠错码理论中的卷积码的编码方案和解码方案,重点分析了它存在的缺陷,然后利用循环码的编码方案改进了卷积码的编码方案,得到了循环卷积码,提高了它编码的纠错能力。     

基于U-Net的盐体识别方法

卷积神经网络在计算机视觉领域取得重大突破，利用其强大的图像处理能力，将地下沉积盐体的识别问题转化为图像语义分割问题，应用深度卷积神经网络实现盐体地震图像的像素级语义分割.本文在U-Net基础上，增加网络深度并同时引入批归一化和Dropout处理，使得神经网络模型具有更高的可信度和更强的泛化能力.通过实验发现，在卷积层之后引入批归一化处理，并在池化层和叠加层之后引入Dropout可以稳定提升模型对盐体图像的分割性能.     

Measurement Accuracy in Phase-Shifting Point Diffraction Interferometer with Two Optical Fibers

A phase-shifting point diffraction interferometer (PS/PDI) with point sources of two single mode optical fibers has been developed, which will be appropriate for the surface figure measurement of large aperture optics on a sub-nanometer scale. To reduce the measurement error factors, a fiber optic plate (FOP) is used as a projection plane for interference pattern. Errors caused by imperfection of optical alignment, such as position of point sources and tilt of FOP, are minimized by analyzing the measured phase data with an original method. Measurement accuracy in the PS/ PDI is estimated with the interference pattern produced by the two optical fiber sources. If inhomogeneity of the FOP and a systematic error of the PS/PDI are eliminated, the measurement accuracy of the present system is estimated to be less than 4nm P-V and 0.7nm rms, respectively, at a measurement wavelength of 632.8 nm.     

球形钴酸锂的乳液法合成及其结构、性能研究

采用一种特殊乳液法, 制备了锂离子电池正极材料球形LiCoO₂. 分别研究了一次颗粒尺寸、烧结时间对于二次造粒的影响. 还对LiCoO₂球体内部的结构进行了SEM和HRTEM研究, 发现一次颗粒非常紧密地融合在一起, 晶体边界区域没有明显的非晶相区, 也没有纳米级别的孔洞、裂缝. 电化学性能研究表明, 在900 ℃烧结4 h的材料具有较高的首次放电容量(143 mAh•g^－1)和良好的循环性能     

Determination of the refractive index of spherical erythrocytes of human blood in the range 0.3–1.2 µm

A proposed procedure for determining the spectral values of the refractive index of spherical erythrocytes of human blood is based on angular light scattering and the use (for one-time scattering) of a multiple regression equation. A theoretical study of the stability of the relationship between the light scattering indicatrix and the refractive index of erythrocytes toward possible variations of their microstructure parameters is performed in order to select the optimal angles. The precision of the proposed method and its sensitivity to indicatrix measurement uncertainties are estimated. __________ Translated from Zhurnal Prikladnoi Spektroskopii Vol. 74, No. 3, pp. 382–386, May–June, 2007.     

四边固支球面扁壳的解析解

本文用解析法四边固支球面扁壳的在均布荷载作用下的内力和挠度进行了计算。     

Variational extension of the mean spherical approximation to arbitrary dimensions

We generalize a variational principle for the mean spherical approximation for a system of charged hard spheres in 3D to arbitrary dimensions. We first construct a free energy variational trial function from the Debye-Hückel excess charging internal energy at a finite concentration and an entropy obtained at the zero-concentration limit by thermodynamic integration. In three dimensions the minimization of this expression with respect to the screening parameter leads to the mean spherical approximation, usually obtained by solution of the Ornstein-Zernike equation. This procedure, which interpolates naturally between the zero concentration/coupling limit and the high-concentration/ coupling limit, is extended to arbitrary dimensions. We conjecture that this result is also equivalent to the MSA as originally defined, although a technical proof of this point is left for the future. The Onsager limitT ΔS ^MSA /ΔE ^MSA → 0 for infinite concentration/coupling is satisfied for all d ≠ 2, while ford=2 this limit is 1. On leave from Department of Physics, University of Puerto Rico, Mayagüez Campus, Mayagüez, Puerto Rico, 00681.     

深度卷积网络的多品种多厂商药品近红外光谱分类

近红外光谱（NIR）分析具有分析高效、样品无损、环境无污染以及可现场检测等优点,特别适合药品的快速建模分析。但NIR存在吸收强度弱以及谱带重叠等缺点,需要建立稳健可靠的化学计量学模型对其进行分析。深度卷积神经网络是深度学习方法中一个重要分支,它通过逐层抽取数据特征并进行组合、转换,形成更高层的语义特征,具有极强的建模能力,广泛应用于计算机视觉、语音识别等领域,而在药品NIR分析方面尚未见报道。基于深度卷积网络模型,对药品NIR多分类建模进行研究。针对药品NIR数据的特点,设计若干个面向多品种、多厂商药品NIR分类的一维深度卷积网络模型。模型中卷积层和池化层交叠排列用于逐层抽取NIR数据特征,输出层连接softmax分类器,对药品NIR数据进行分类概率预测。在输出层之前采用全局最大池化层,将特征图进行整体池化,形成一个特征点,用于解决全连接层存在的限制输入维度大小,参数过多的问题。同时,在网络模型中引入批处理操作和dropout机制,以防止梯度消失和减小网络过拟合的风险。在网络模型的设计过程中,通过设计不同的卷积网络层数以及不同的卷积核尺寸大小,分析其对建模效果的影响,同时分析五种经典数据预处理方法对NIR分析的影响。以我国7个厂商生产的头孢克肟片和11个厂商生产的苯妥英钠片样本NIR为实验对象, 建立药品的多品种、多厂商分类模型,该模型在二分类、多分类实验中取得了良好的分类效果。在十八分类实验中,当训练集与测试集比例为7∶3时,分类准确率为99.37±0.45,比SVM, BP, AE和ELM算法取得更优的分类性能。同时,深度卷积神经网络模型推理速度较快,优于SVM和ELM算法,但训练速度慢于二者。大量实验结果表明,深度卷积神经网络可对多品种、多厂商药品NIR数据准确、可靠地判别分类,且模型具有良好的鲁棒性和可扩展性。该方法也可推广到烟草、石化等其他领域的NIR数据分类应用中。