基于并行多通道卷积长短时记忆网络的轴承寿命预测方法

在预测轴承剩余使用寿命时，数据间的时序特性是一个可以利用的重要隐藏信息。为了更好地提取具有时序信息的特征用于预测，提出了一种基于并行多通道卷积长短时记忆网络（PMCCNN-LSTM）的剩余使用寿命预测模型。该模型主要由两部分组成：前端为并行多通道卷积网络（PMCCNN），提取信号特征，挖掘数据的时序特性，并采用逐层训练和微调的方式提升参数的收敛性；后端为长短时记忆（LSTM）网络，基于特征进行剩余使用寿命预测，并采用加权平均的方法对预测结果进行平滑处理。在一个轴承加速寿命实验的公开数据集上使用留一法验证了该模型的准确性，实验结果表明：所提模型的平均误差与最大误差分别比传统的卷积神经网络（CNN）低23.38%和15.84%，比传统的LSTM低24.14%和19.01%，比卷积长短时记忆网络（CNN-LSTM）低30.32%和23.09%。     

煤矿主提升机平衡尾绳更换工艺的改进

更换煤矿主提升机平衡尾绳是煤矿重大检修项目之一,原施工方案为两条平衡尾绳同时更换,存在较多不合理之处,改进为以新绳带旧绳、单绳依次更换的方案,实现了安全、快捷地更换平衡尾绳,同时减少施工对矿井生产的影响,提高煤矿经济效益。     

热处理对挤压态Mg-6Gd-5Y-1Zn合金组织与性能的影响

通过对Mg-6Gd-5Y-1Zn(质量分数,%)合金在固溶和时效处理状态下显微组织和力学性能的研究发现,α-Mg基体、沿挤压方向分布的条状18R-LPSO相、少量的Mg₂₄(GdYZn)₅ 相以及细层片状的14H-LPSO相构成了挤压态合金的组成相。挤压态合金经固溶(T4)处理后,一部分18R-LPSO相溶入基体,并且基体中的14H-LPSO相伸长同时粗化。挤压态合金经过固溶加时效(T6)处理后,大量β′相从α-Mg基体中析出。T6态合金的室温力学性能最好,其屈服强度、抗拉强度及伸长率分别为272 MPa、406 MPa和6.1%。β′相沉淀也发生在挤压态合金的直接人工时效(T5)处理过程,但相比于T6处理,14H-LPSO相和β′相在基体中的体积分数均偏低。     

覆岩结构类型与开采覆岩损害特征分析

煤层开采打破了上覆岩层原有的平衡,必然引起岩层的移动、破断,导致应力场和裂隙场的改变,工作面出现矿压显现。上覆岩层的岩性和组合结构对工作面开采覆岩损害特征分布影响显著,一般岩性越坚硬,上覆岩层应力集中程度越高,采动裂隙越发育,矿压显现也越明显;坚硬岩层尤其是关键层对覆岩的破断运动起控制作用,覆岩破断发展到一定层位后软弱岩层往往是抑制覆岩破断进一步向上发展的关键。当工作面上覆岩层同时存在远近场关键层时,近场关键层的周期性破断引起裂隙带范围内岩层的运动和工作面周期来压;而远场关键层的周期性破断导致地表裂缝或台阶下沉,工作面压力急剧上升,有时甚至导致煤矿动压灾害,尤其是远近场关键层同步破断时动力现象更明显。通过地面钻孔使远场关键层提前破断是防止此类动力灾害的关键。     

Study on estimating electric potential in space between parallel electrodes

We have been studying on estimating distribution of permittivity between measurement electrodes using capacitance and electric potential. Two arc electrodes were separated by long distance and there electrodes were surrounded by additional electrodes respectively. In past research work, we carried out numerical electric analysis for calculating the capacitance and electric potential using Finite Element Method (FEM) and compared with experimental and numerical results. The capacitance values were almost agreed with experimental and numerical results. However, the electric potential values were different between experimental and numerical results in conventional studies. In this paper, we proposed an equivalent circuit including the stray capacity and measurement method for capacitance, the electric potential in space between long distance electrodes was estimated.     

基于深度学习的彩色以及近红外图像去马赛克

单传感器捕获的彩色-近红外（RGB-NIR）图像存在光谱干扰，从而导致重建出的标准彩色图像（RGB）图像与近红外（NIR）图像存在色彩失真以及细节信息模糊。针对这个问题提出一种基于深度学习的去马赛克方法，通过引入跳远连接与稠密连接解决了梯度消失和梯度弥散问题，使得网络更容易训练，并且提升了网络的拟合能力。首先，用浅层特征提取层提取了马赛克图像的像素相关性以及通道相关性等低级特征；然后，将得到的浅层特征图输入到连续多个的残差稠密块以提取专门针对去马赛克的高级语义特征；其次，为充分利用低级特征与高级特征，将多个残差稠密块提取到的特征进行组合；最后，通过全局跳远连接恢复最终的RGB-NIR图像。在深度学习框架Tensorflow上使用公共的图像与视觉表示组（IVRG）数据集、有植被的户外多光谱图像（OMSIV）数据集和森林（Forest）三个公开数据集进行实验。实验结果表明，所提方法优于基于多级自适应残差插值、基于卷积卷积和神经神经网络以及基于深度残差U型网络的主流的RGB-NIR图像去马赛克方法。     

饱和尾粉砂边界面模型在FLAC3D中的开发

由于岩土材料的复杂性，FLAC3D软件自带的本构模型不能满足实际数值分析的全部要求。利用FLAC3D预留的UMD接口，在C++的编译环境下实现了一种基于临界状态模型CASM边界面模型的开发。介绍了边界面模型的基本组成，给出了开发的关键步骤。并基于三轴试验结果，利用该模型对饱和尾粉砂静/动力学特性进行了分析，论证了该模型对饱和尾粉砂的适用性。     

Performance assessment of gas crossover phenomenon and water transport mechanism in high pressure PEM electrolyzer

To improve proton exchange membrane (PEM) electrolyzes’ performance the voltage loss through them should be avoided. In this work, it is intended to analyze losses including of diffusion loss, ohmic loss due to electrode, bipolar plate (BP), and membrane resistances, and gas crossover associated with the water transferring mechanisms. All of the losses are associated with water transferring mechanisms, which is created due to electro-osmoic drag, pressure differential between the anode and cathode sides, and diffusion. Furthermore, the effect of membrane thickness, cathode pressure, and operating temperature on the hydrogen crossover is examined. In addition, the contribution of ohmic loss due to electrode bipolar plate (BP), and membrane resistances is studied and, the contribution of different losses on the cell performance is discussed. Results show that raising cathode pressure from 1 to 40 bar lead to the increment of anodic hydrogen content from 1.038% to 21% at the specific current density of 10,000 A/m2. Enhancing the thickness of membrane has considerable impact on decrementing anodic hydrogen content, but the mass transfer loss rises from 0.022 to 0.027 V with enhancing membrane thickness from 50 to 300 μm, respectively. Furthermore, the contribution of voltage losses, assigned to each of losses are equal to 85%, 3%, and 12% for activation, diffusion and ohmic losses, respectively. It is found that, from the reported contribution for ohmic loss, the contribution of electrode BP, and membrane resistances are 31% and 69%, respectively.     

Controllable 1D Patterned Assembly of Colloidal Quantum Dots on PbSO4 Nanoribbons

Control of the 1D self‐assembly pattern of colloidal quantum dots (QDs) on PbSO₄ nanoribbon (NRb) templates is achieved. The internal structure of the NRbs is investigated by X‐ray diffraction, revealing the isotropic packing of the PbSO₄ nanoclusters in the NRbs. Colloidal QDs in a chloroform/hexane mixture are adsorbed onto the region close to the edges of the NRbs and form a 1D assembly of straight single line or double lines by controlling the amount of OAm. This is the first demonstration of a densely packed 1D self‐assembly of colloidal QDs with a straight line pattern without the use of any molecular bridge or adhesive. Atomic force microscopy measurements of the NRbs show depressions in the phase profile along the width of the NRbs, corresponding to the position of the 1D QD chain. The amount of adsorbed QDs on the NRbs in solution decreases as the addition of OAm increases, suggesting that additional OAm prevents interaction between the QDs and NRbs but facilitates the uniform adsorption of the 1D assembly. The low‐dimensional self‐assembly of colloidal QDs in this study opens up the possibility for the creation of anisotropically assembled QD superstructures.