AFDCNN方法及其在齿轮故障定量诊断中的应用

传统的浅层故障诊断方法需要根据先验知识人为选取故障特征和设计分类器,基于深度学习理论的深度卷积神经网络(deep convolutional neural network,DCNN)能够完成对故障特征的自适应提取与健康状况的智能诊断,但是作为一种大数据处理工具,DCNN无法适应于小样本数据的故障诊断。在Fisher准则深度卷积神经网络识别算法(fisher-based deep convolutional neural network,FDCNN)的基础上提出了一种自适应Fisher准则的深度卷积神经网络(adaptive fisher-based deep convolutional neural network,AFDCNN)方法。在DCNN的基础上引入类内和类间约束的能量函数模型并采用优化算法得到最优能量函数模型参数,能够直接从原始时域信号中自动提取故障特征,实现智能诊断。对齿轮故障实验信号的分析结果表明,所提出的方法可以有效地实现小样本下的齿轮故障智能定量诊断。

基于自适应分段跳过规则的视点合成优化方法

对目前三维高效视频编码(3D-HEVC)标准的性能进行了分析,针对其深度图编码采用的合成视点失真变化(SVDC)计算复杂度高的问题,提出了一种基于自适应分段跳过规则的视点合成优化(VSO)方法。考虑到原始方法跳过了深度数据没有变化的模块或像素行,但忽略了可以在VSO中跳过的较小区域,该方法首先根据深度零失真准则和纹理平滑度准则判别出深度图中不会造成合成视点失真的区域,然后在深度图编码进行SVDC计算前基于自适应分段规则来提前跳过上述零失真区域,最后终止该区域的SVDC计算过程,有效地降低编码复杂度。对比实验结果证明了该方法的有效性,能在保持编码质量不变的前提下,平均减少14.2%的编码时间。

Critical embedment depth of a rigid retaining wall against overturning in unsaturated soils considering intermediate principal stress and strength nonlinearity

The overturning stability is vital for the retaining wall design of foundation pits, where the surrounding soils are usually unsaturated due to water draining. Moreover, the intermediate principal stress does affect the unsaturated soil strength; meanwhile, the relationship between the unsaturated soil strength and matric suction is nonlinear. This work is to present closed-form equations of critical embedment depth for a rigid retaining wall against overturning by means of moment equilibrium. Matric suction is considered to be distributed uniformly and linearly with depth. The unified shear strength formulation for unsaturated soils under the plane strain condition is adopted to characterize the intermediate principal stress effect, and strength nonlinearity is described by a hyperbolic model of suction angle. The result obtained is orderly series solutions rather than one specific answer; thus, it has wide theoretical significance and good applicability. The validity of this present work is demonstrated by comparing it with a lower bound solution. The traditional overturning designs for rigid retaining walls, in which the saturated soil mechanics neglecting matric suction or the unsaturated soil mechanics based on the Mohr-Coulomb criterion are employed, are special cases of the proposed result. Parametric studies about the intermediate principal stress, matric suction and its distributions along with two strength nonlinearity methods on a new defined critical buried coefficient are discussed.

Analytical solution for stress and deformation of the mining floor based on integral transform

Following exploitation of a coal seam, the final stress field is the sum of in situ stress field and an excavation stress field. Based on this feature, we firstly established a mechanics analytical model of the mining floor strata. Then the study applied Fourier integral transform to solve a biharmonic equation,obtaining the analytical solution of the stress and displacement of the mining floor. Additionally, this investigation used the Mohr–Coulomb yield criterion to determine the plastic failure depth of the floor strata. The calculation process showed that the plastic failure depth of the floor and floor heave are related to the mining width, burial depth and physical–mechanical properties. The results from an example show that the curve of the plastic failure depth of the mining floor is characterized by a funnel shape and the maximum failure depth generates in the middle of mining floor; and that the maximum and minimum principal stresses change distinctly in the shallow layer and tend to a fixed value with an increase in depth. Based on the displacement results, the maximum floor heave appears in the middle of the stope and its value is 0.107 m. This will provide a basis for floor control. Lastly, we have verified the analytical results using FLAC3 Dto simulate floor excavation and find that there is some deviation between the two results, but their overall tendency is consistent which illustrates that the analysis method can well solve the stress and displacement of the floor.

Mechanism of zonal disintegration phenomenon in enclosing rock mass around deep tunnels

In order to study the mechanism of the zonal disintegration phenomenon(ZDP),both experimental and theoretical investigations were carried out.Firstly,based on the similarity law,gypsum was chosen as equivalent material to simulate the deep rock mass,the excavation of deep tunnel was modeled by drilling a hole in the gypsum models,two circular cracked zones were measured in the model,and ZDP in the enclosing rock mass around deep tunnel was simulated in 3D gypsum model tests.Secondly, based on the elasto-plastic analysis of the stressed-strained state of the surrounding rock mass with the improved Hoek-Brown strength criterion and the bilinear constitutive model,the maximum stress zone occurred in vicinity of the elastic-plastic interface due to the excavation of the deep tunnel,rock material in maximum stress zone is in the approximate uniaxial loading state owing to the larger tangential force and smaller radial force,the mechanism of ZDP was explained,which lay in the creep instability failure of rock mass due to the development of plastic zone and transfer of the maximum stress zone within the rock mass.Thirdly,the analytical critical depth for the occurrence of ZDP was obtained,which depended on the mechanical indices and stress concentration coefficient of rock mass.

指向深度学习的深度分析研究

深度学习是指在真实情境中,依据学习分析,由学习者主动建构新知识,迁移运用并解决问题的学习方式。所依据的不只是空间上的分析,还应是时间和空间相结合的深度分析。而深度分析作为深度学习的基础和前提,又可分为目标分析深度,内容分析深度和评价分析深度。这几者不是独立的,而是紧密联系而又不可分割的深度准则。因此,指向深度学习的深度分析必须强调系统性。