分层有杆抽油系统诊断模型的反问题计算法

建立了分层有杆抽油系统的诊断模型 ,研究了在实心抽油杆与空心抽油杆的界面处空心泵对应力波的影响 ,给出了抽油杆柱不同杆段的波动方程的显示差分格式 ,提出了一种诊断模型的反问题计算法。用此算法可在缺少一个边界条件的情况下求解诊断模型 ,获得抽油系统各个工况下抽油杆柱任意截面的受力状态 ,从而可绘制出分层有杆抽油系统的空心泵示功图和管式泵示功图。将此算法与模式识别技术相结合就可对分层有杆抽油系统的故障进行诊断。     

定向井有杆抽油系统动态性能分析方法

由于杆管间库仑摩擦的影响,定向井有杆泵抽油系统动态参数预测模型是一个非线性的偏微分方程,求解复杂。鉴于此,提出了一种新的分析方法。该方法以定向井有杆抽油系统中的抽油杆柱作为研究对象,根据三次样条插值模拟得到的定向井的井眼轨迹,利用静力有限元法计算出了油管对抽油杆柱的支反力,进而得出了杆柱与油管之间的库仑摩擦力;给出了杆柱单元的受力分析;建立了有限元形式的杆柱系统动力学方程并利用状态空间法对其进行了数值求解,获得了悬点示功图。文末给出了两口油井的预测实例,并将预测结果与实测结果进行了对比。对比结果表明本文所提的分析方法是正确和有效的。     

有杆抽油泵井下抽油示功图计算机诊断系统

本文研制了有杆抽油泵井下示功图计算机诊断系统。能直接根据所测地面光杆的载荷和位移信号,计算及绘制井下示功图,并据此诊断整个抽油系统的工作状况。     

有杆抽油系统动态特性的计算机仿真

本文建立了有杆抽油系统动态特性仿真的综合数学模型．即应用数值积分法建立转子运动规律的仿真模型和应用差分法建立抽油杆柱纵向振动的仿真模型。本文提出了一种叠代方法较好地解决了两子系统之间相互耦合的问题，给出了一种便于应用的井下边界条件的仿真方法。     

薄板弯曲问题的神经网络方法

为发展神经网络方法在求解薄板弯曲问题中的应用,基于Kirchhoff板理论,提出一种采用全连接层求解薄板弯曲四阶偏微分控制方程的神经网络方法。首先在求解域、边界中随机生成数据点作为神经网络输入层的参数,由前向传播系统求出预测解;其次计算预测解在域内及边界处的误差,利用反向传播系统优化神经网络系统的计算参数;最后,不断训练神经网络使误差收敛,从而得到薄板弯曲的挠度精确解。以不同边界、荷载条件的三角形、椭圆形、矩形薄板为例,利用所提方法求解其偏微分方程,与理论解或有限元解对比,讨论了影响神经网络方法收敛的因素。研究表明,本文方法对求解薄板弯曲问题的四阶偏微分控制方程具有一定的适应性,其收敛性受多种条件影响。相比有限元,该方法收敛速度较慢,在复杂的边界条件下收敛性不佳,然而其不基于变分原理,无需计算刚度矩阵,便可获得较高的计算精度。     

水电站压力钢管外压稳定问题的数值仿真研究

水电站压力钢管因失稳屈曲破坏过程的复杂性及影响失稳因素的多样性,导致对该问题的研究很难进行.目前该问题的传统计算方法存在一定的局限性,有时甚至计算结果相差较大.本文用神经网络模型结合模拟退火算法对该问题进行数值仿真求解,并通过测试样本验证了算法的可行性,为压力钢管外压稳定问题的分析方法开拓了一条新的研究思路.     

基于径向基函数神经网络的磨粒识别系统

应用磨粒形状特征参数、颜色特征参数和表面纹理特征参数对磨粒形态进行量化表征，并以此为输入矢量，引入径向基函数神经网络对磨损微粒进行自动分类识别，建立了适用于磨粒识别的径向基函数神经网络模型，并给出了具体算法．应用实例表明，径向基函数神经网络的收敛速度和识别率优于传统的BP神经网络．     

基于卷积神经网络的涵洞式直立堤波浪透射预测

涵洞式直立堤是一种具有特殊用途的海岸工程结构物,对其透浪特性的研究具有重要工程意义. 然而,目前众多学者对于涵洞式直立堤波浪透射问题的研究主要以理论分析、实验模拟及数值计算为主.随着机器学习技术的发展, 传统水动力学问题迎来了新的求解理念.机器学习算法可根据训练数据集自主学习相应的规律,以数据映射的方式建立水动力学特征预测模型,在实际应用中无需对流体运动控制方程进行求解, 具有较高的计算效率. 因此,本文基于卷积神经网络(convolutional neural network, CNN),对不同开孔条件下的涵洞式直立堤透浪特征进行预测.首先利用模型试验验证计算流体力学(computational fluid dynamics, CFD)模型的有效性,然后基于CFD模型生成相应的训练数据集, 通过训练卷积神经网络模型,建立相应的波浪透射结果之间的数据映射关系,实现在新的工况下对波浪透射系数以及透射波波形等特征的快速预测. 结果表明,经过训练的卷积神经网络可在极短时间内计算得到相应的结果, 并具有较高的准确性.研究成果可为波浪与海岸结构物相互作用的问题提供新的求解理念.      

抽油杆杆体及过渡段的剩余寿命研究

本文通过抽油杆材料疲劳裂纹扩展速率及抽油杆工作载荷的研究，对抽油杆杆体及过渡段存在表面缺陷时的剩余寿命进行了分析、推导、计算，得到了便于工程应用的剩余寿命图线，为抽油杆缺陷评定提供了依据。     

基于物理信息神经网络的功能梯度材料稳态/瞬态热传导分析

采用物理信息神经网络PINN(Physics-informed Neural Networks)求解稳态和瞬态功能梯度材料(FGMs)热传导问题。该方法利用控制方程、边界及初始条件的残差构造损失函数,在无任何响应数据的情况下得到了更具泛化能力的神经网络模型,同时避免了传统数值方法在求解计算力学问题时所需的微分、积分公式推导以及繁重的建模和划分网格等前处理工作。本文探究了PINN及其域分解的扩展物理信息神经网络XPINN(eXtended Physics-informed Neural Networks)在求解稳态和瞬态FGMs热传导问题时的适用性,讨论了网络结构对预测结果的影响。研究结果表明,PINN/XPINN在解决几何复杂的稳态和瞬态FGMs热传导问题时仍具有较高的可靠性和简洁的求解流程,同时,为极端环境下求解复杂多场耦合和夹杂等问题提供了新思路。     

The effects of Coulomb friction on the performance of centrifugal pendulum vibration absorbers

We investigate analytically and experimentally the effects of Coulomb friction on the performance of centrifugal pendulum vibration absorbers (CPVAs), which are used to reduce torsional vibrations in rotating machinery. The analysis is based on perturbation methods applied to the nonlinear equations of motion for a rotor subjected to an engine order applied torque and equipped with a circular path CPVA with viscous and Coulomb damping. The experimental work is based on quantifying parameters for the damping model using free vibration measurements with a viscous and Coulomb damping identification scheme that is enhanced to better handle measurement noise, and running tests for steady-state operation under a range of loading conditions. The level of Coulomb damping is varied by adjusting the friction of the absorber connection bearing. Good agreement is found between the analytical predictions and the experimental data. It is shown that the absorber sticks up to a level of excitation that allows it to release, after which the Coulomb damping acts in the expected manner, resulting in lowered response amplitudes. The results obtained are of general use in assessing absorber performance when dry friction is present in absorber suspensions.     

用地面模拟试验识别卫星太阳电池翼伸展运动的各种阻尼参数

本文利用Lagnrange方程建立了卫星太阳电池翼伸展运动的数学模型.这是一个包含摩擦阻尼、结构阻尼和空气阻尼参数的非线性微分方程组.在用气垫消除了重力影响但存在空气阻尼的地面环境中,对实尺太阳电池翼模型进行了伸展运动的模拟试验并用频闪摄影技术记录了运动的时间历程.再利用最小p乘优化方法识别出该系统的各个阻尼参数.通过对系统运动方程进行数值积分求得的太阳电池翼的伸展运动历程与实测值相当吻合.取空气阻尼等于零,得出了太阳电池翼在无空气阻尼的太空中的伸展运动规律.     

伴随变阻尼作用的干摩擦下的车辆系统非线性动力学分析

对分段线性阻尼和干摩擦共同作用下的车辆悬挂系统进行了非线性动力学分析研究，阐述了判定系统周期运动稳定性的理论方法；利用数值模拟方法分析了具有不同阻尼参数组合的系统对简谐激励的振动响应，并分析了由干摩擦引起的粘-滑振动行为．结果表明：提高摩擦力对抑制响应有利，但车辆系统在低速下运行时会出现复杂的粘-滑振动，轮轨之间产生较大的瞬时刚性冲击；而通过增加轮对与侧架的弹性悬挂可以有效减弱这种瞬时刚性冲击．     

机器人关节非线性摩擦模型关键参数反求

机器人关节非线性摩擦的准确描述对提高机器人轨迹精度、定位精度及其可靠性等具有重要理论意义和科学价值. 然而, 机器人关节通常包含电机、减速器、驱动器和传感器, 是一个复杂的机电耦合系统, 随服役时间及工况的变化, 机器人关节的摩擦参数也存在显著时变效应, 难以准确描述, 造成轨迹精度下降, 为机器人后期精度维护造成巨大困难. 因此, 本文定量评价了摩擦参数对机器人输出力矩的影响, 提出考虑时变效应的机器人关节非线性摩擦参数反求方法. 首先, 建立机器人关节一般非线性摩擦模型. 设计机器人关节恒速跟踪实验, 通过卡尔曼滤波对实验采集的数据进行处理, 进而建立关节速度和驱动电机电流之间的关系, 完成关节一般非线性摩擦模型建立. 其次, 择取非线性摩擦模型关键参数. 建立包含非线性摩擦的机器人动力学模型, 基于激励轨迹计算各关节力矩, 并对其开展灵敏度分析, 择取对关节力矩灵敏性较高的摩擦参数. 再次, 建立关节输出力矩和摩擦参数一一对应的数据集. 基于实际工况构建摩擦参数取值空间, 采用最优拉丁超立方法对摩擦参数采样, 并将其代入机器人动力学模型计算出相应的力矩, 从而求得关节输出力矩和摩擦参数一一对应的数据集. 最后, 建立反问题神经网络并对其进行训练, 实现非线性摩擦模型关键参数反求, 并进行验证. 研究结果表明关节非线性摩擦的准确描述减小了机器人低速运动换向时摩擦力矩突变对机器人轨迹的影响, 显著提升了机器人轨迹精度.      

Neural network solution to an inverse problem associated with the eigenvalues of the Stokes operator

A numerical method, based on the design of two artificial neural networks, is presented in order to approximate the viscosity and density features of fluids from the eigenvalues of the Stokes operator. The finite element method is used to solve the direct problem by training a first artificial neural network. A nonlinear map of eigenvalues of the Stokes operator as a function of the viscosity and density of the fluid under study is then obtained. This relationship is later inverted and refined by training a second artificial neural network, solving the aforementioned inverse problem. Numerical examples are presented in order to show the effectiveness and the limitations of this methodology.     

A high-order discontinuous Galerkin method for the incompressible Navier-Stokes equations on arbitrary grids

A high-order discontinuous Galerkin (DG) method is proposed in this work for solving the two-dimensional steady and unsteady incompressible Navier-Stokes (INS) equations written in conservative form on arbitrary grids. In order to construct the interface inviscid fluxes both in the continuity and in the momentum equations, an artificial compressibility term has been added to the continuity equation for relaxing the incompressibility constraint. Then, as the hyperbolic nature of the INS equations has been recovered, the local Lax-Friedrichs (LLF) flux, which was previously developed in the context of hyperbolic conservation laws, is applied to discretize the inviscid term. Unlike the traditional artificial compressibility method, in this work, the artificial compressibility is introduced only for the construction of the inviscid numerical fluxes; therefore, a consistent discretization of the INS equations is obtained, irrespective of the amount of artificial compressibility used. What is more, as the LLF flux can be obtained directly and straightforward, no numerical iteration for solving an exact Riemann problem is entailed in our method. The viscous term is discretized by the direct DG method, which was developed based on the weak formulation of the scalar diffusion problems on structured grids. The performance and the accuracy of the method are demonstrated by computing a number of benchmark test cases, including both steady and unsteady incompressible flow problems. Due to its simplicity in implementation, our method provides an attractive alternative for solving the INS equations on arbitrary grids.     

基于神经网络的差分方程快速求解方法

近年来, 人工神经网络(artificial?neural?networks, ANN), 尤其是深度神经网络(deep?neural?networks, DNN)由于其在异构平台上的高计算效率与对高维复杂系统的拟合能力而成为一种在数值计算领域具有广阔前景的新方法. 在偏微分方程数值求解中, 大规模线性方程组的求解通常是耗时最长的步骤之一, 因此, 采用神经网络方法求解线性方程组成为了一种值得期待的新思路. 但是, 深度神经网络的直接预测仍在数值精度方面仍有明显的不足, 成为其在数值计算领域广泛应用的瓶颈之一. 为打破这一限制, 本文提出了一种结合残差网络结构与校正迭代方法的求解算法. 其中, 残差网络结构解决了深度网络模型的网络退化与梯度消失等问题, 将网络的损失降低至经典网络模型的1/5000; 修正迭代的方法采用同一网络模型对预测解的反复校正, 将预测解的残差下降至迭代前的10?5倍. 为验证该方法的有效性与通用性, 本文将该方法与有限差分法结合, 对热传导方程与伯格方程进行了求解. 数值结果表明, 本文所提出的算法对于规模大于1000的方程组具有10倍以上的加速效果, 且数值误差低于二阶差分格式的离散误差.      

Nonlinear dynamics of an inclined beam subjected to a moving load

In this paper, the nonlinear dynamic response of an inclined pinned-pinned beam with a constant cross section, finite length subjected to a concentrated vertical force traveling with a constant velocity is investigated. The study is focused on the mode summation method and also on frequency analysis of the governing PDEs equations of motion. Furthermore, the steady-state response is studied by applying the multiple scales method. The nonlinear response of the beam is obtained by solving two coupled nonlinear PDEs governing equations of planar motion for both longitudinal and transverse oscillations of the beam. The dynamic magnification factor and normalized time histories of mid-pint of the beam are obtained for various load velocity ratios and the outcome results have been illustrated and compared to the results with those obtained from traditional linear solution. The appropriate parametric study considering the effects of the linear viscous damping, the velocity of the traveling load, beam inclination angle under zero or nonzero axial load are carried out to capture the influence of the effect of large deflections caused by stretching effects due to the beam’s immovable ends. It was seen that quadratic nonlinearity renders the softening effect on the dynamic response of the beam under the act of traveling load. Also in the case where the object leaves the inclined beam, its planar motion path is derived and the targeting accuracy is investigated and compared with those from the rigid solution assumption. Moreover, the stability analysis of steady-state response for the modes equations having quadratic nonlinearity was carried out and it was observed from the frequency response curves that for the considered parameters in the case of internal-external primary resonance, both saturation phenomenon and jump phenomenon can be predicted for the longitudinal excitation.     

确定非线性隔振装置参数的一种方案

对单自由度刚度可调非线性隔振模型进行数学建模仿真和实验，对于待训练的神经网络模型，将数值仿真结果作为学习样本，将实验结果作为检验样本，训练成功的神经网络模型具有较好的内延能力和一定的外推能力，在设定非线性隔振装置要求的隔振效能参数后，可以给出隔振装置对应的刚度和阻尼参数，从而为设计非线性隔振装置提供了一种方案。