倾斜油水两相流水为连续相的逆塞流及过渡流型非线性特性

在倾斜油水两相流模拟井装置上,利用自制的纵向多极阵列电导传感器,测量得到了66组不同倾斜角度的油水两相流电导波动信号.采用Hurst指数和吸引子形态描述方法提取了波动信号的非线性特征,通过研究Hurst指数、吸引子形态特征量与流型之间关系,发现倾斜油水两相流电导波动信号的Hurst指数介于0.6～0.9之间,4阶吸引子形态特征量对流型转化敏感.结果表明,4阶吸引子形态特征量的组合能够有效识别水为连续相的逆塞流型和过渡流型.     

多类型特征量的油水两相流流型识别算法

本文以收敛型微通道中油水两相流的流型识别为对象,将高速摄像法与神经网络算法相结合,提出了高效的、可视化的、智能化的两相流流型识别方法。该方法采用了包含图像纹理参数和流型无量纲参数的多类型特征量,更精准区分6种流型的不同特点,流型识别的收敛速度和准确率更高。BP神经网络的识别率为92. 5%,Elman神经网络的识别率为93. 7%,Elman神经网络在收敛速度与准确率方面优于BP神经网络。     

起伏振动状态下水平管内两相流多尺度熵分析

将振动装置与气液两相流实验回路相结合,对起伏振动状态下水平通道内气液两相流进行实验研究,观察记录流型并此采集压差波动信号。同时采用多尺度熵方法对102种流动条件6种流型的压差波动信号进行分析。研究结果表明:起伏振动状态下水平管内主要流型包括泡状流、珠状流、炮弹流、弹状-波状流、沸腾波状流及环状流。利用小尺度下样本熵的变化速率特征可以明显区分水平通道内气液两相流型,而大尺度下样本熵的变化特征可以反映不同流型的动力学特性。     

基于Choi-Williams分析与神经网络的两相流流型识别北大核心CSCD

提出一种基于Choi-Williams分析和神经网络的流型识别方法。使用阵列电导传感器获取垂直上升管道气液两相流流型信息,并将多元测量数据进行去噪和降维处理,进一步采用Choi-Williams分析将其转换为时频谱图,并构建数据集。分别搭建CNN、VGG-16和ResNet-18网络模型,将不同流型的时频谱图作为网络输入进行训练、测试。识别结果表明,Choi-Williams分析可以有效提取不同流型信号的特征,3种网络均具有较高的识别精度,其中ResNet-18网络准确率最高,平均流型识别率达99.4%。     

基于流体诱发管道振动的气液两相流型识别

通过对不同流型的气液两相流流经管道时诱发的管道振动特性的实验研究,提出了基于流体诱发振动的非接触式在线两相流流型识别新方法.通过安装于测试管道外壁的振动传感器测量不同流型下气液两相流诱发的振动信号;采用小波包分析提取了表征流型变化的振动信号能量特征向量;以能量特征向量作为模型输入,建立了概率神经网络模型用于识别分层流、...     

基于小波包与Elman神经网络的气力提升装置流型识别技术研究

气力提升装置流型对气液流动特性及提升系统性能均有很大的影响,但由于气液两相交界面形态以及截面含气率动态变化、气液两相速度复杂难测等原因,致使提升管流型亦交替变化且不易识别。针对这一难题,提出了基于小波包分析与Elman神经网络的流型辨识策略:利用小波包分析方法提取提升装置压差信号各频带能量特征值,借助Elman神经网络辨识技术,以各频带能量为Elman网络输入变量,以流型为输出变量,通过对Elman神经网络进行大量数据训练。从而对提升系统流型进行辨识。实验结果表明,该方法对流型辨识精度达到了92.6%,比BP网络高6.5%,能有效对提升管流型进行辨识。     

气液两相流流型振荡诱发制冷循环不稳定性的实验研究

研究利用变制冷剂流量制冷循环的实验平台结合流动显示方法发现:1)蒸发器出口的制冷剂气液两相流流型存在过热蒸汽流和雾状流两种型式,二者之间存在一个转变过渡区域,此时,两种流型闪动交替出现,回气温度随机性波动;2)膨胀阀出口(即蒸发器入口)的制冷剂气液两相流的流型存在液一气分相流、泡一气分相流两种型式,二者之间也存在一个转变过渡区,此时两种流型交替出现,制冷循环周期性振荡,蒸发温度、回气温度、排气温度等参数周期性波动:且振荡周期随着制冷循环制冷剂流量的增大而缩短.     

基于电容层析成像系统测量信号稀疏性的两相流流型辨识

提出了基于电容层析成像(ECT)测量电容信号稀疏性的两相流流型辨识算法,该算法首先使用所有流型对应的归一化测量电容值信号构建一个过完备字典,并将待辨识样本通过该过完备字典进行稀疏表示,使其具有稀疏性并满足稀疏重构的基本要求,然后以压缩感知的正交匹配追踪(OMP)算法求取各标准样本对应于完备样本集的稀疏解,最后根据待辨识样本与标准样本稀疏解之间的线性相关程度进行流型辨识。使用该方法对5种典型的两相流流型识别进行了仿真及实验研究,结果表明:该方法的流型正确识别率均高于98%。     

基于电容层析成像技术重构图像的两相流流型识别

针对工业多相流流型识别的需求,在过程层析成像技术的研究基础上,将重构图像信息即像素分布进行简单有效地分析处理,利用二维最大熵阈值分割技术及遗传算法优化的神经网络分类器对重构的图像进行处理以实现流型识别。将该方法在气固(空气/煤粉)两相流气力输送平台上对3种典型气固流型进行验证,实验结果表明:50组测试样本中识别正确率为94.7%。     

基于多元经验模态分解与卷积神经网络的气液两相流流型识别

提出了一种基于多元经验模态分解(MEMD)与卷积神经网络(CNN)的垂直管道气液两相流流型识别方法。该方法基于数字化电阻层析成像(ERT)系统采集的测量数据,预处理后进行MEMD分析,通过求取各分量与原始信号的皮尔逊相关系数选取本征模函数(IMFs)并求解Hilbert边际谱,提取Hilbert边际谱的标准差与均值作为卷积神经网络(CNN)输入以识别流型。结果表明,该方法能够有效识别泡状流、弹状流、段塞流,平均识别准确率可达96.43%。     

基于MO-PLP-ELM及电容层析成像的两相流流型辨识

提出一种基于多目标优化并行感知器的极限学习机(MO-PLP-ELM)及电容层析成像(ECT)技术的两相流流型辨识算法。首先,为保证样本具有代表性,采用随机思想生成7类流型的训练及测试样本集;其次,对样本模型的电容数据归一化处理;最后,采用MO-PLP-ELM算法进行流型辨识,并与常用的BP神经网络、支持向量机、极限学习机及并行感知器改进极限学习机算法进行比较,结果表明,提出的MO-PLP-ELM算法其辨识率明显高于其它算法,平均辨识率达96.1%。     

基于边界不变特征的扩展目标识别方法

提出一种新的以边界不变矩作为识别特征,运用BP网络识别扩展目标的方法。首次通过详细的理论证明和实验分析,揭示了离散边界不变矩不再具有严格的比例不变性,而位移和旋转不变性保持相对稳定,并对该不变矩作为识别特征的误差进行了深入分析,给出了正确计算边界不变矩的途径。在此基础上,以该边界不变矩作为识别特征,输入BP网络,采用合理的网络结构,实现对发生位移、旋转和尺度变化的扩展目标的识别。边界不变特征的引入,减少了数据运算量,实验结果表明,识别率达到95.9%。     

槽式孔板差压信号的高阶统计量特征

针对气液两相流流过槽式孔板时产生的差压信号呈现出的非高斯特性,提出了一种基于高阶谱技术的信号分析方法.通过对槽式孔板两端差压信号进行的双谱分析,提取了不同流量下信号的非高斯特征.以信号的双谱幅度最大值和双谱切片幅度最大值作为特征参数,对不同流型、流量下差压信号偏离高斯分布的程度进行了定量估计.结果表明,提取的特征值不仅可作为气液两相流流型转变的判据,而且还为进一步优化两相流流量计的设计提供了参考.     

Research on automatic identification system of tobacco diseases

In order to improve recognition accuracy of tobacco diseases, an identification method based on multi-feature and genetic algorithms optimizing BP neural network was proposed. First, Otsu method was used to obtain disease location information and GrabCut function was initialized for extracting diseased area effectively. Second, colour moments, disease contour and GLCM were used to get colour, multi-contour and texture features. Once again, BP neural network was optimized by genetic algorithm, and the optimal initial weights and thresholds were obtained, which shortened the training time and improved the accuracy of disease identification. Finally, BP neural network model for tobacco diseases diagnosis was established with the mobile client as input and the user services as output. The field experiment showed that the method could diagnose eight types of tobacco diseases effectively and automatically. The average recognition accuracy rate of selected tobacco diseases was about 92.5%.     

Prediction and Control of Internal Condensing Flows in the Experimental Context of their Inlet Condition Sensitivities

The reported experimental results involve fully condensing flows of pure FC-72 vapor on a horizontal condensing surface (316 stainless steel) which is the bottom surface (wall) of a rectangular cross-section duct of 2 mm height, 15 mm width, and 1 m length. The sides and top of the duct are made of clear plastic. The experimental system in which this condenser is used is able to control steady-in-the-mean (termed quasi-steady) mass flow rate, exit pressure, and wall cooling conditions. It has been found that, with the condenser mean (time averaged) inlet mass flow rate, exit pressure, and wall cooling condition held fixed at steady values, there is a very strong sensitivity to high amplitude pressure fluctuations and flow rate pulsations at the condenser inlet. This sensitivity often significantly alters the condenser mean inlet pressure, pressure drop, local heat transfer rates (>200% increase at certain locations), and the condensing flow morphology. These effects are representative of fluctuations/pulsations that are typically encountered in applications but are either not accounted for or are not detected. The effects of imposed fluctuations/pulsations, as opposed to cases involving negligible imposed fluctuations/pulsations, are dependent on the amplitude and the frequency content of the imposed fluctuations and this is discussed in a separate paper. A significant upstream annular regime portion of the reported shear/pressure driven fully condensing flows operate under conditions where the laboratory??s transverse gravity effects are negligible and, therefore, the identified sensitivity phenomenon is highly relevant to zero- or micro-gravity conditions. The micro-gravity relevance of this sensitivity for the annular regime phenomenon is currently also being demonstrated with the help of computations and simulations.     

基于层析成像的气液两相流相关流量测量方法

对电阻层析成像技术和图像的小波纹理特征进行了研究,提出一种基于层析成像的气液两相流相关流量测量方法,实现了液相流量的准确测量。利用电阻层析成像技术和相关算法对不同泡型下的相含率、渡越时间进行检测,得到气相流量;利用小波分析提取出层析成像的流型纹理特征;从而基于BP神经网络建立不同泡型下的气液两相流的相关流量测量模型。实验结果表明,液体流量的测量精度可以达到3%以内。     

Dominant Flow Structures in Gas–Solid Fluidized Beds Using Time and Frequency Domains Analyses

Time series analysis techniques in time domain and average cycle frequency were applied to characterize bubbling fluidization. The experiments were carried out in a laboratory scale fluidized bed, operated under ambient conditions and various sizes of particles, measurement heights, and different superficial gas velocities. It was found that a minimum in average cycle frequency and flatness and a shift of skewness from negative to positive against velocity correspond to shift from macrostructures and finer structures of the flow rather than transition velocity from the bubbling to turbulent regime. The power spectrum estimation of the measured pressure fluctuations shows that the peak dominant frequency of the pressure fluctuations is about 1.5–2.5 Hz which is corresponding to the macrostructures of the bed. Accordingly, the onset of turbulent fluidization regime was detected through standard deviation analysis. It was shown that the simple analysis techniques still have interesting information about hydrodynamics of fluidization and they can accurately estimate transition between dominant flow structures of a gas–solid fluidized bed.     

融合时频域特征的舰船识别方法及实验研究

文章提出了一种融合舰船辐射噪声时频域特征的识别方法,将舰船辐射噪声的线谱特征和线性预测倒谱特征作为输入,分别利用反向传播（Back Propagation,BP）神经网络进行训练、降维及初步判别,并采用加权投票方式,引入置信度算法和拒判机制实现决策级融合识别。实验结果表明,对比基于舰船单一特征的识别方法,利用舰船辐射噪声时频域特征的互补性进行融合识别,减小了单一识别方法误判对总识别率的影响,具有较强的鲁棒性,可有效提高对目标的识别率。     

Effect of Dune Formation on Pressure Drop in Horizontal Pneumatic Conveying System

Pneumatic conveying is widely used in industries handling large amount of granular materials to transport the solid particles; however, the process is energy intensive as an instability of flow sets in the transportation line even in the dilute regime, causing large fluctuations in the line pressure drop, the reason of which is not clearly understood. Here, we investigate, both by experiments and by using numerical simulations, the instability transition regimes and identify the reasons of the fluctuations observed in the line pressure drop in a horizontal pneumatic transport system operating at near-saltation conditions. It is observed that the increase in the pressure drop (immediately after the saltation) is accompanied by the formation of distinct dunes. It is also observed that the line pressure drop depends on the axial location of the dune and shows large fluctuations in the regime where the dunes are unstable. Results obtained from the numerical simulations suggest that the increase in the line pressure drop in the presence of dunes is essentially due to the shear stresses at the dune surface which are larger than that for the flows in clean pipe.     

基于足底压力传感器的不控制减重比例下步态相位识别

下肢外骨骼机器人是帮助下肢运动功能障碍患者步行训练的新手段，能够减轻医护人员的劳动强度，它常采用减重方式完成辅助训练。然而，对于地面行走减重外骨骼机器人系统而言，其减重比例随步态及穿戴方式变化而变化，因此不控制减重比例下的步态相位识别具有重要意义。通过搭建基于Arduino Mega2560板卡和单侧鞋内8个薄膜式压力传感器的足底压力采集系统，分别采集了正常行走、不控制减重比例减重带减重状态下行走时的足底压力信息，并采用神经网络算法进行步态相位识别。结果表明：在减重状态下行走与正常行走相比，左右脚压力值均出现明显下降且两侧具有对称性；足底每个压力传感器处的压力减小比例不同；采用神经网络算法对正常行走时步态相位总体识别率达到96.8%，对减重行走时步态相位总体识别率达到94.8%。研究结果表明该足底压力采集系统可以有效测量减重行走时的足底压力，为在地面不控制减重比例下减重带减重的外骨骼机器人控制策略的制定提供一定支持。