基于止回阀声音信号的气液两相流流型智能辨识方法研究

发布时间：2018-07-14 09:31

【摘要】：气液两相流广泛存在于自然界和工业生产中,在石油的输送管道中更为常见。石油作为一种重要能源,在国民经济和工业发展中都占有重要地位,因此对于石油输送管道的安全监测就显得尤为重要。石油管道的安全监测从本质上讲就是检测管道内的气液两相流动状态,识别出故障信息并及时发出预警。石油管道内的气液两相流型是油气管道输送安全监测的重要参数之一,由于气液两相流的复杂性和易燃易爆的危险性,至今流型识别还没有得到很好的解决。本文提出了一种基于止回阀内声音信号,辨识气液两相流型的新方法。参照石油输送的实际工况,设计制作了适用于实验研究的止回阀,搭建了可视化的实验平台,利用水和空气模拟产生了液相流、气相流、弹状流和混状流四种不同的流型。采集了这四种流型经过止回阀时产生的声音信号。通过设置合理的采样时间和频率,对获得的信号进行了采样,提高了信号处理效率。根据气液两相流的非线性特性,选用希尔伯特-黄变换(HHT)作为声音信号的处理方法。声音信号经过采样之后,利用EMD去噪算法降低了采样信号的噪声,并对去噪后的声音信号进行了重构。再对重构信号进行EMD和希尔伯特谱分析,得到了信号的IMF分量、希尔伯特谱和希尔伯特边际谱等能量特征。研究发现,第3个IMF分量及其希尔伯特谱的能量和重构信号边际谱的能量能够反映不同流型的特征。因此,选择这三项特征数据绘制了二维和三维流型图。实验表明,利用止回阀内声音信号特征所绘制的流型图,辨识准确率在98.1%以上。为了进一步完善本方法,实现对气液两相流型的智能辨识。本文选择了人工神经网络作为流型智能辨识的分类器,设计了误差反向传播(BP)神经网络模型。设定用于绘制流型图的三种能量特征作为输入,四种流型为输出,选择多组数据对神经网络进行了训练。学习后的神经网络对流型的辨识准确率达到了 97.5%,实现了对流型的智能识别。本文提出的基于止回阀声音信号的气液两相流流型识别方法具有较好的推广应用前景。
[Abstract]:Gas-liquid two-phase flow is widely used in natural and industrial production, and is more common in the pipeline of petroleum. As an important source of energy, petroleum is important in both national economy and industrial development. Therefore, the safety monitoring of oil pipeline is particularly important. The gas liquid two phase flow status in the pipeline is detected and the fault information is identified and early warning is issued. The gas liquid two-phase flow pattern in the oil pipeline is one of the important parameters for the safety monitoring of oil and gas pipeline transportation. Because of the complexity of the gas and liquid two phase flow and the danger of flammable and explosive, the flow pattern recognition has not been well solved. A new method for identifying gas-liquid two phase flow patterns based on the sound signals in the check valve is proposed. According to the actual working conditions of the oil transport, a check valve suitable for experimental research is designed and made. A visual experimental platform is built. Four different flow patterns are produced by simulation of water and air, and the flow patterns of gas phase flow, elastic flow and mixed flow are collected. These four flow patterns are produced by the check valve. By setting the reasonable sampling time and frequency, the obtained signal is sampled and the signal processing efficiency is improved. According to the nonlinear characteristics of the gas-liquid two-phase flow, the Hilbert yellow transform (HHT) is selected as the processing method of the sound signal. After the sound signal is sampled, The EMD denoising algorithm is used to reduce the noise of the sampled signal and reconstruct the noise after the denoising. Then the reconstructed signal is analyzed by EMD and Hilbert spectrum, and the energy characteristics such as the IMF component of the signal, the Hilbert spectrum and the Hilbert marginal spectrum are obtained. The energy and weight of the third IMF components and their Hilbert spectrum are found. The energy of the marginal spectrum of the signal can reflect the characteristics of different flow patterns. Therefore, the three feature data are selected to draw two-dimensional and three-dimensional flow patterns. The experiment shows that the identification accuracy is above 98.1% by using the flow pattern drawing of the sound signal characteristics in the check valve. In order to further complete the method, the intelligence of the gas-liquid two-phase flow pattern is realized. Identification. In this paper, artificial neural network is selected as classifier of flow pattern intelligent identification, and the error back propagation (BP) neural network model is designed. The three energy characteristics used to draw flow pattern are set as input, four flow patterns are output, and multiple sets of data are selected to train the neural network. After learning, the neural network convection model is identified. The recognition accuracy is 97.5%, and the intelligent recognition of the convective type is realized. The method of identifying the flow pattern of gas-liquid two phase flow based on the sound signal of the check valve has a good prospect for popularization and application.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：O359

【相似文献】