基于Choi-Williams分析与神经网络的两相流流型识别

Two-Phase Flow Pattern Identification Based on Choi-Williams Analysis and Neural Network

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%。

A flow pattern recognition method based on Choi-Williams analysis and neural network is proposed.The array conductivity sensor is used to obtain the flow pattern information of gas-liquid two-phase flow in vertical rising pipeline,and the multivariate measurement data are denoised and dimensionally reduced.Further,Choi-Williams analysis is used to convert it into time-frequency spectrogram,and the data set is constructed.CNN,VGG-16 and ResNet-18 network models are built respectively,and the time-frequency spectrograms of different flow patterns are used as network input for training and testing.The recognition results show that Choi-Williams analysis can effectively extract the characteristics of different flow pattern signals,and the three networks have high recognition accuracy,among which ResNet-18 network has the highest accuracy,with an average flow pattern recognition rate of 99.4%.