Online Two-Phase Flow Pattern Identification Based on Deep Learning and Time-Domain Signal of Coriolis Flowmeter

Gas-liquid two-phase flow is commonly encountered in both nature and industrial processes. Identifying the flow pattern plays a crucial role in gaining a deeper understanding of the flow structure and ensuring the safe and stable operation of industrial production. Traditional machine learning-based methods for flow pattern identification generally overlook the influence of time-sequenced factors and involve excessively complicated computations, making them difficult for online recognition. To address these challenges, this paper proposes a flow pattern identification method based on deep learning and time-domain signals derived from Coriolis flowmeter. Deep learning neural networks can automatically extract features from data, which can avoid the limitations of manual feature extraction and demonstrate a certain level of robustness. The extracted features are computationally simple and suitable for online flow pattern identification. The results indicate that this method yields excellent flow pattern identification results, with an offline training set recognition accuracy of 97.9 %, a test set accuracy of 95.3 %, and an online recognition accuracy of 93.3 %.