Multiphase flow measurement and data analytic based on multi-modal sensors
Accurate multiphase flow measurement is crucial in the energy industry. Over the past decades, separation of the multiphase flow into single-phase flows has been a standard method for measuring multiphase flowrate. However, in-situ, non-invasive, and real-time imaging and measuring the key parameters of multiphase flows remain a long-standing challenge. To tackle the challenge, this thesis first explores the feasibility of performing time-difference and frequency-difference imaging of multiphase flows with complex-valued electrical capacitance tomography (CVECT). The multiple measurement vector (MMV) model-based CVECT imaging algorithm is proposed to reconstruct conductivity and permittivity distribution simultaneously, and the alternating direction method of multipliers (ADMM) is applied to solve the multi-frequency image reconstruction problem. The proposed multiphase flow imaging approach is verified and benchmarked with widely adopted tomographic image reconstruction algorithms. Another focus of this thesis is multiphase flowrate estimation based on low-cost, multi-modal sensors. Machine learning (ML) has recently emerged as a powerful tool to deal with time series sensing data from multi-modal sensors. This thesis investigates three prevailing machine learning methods, i.e., deep neural network (DNN), support vector machine (SVM), and convolutional neural network (CNN), to estimate the flowrate of oil/gas/water three-phase flows based on the Venturi tube. The improvement of CNN with the combination of long-short term memory machine (LSTM) is made and a temporal convolution network (TCN) model is introduced to analyse the collected time series sensing data from the Venturi tube installed in a pilot-scale multiphase flow facility. Furthermore, a multi-modal approach for multiphase flowrate measurement is developed by combining the Venturi tube and a dual-plane ECT sensor. An improved TCN model is built to predict the multiphase flowrate with various data pre-processing methods. The results provide guidance on data pre-processing methods for multiphase flowrate measurement and suggest that the proposed combination of low-cost flow sensing techniques and machine learning can effectively translate the time series sensing data to achieve satisfactory flowrate measurement under various flow conditions.