Gas-liquid two-phase flow in rectangular channels involves thermal energy, chemical industry, and other fields. However, it is difficult to identify steady flow status and unify heat transfer model owing to the changeable mixing process and the complex industrial environment. In this work, the time series decomposition technique (i.e., variational mode decomposition) and neural network algorithm (i.e., long short-term memory) are employed to identify the steady flow pattern of the gas-liquid mixing process in vertical rectangular channels. The heat transfer model of a single bubble in a rectangular channel is established based on the geometrical characteristics of the bubble. Results demonstrate that a 30-dimensional feature vector is extracted from the conductivity time series for flow pattern recognition. The feature parameters of each intrinsic mode function are period, minimum, maximum, standard deviation, skewness, and kurtosis. Average accuracy of the new model is 94.58% with the highest at 95.83% and the proposed new model improves the recognition accuracy of steady flow pattern by 2.54–6.93% compared with other models adopted in this work. The precision of the new hybrid model is the highest while the number of decomposition layers for the conductivity time series is 5 and the bubble flow exhibits the largest heat transfer area under the same conditions. The proposed hybrid model can be used to enhance the identification accuracy of bubble flow in rectangular channels, contributing to curtailing unneeded flow patterns in industrial processes. The heat transfer model of a single bubble can provide a unified and convenient method for calculating industrial heat transfer. These findings provide a useful guideline for engineering design and energy-saving in the fields of thermal energy and chemical industry.
Read full abstract