Development of two-phase flow regime map for thermally stimulated flows using deep learning and image segmentation technique

Deep learning based liquid level extraction from video observations of gas–liquid flows

“…These accuracies are high compared to the reported training and test accuracies in between 89% and 98.7% of other successfully trained deep-learning based gas and liquid segmentation models, see e.g. Cerqueira and Paladino (2021), Yu et al (2021) and Ahmad et al (2022).…”

“…Furthermore, in Lin et al (2020), a deep learning model is used to predict different two-phase flow patterns in inclined pipes based on superficial velocities of the individual phases and inclination angles. Moreover, image processing techniques based on deep convolutional neural networks have been presented in Poletaev et al (2020), Haas et al (2020) and Cerqueira and Paladino (2021) for the detection, reconstruction, and analysis of gas bubbles in vertical pipes and micro-channels, for the recognition of flow patterns in micro pulsating heat pipes (Kamijima et al, 2020;Ahmad et al, 2022) as well as for the extraction of relevant water regions as pre-processing step for two-phase PIV-measurements in the field of ship and ocean engineering (Yu et al, 2021). For the quantification of separated and intermittent gas-liquid flow patterns in horizontal pipes, such as stratified wavy or slug flow, such advanced image processing techniques have not been reported.…”

Optimizing Enhanced Extended Topological Active Nets Model Using Parallel Processing

“…In conjunction with parallel processing, the integration of optimal clustering techniques into the EETAN model plays a crucial role in refining segmentation accuracy and addressing challenges associated with complex anatomical structures. Various clustering algorithms are explored to enhance the robustness of the segmentation process [4]. The k-means clustering is applied to group pixels based on intensity similarities.…”

Review of Machine Learning for Hydrodynamics, Transport, and Reactions in Multiphase Flows and Reactors

“…190 More recent investigations were extended to develop a novel DL method aided by an image segmentation technique for identification of thermally gas−liquid two-phase flow regimes including annular/semiannular, elongated plug, slug-plug, and bubbly flows. 191 To sum up, a major drawback is that most researchers have trained a pure ML model based on relatively limited experimental data sets, which leads to a risk of overfitting and thus reduces the model generalization capability. On the one hand, it could be an effective solution to establish a vast database for flow images and some recent works have presented an excellent example of how to improve this kind of weakness in identification of condensing two-phase flow patterns using CNN.…”

“…So, PCA can be used to extract the feature vector that still well represents the feature space for identifying various flow regimes. − Recently, researchers also used RNN approaches such as LSTM to predict the time-series chaotic dynamics and forecast two-phase flow regimes . More recent investigations were extended to develop a novel DL method aided by an image segmentation technique for identification of thermally gas–liquid two-phase flow regimes including annular/semiannular, elongated plug, slug-plug, and bubbly flows . To sum up, a major drawback is that most researchers have trained a pure ML model based on relatively limited experimental data sets, which leads to a risk of overfitting and thus reduces the model generalization capability.…”