Comparing temporal characteristics of slug flow from tomography measurements and video observations

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

“…With this procedure the liquid level time series are obtained from the video recordings of multiphase flows. The code for the extraction of the liquid level time series from the RGBpixelcolumns over time is available as Jupyter Notebook in Olbrich et al (2021c).…”

“…The model was trained and tested using python version 3.7, tensorflow version 2.3 and Keras version 2.4.3, see Chollet et al (2015) and Abadi et al (2015). The implementation of the model, the code for the training of the model as Jupyter Notebook, and the weights of the trained model are available in Olbrich et al (2021c).…”

“…Furthermore, for foamy/bubbly regions in the flow, the segmentation often includes foam into the liquid phase, leading to overestimated absolute values for the liquid level time series. This overestimation was investigated in detail in Olbrich et al (2021b), where hand-labelled liquid level time series of slug flow have been compared to reference parameters of a conventional tomography measurement system.…”

“…This non-dimensional parameter has a range of [0, 1] with respect to the inner pipe diameter 𝐷 and is hereinafter referred to as the liquid level time series at a fixed position 𝑥, denoted by ℎ 𝐿 (𝑡) (Olbrich et al, 2021a;Schmelter et al, 2021a). It reveals the dynamics of the spatial distribution of the two phases in the pipe and can therefore reliably indicate slugs, waves or other liquid structures, similarly to the hold-up parameter, see Olbrich et al (2021b). Typically, for analyses of hold-up or liquid level time series, the conventional length and time scales of slug flow are calculated by simple thresholding procedures, see Zhao et al (2015), Baba et al (2018), Schmelter et al (2021a) and Olbrich et al (2020Olbrich et al ( , 2021b and Fig.…”

“…It reveals the dynamics of the spatial distribution of the two phases in the pipe and can therefore reliably indicate slugs, waves or other liquid structures, similarly to the hold-up parameter, see Olbrich et al (2021b). Typically, for analyses of hold-up or liquid level time series, the conventional length and time scales of slug flow are calculated by simple thresholding procedures, see Zhao et al (2015), Baba et al (2018), Schmelter et al (2021a) and Olbrich et al (2020Olbrich et al ( , 2021b and Fig. 1.…”