For many years interest in multiphase flow has been increasing in heat science and flow mechanics. Multiphase flow appears in many technique domains. It is a very common phenomenon in chemical, agricultural, food, cooling and power industries, and environmental or processing engineering. The techniques described here are as valid in gas–liquid as in gas–solid flow. The recognition of two-phase flow patterns is essential in process engineering. This kind of flow pattern is the most important criterion that enables correct calculation of processes of the heat and mass transfer. Researches of fluidization are very significant from the practical point of view, because they enable, for example, minimization of energy consumption in pneumatic transport or of selection and keeping proper parameters in fluidic power boilers.
For the proper running of most apparatus, in which two-phase flow appears, generation of strictly specified two-phase flow pattern is required. Investigation of the process in industrial conditions requires the use of an objective method of identification, which is based on measurable features, which are characteristic for the specified two-phase flow. As far as single-phase flow is concerned, definition of distribution of velocity is enough but in two-phase flow, the definition of phase concentration is also very important (1). There has been significant research in to the recognition of phase concentration and various measurement techniques have been developed.
Most of these measurement methods are based on a visual estimation of the flow structure, which is performed by hand using a comparison scale to draw conclusions. Unfortunately, the results of this approach are very subjective. In this connection it is necessary to develop a universal, precise, quick, and cheap method of measurement. Digital processing image methods have this advantage. In comparison to commonly used methods of recognition of two-phase flow structures the distinguishing marks of digital methods are fast time of response, repeating results, and elimination of the subjective human factor for direct measurements. This method can be used in a range of hostile environments where human intervention is not possible (2).
The research work aimed at developing techniques for the recognition of two-phase flow patterns with the use of dynamic image analysis. In the case of gas–liquid flow the continuous phase was coloured methyl alcohol and the dispersed phase was air (this choice was made because of combination of air and alcohol gives better contrast between two phases then air and water). In the case of gas–solid flow the continuous phase was air, and the solid phase were spheres of 2.5 mm diameter and 800 kg/m3 density. These two kinds of flow have been realized in flat channels with rectangular profile. Proper regulation of intensity of airflow (in both cases) and proper dosing of liquid and solid phases allows realization of each type of flow.
The first step was to record each individual flow pattern using a video camera, in this case a SVHS C JVC GR-S 707 video camera. The structures obtained were recorded with the shutter speed of 1/1000 s. The acquisitioned object was illuminated with a halogen spotlight. A light dispersion filter was placed, between the lamp and the channel. After recording the dynamic changes of the flow on videotape, digitalization was started to change the analogue form of the image (magnetic record) to a digital form. With the use of a frame grabber video card and ADOBE PREMIERE® software on a PC computer in WINDOWS® 98 environment, this step in the image processing was realized. Image preprocessing included several improvements such as contrast and brightness adjustment for each realization of flow structures. In addition the colour image was converted in to monochromatic. After the internal MPEG compression was made by the video card, realizations were saved as AVI files. The recorded sequences have 24-bit depth of colours, and thanks to a low compression coefficient, they were very good quality.
The main part of the digital image processing research was realized with the use of our own software. In this program recorded changes of phase concentration distribution were put under analysis. The program allows the flow structure analysed at any one part of the film and any time. By using a suitable size of zone – probe (there is possibility of adjusting the area of the probe and choosing between single and multiple probe) – analysis of the most interesting part of flow was possible. The result of this analysis was a two- or three-dimensional matrix of data (it depends on the probe used), which represents the sequence of grey level, changes. In the area of the probe, the mean grey level counting, have been made. Obtained values of grey level were included in range of 0–255, where 0 is white and 255 is black. A zero value can be treated as 100 per cent gas-phase (but fully transparent). On the other hand a value of 255 is a pure solid phase (black) in the fluidization process. The liquid phase has much lower grey level values than solid phase (3).
The dynamics of change, while the two-phase flow persists, has been signified as the function of grey level rate in time domain. The unit was frames. Knowing the frequency rate of the video camera recording, which is 25 Hz, this rate was converted in to seconds and the new received unit was 1 frame = 0.04 s. The results of the analysis have been transformed in to graphical form.
Figure 18.1 represents classified individual structures of the gas–liquid flow, Fig. 18.2 shows the sizes and situations of the probes used. Realizations were the subject of an examination with four types of zone probes – the superficial probe [Fig. 18.2(a)], the elongated probe [Fig. 18.2(b)], the spot probe [Fig. 18.2(c)], and the multiple probe [Fig. 18.2(d)]. Each of these probes allows different kinds of flow pattern to be obtained. In this case characteristics for each probe has been made.
The size of the elongated and spot probes have been adjusted in order to get the object like a bubble that appears and disappears in the area of the probe during the movement from one frame to another. The distance that the bubble passes through is ten pixels long. That is why the width of these probes is set up on ten pixels. With the superficial probe the largest area of the channel has been analysed. Considering this probe, each acquired graph has the mean gray level placed in the lowest level among all other types of probes for the structure taken. This level was chosen as a basis from which positive deviation took place (upwards in gray level values).
The individual feature of this probe is that it has the lowest, among all other, amplitude of realizations obtained. The other characteristic is that it has the lowest sensitivity. It is the result of the biggest averaging area. The big measurement area implies that the time of being single objects is long enough to collect many in this area. The time of acquisition of such objects was quite long. The objects, which flow in and out from the examined area, are hardly seen on the graph. Only big amounts of these objects or one big object can be seen. In case of small objects or big ones, consisting of many smaller ones (agglomerates), when their amount flowing in to the examined area is bigger than the amount flowing out, then the impulse is increasing. It is increasing until the time, when the proportions of the flow in and flow out are inverted. Then the impulse is decreasing. In the case of big homogeneous objects, such as plugs and bubbles the situation inverts.
To sum up, this probe gives a general view dealing with the specific flow structures. It describes changes of grey level value with global characteristics rather than local ones. This means that it is much easier to notice the presence of areas with greater density of gas phase in the form of various objects such as single bubbles or agglomerates than single objects.
The elongated probe, because of its similar shape but smaller area, is a more sensitive version of superficial probe. The increased sensitivity results from the fact that objects in the time of passing from one frame to another appear in the area of this probe and then disappear. The frames obtained by this probe feature greater amplitude in comparison to the superficial probe. The explanation of this fact is that the proportion of the area of the single object, like a bubble, to the area of the whole probe is much greater than in the case of the superficial probe. That is why the object is better visible on the frame. Thanks to the length of the probe, which is ten pixels, the density of the impulses on the graph is much higher, which means that even single objects, such as bubbles, are shown. Due to the much smaller length of the described probe, shorter time of acquisition of the object implies the reduced width of its impulse. In spite of the reduced length of this probe, the fact that its width remains unchanged allows the conclusion of the character of the examined structure. That is why the graphs of this and the previous probe cover each other (3).
The spot probe covers the smallest area among all others. In this connection, the relation of the area of the examined object with the area of the probe is the highest and is almost unity (this happens when a whole object covers the area of the probe). It influences the highest amplitudes most. In other words, the spot probe has the highest sensitivity. Because it has the same length as the elongated probe, the width of the generated impulse, in both cases, is the same. Additionally the grey level value obtained by the probe for a specific object corresponds with its real value. The reason of this is the fact that the probe does not average this value with the values of the other objects, which falls in to the probe area (the probe's size is slightly larger then the smallest possible object). On the basis of the value obtained we can conclude the type of examined object, but we can not conclude the type of flow structure. Three examples of the results are shown on the Fig. 18.3.
On the received courses the following kinds of objects have been observed:
The important feature related to the two-phase flow is the interfacial surface. The refraction of light is occurring on it. Because of this, the interfacial surface is recorded as a part of the image with the greatest grey level value. The fact that the objects (gas structures) are in reality convex, causes that observed line of the interfacial surface to be thick (and that is why it is dark).
Single bubbles are usually of small size. Because of this, their edges are close to each other. This is why these objects have a thick border that causes their darker colour from the enclosed liquid. The form of the graphical representation of the objects is the peek, which goes upwards – in the direction of ascending grey level, values. The superficial probe doesn't detect the single bubble – we have no clear signals on the trace recorded by this probe because the value, which represents this object, is getting lost among the averaged background and values of the other objects. The elongated probe gives the response in form of peeks with 0.7–1.5 amplitude in grey level scale. Only for SB1-flow is it possible to give the amplitude for the single bubbles. In the case of other kinds of flow, we are then counting the larger quantity of the objects crossing the area of the probe in one moment. The spot probe gives the clearest peeks, whose amplitude is placed in the 4–14 range of the grey level scale. It always gives the right information about the single bubble.
Agglomerates are conglomerations of the single bubbles, and therefore their view has a form of a larger object, but with the similar grey level value, as for the single bubble. Also the view of the impulse is similar – which is oriented upwards on the ascending grey level values. Agglomerates, treated as a single object, are properly represented on the traces of the superficial and elongated probes. The result of its analysis is an increasing impulse. The signal strength depends on the size of the agglomerate and its interfacial surface. The interfacial surface depends on the packing density of single bubbles. The spot probe gives us the view of the agglomerate that depends on which part of the agglomerate crosses the probe area. If the interfacial surface is crossing the area of the probe, the impulse increases, and if the body of the big bubble crosses the probe area, the impulse decreases.
Big bubble and the plug are big and homogenous objects, so their interfacial surface has no influence on the character of impulse. That is why the graphic representation on the signal is the clear peek directed downwards on the grey level values. The superficial probe records those objects as a wide decreasing impulse with big amplitude. The signal of these objects obtained by the elongated and spot probe is similar. They are narrow peeks, directed downwards, but they differ in amplitude: the elongated probe gives the impulses of amplitude from 8–10, and the spot probe from 25–30 in grey level value scale.
The above description and analysis of particular impulses on the given records, are just samples of recognition of the particular patterns, based on such characteristics as amplitude, frequency, width, and kind of impulse, which is shown below in Table 18.1. To show other characteristics, which can be used to describe individual structure, composition of courses given by three probes has been made for each concerned flow type.
The first conclusion that can be drawn from these compositions was the order of appearance of the traces: the lowest was the trace obtained by the superficial probe, next was the elongated probe trace, and the highest was the trace recorded by the spot probe. This configuration was characteristic for each of the structures concerned. Although they reveal other interesting characteristic these traces draw aside while the velocity of flow increases, and after the extreme values of the flow velocity have been reached these traces are brought together. This allowed the making of a juxtaposition of the mean grey level values for each realization, related to each individual structure, segregated by increasing velocity of the mixtures flow (see Fig. 18.4). After the analysis of juxtaposition on Fig. 18.4 confirmation has been made. For the spot probe we noticed that from the SB1- to P-structure, the graphs curve increases minimally which is caused by the sequence appearance of the greater number of the single bubbles and/or agglomerates. Their high grey level value is compensated by big bubbles, which appear in the next structures (their low grey level value). This is why the curve has a small slope in the examined range.
From P- to F2-structure we now have very high velocity of flow. It is the reason that the big bubbles have irregular shape and therefore superficial surface is bigger. It is also the reason of the larger part of agglomerates and big quantity of small bubbles, which are often rotating backwards (it causes them to remain longer in the probe area). This is why the curve in this area has a violent increase. After the next increase intensity of flow and creating A-structure, disappearing of objects such as single bubble, agglomerate, or foam has been observed.
However in the centre there persists a bright gas core, which is periodically drowned (because of immovability of the liquid phase). Now gas is creating a film of liquid on the channel walls. From the analysis of grey level value, we can speak about the replacement of the dispersed phase in to a continuous phase – and now low values are averaged by high values. This result is the decreasing slope of the curve (3).
The elongated probe better shows the amount of small objects such as agglomerates or single bubbles. This can be seen on DB2- and BB-structures where the curve gently increases. Also increased amounts of those objects in type P-structure is accented clearly. Whereas in case of F-structure we can see that breakdown and violent decreasing of curve until the A-structure took place.
Together with stream volume velocity of gas, more and more fine bubbles appear – dark objects. In case of SB1- and SB2-structure the image is similar, because the recording is made in chosen channel fragments. For SB2-structure, bubbles flow in three rows and the recording is made only in one of them, which equals the results with SB2. From DB1- to BB-structure we have progressive increasing of single bubbles and agglomerates. All of those objects are characterized by a greater optical density than the liquid. This happens because of a thicker interfacial surface line. Therefore the curve increases until big bubbles appear. When, in the flow structure, big bright bubbles begin to dominate, the compensation of dark small objects starts, and graph curve decreases a bit. This happen in the BB- to F-structure range. For the F-structure, where long and big bight objects appear in the probe area, a considerable drop of grey level value follows. For the A-structure, where there are long periods of gas core appearance, the curve on the graph falls violently (brightening).
Crossing of the spot probe curve and superficial and elongated probe curves in case of A-structure, comes from the spot probe characteristics. This probe is situated in the middle of the channel, and during the recording it detects only gas core or flooding moment. It does not record (and in reality does not average) the influence of the liquid film that remains on the channel walls. Superficial and elongated probes as insignificant blackening record this liquid film.
In the case of the gas–solid flow, except the digital image processing and analysis method, the analysis of random signals has been used. Theoretically, these two methods can work together for the best results in two-phase flow pattern recognition. Therefore the research in this area was oriented to verify if the connection of these two methods would give a rational effect. In practice many of the physical phenomena, which can be described with mathematical dependences, has determined character. However there are a great deal of physical phenomena, which have non-determined character, and can not be described with mathematical dependences. There is no possibility of predicting the value of such a signal in any moment in the future. These signals are random by nature, therefore the result of each particular observation is only one of the infinitive numbers of possible results, which can happen. That is why these phenomena have to be described with averaged statistical characteristics (4).
So far, the description of the phenomena, which accompanies the two-phase flow, is the result of simplification of a complicated character of the process. The next reason for the simplification is limited experimental verification of the results, because analog measurement methods enable the comparison of stabilized processes only. We made an assumption that fluidization is a stationary process, because nonstationary processes preclude the extraction of a sufficient number of time functions (series or random functions). And because characteristics of unsteady random processes are time functions, which can be specified only by averaging actual values in random functions ensamble which form the process, we are not able to acquire exact measurement of these characteristics. This circumstance makes the development of practical measurement and analysis methods difficult (5).
To describe the main qualities of random signals we apply, among others, the statistic functions such as probability, density, and autocorrelation. Probability density function describes the properties of the process in value (amplitude) domain. Autocorrelation function gives us the information of the process in the time domain.
Probability density function of the random signal determines the probability of an event, which means that the values of the signal at any moment is included in the particular range. The main goal of the probability density of the physical signal measurement is to establish the statistical laws concerning the distribution of its actual values. Nevertheless this function can be applied to distinguishing the harmonic signal from the random signal. Besides distribution and shape of the probability density, fiction allows the experienced specialist to reveal the non-linearity of physical effects.
To simplify, the autocorrelation function of the random signal determines the overall dependence of the actual value to the value in other moments. The main application of the autocorrelation function of the random signal is the examination which consists of determining to what degree the value of a process at a particular moment has impact upon the value of that process in a moment in the future. In the case of determined signal, the autocorrelation function ‘lasts’ for all time displacements. And autocorrelation function of the random signal approaches zero for big values of displacement. In that case autocorrelation function is a good instrument for determined processes detection, which can be ‘masked’ by random noise (5).
The experiment was two-way guided. Two methods used and combined at the end. First of all the fluidization process was investigated by visual methods and presented as a change of gray value function [Fig. 18.5(a)]. From the data obtained this way, graphs of probability density distribution [Fig. 18.5(b)] and autocorrelation [Fig. 18.5(c)] were made. The results were elaborated for various two-phase gas–solid flow patterns.
The image that is generated by video camera is two-dimensional. Therefore the events, which happen near the back wall of the channel, can not be recorded by objective of the camera. In connection with that, there is proposal of realizing further research in the apparatus with more flat channel to be able to avoid the processes, which happen deep inside the bed. This will allow focusing exactly on the events that happen next to the front wall of the channel. In almost all cases of the autocorrelation function, the graph presents a curve that resembles a sine curve [see Fig. 18.5(c)]. In all other cases the curves demonstrate repeated cyclic formations. This suggests determinated character of the fluidization process, at least in case of blister and gas trap fluidization. In recapitulate there is the possibility of determination of the flow structures with the use of an image recognition method. In addition, by combining the analysis of gray level and random signal analysis methods, we can achieve satisfactory results.
On the grounds of realization the research work and detailed analysis of theoretical basis of the digital dynamic image process and analysis. And by employing the known fundamentals of two-phase flow pattern recognition using the random signals analysis, it was found that each of these methods and their combination as well, is an appropriate and applicable research tool in the area we talk over. In the multiphased research, where there is the need to predict the specification of the flow structures, the methods presented meet these needs. These methods are very precise and sensitive, which testifies the recording of even the least perturbations, for example the non-uniform lighting, which took place during the research work.
For the matter of further research in the field of two-phase flow, it is possible to determine interfacial surface. During the analysis of acquired results, unequivocally identify property have been noticed. It characterizes with the highest gray level values. Taking this in to consideration we can calculate the number of pixels, where brightness is situated in the former specified range characteristic for the interfacial surface. Then we can count the pixels' surface, that is, we can calculate the interfacial surface. There is another variant possible too. By analysing the given flow pattern stochastically, we can create a probability density graph. From it, we can obtain the fraction of pixels whose brightness is contained in the interfacial surface range. In connection with above the digital image processing and analysis method is the research tool with great possibilities, which is precise and modifiable for the particular applications. This method can be support for the two-phase flow research.
(1) R. Ulbrich, Identification of two-phase gas–liquid flow, qualifying as assistant professor work, Opole Technical University, 1989.
(2) G. P. Celata, P. Di Marco, and R. K. Shah, Two-Phase Flow Modelling and Experimentation 1999, Pisa 1999.
(3) M. Masiukiewicz and D. Zając, The analysis of two-phase flow gas–liquid with use of digital image processing, Thesis, Opole Technical University, 2000.
(4) N. Szmolke, Method of appreciation of the fluidised bed structure, PAN Chemical Engineering Institute, Gliwice 1997.
(3) S. Anweiler, The analysis of two-phase flow gas–solid mixture with use of image recognition, Diploma paper, Opole Technical University, 2000.
R Ulbrich, M Krótkiewicz, N Szmolke, S Anweiler, M Masiukiewicz, and D Zając
Heat Technique and Process Engineering Department, Opole Technical University, POLAND
© With Authors 2002