A Digital Special-Purpose Signal Processor for Two-Phase Flow Real-Time Analysis

254

IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. IM-26, NO. 3, SEPTEMBER 1977

When phase goes from second quadrant to third quadrant, R3 will be set by the phase state logic and simultaneously reset by the XP pulse. The internal operation of

the flip-flop is such that both of its outputs will simultaneously go true for the duration of the XP pulse and revert to normal, correct output after XP ends. The Assertion output will be correct during and after the XP pulse, but the Negation output will be incorrect for the duration of the XP pulse. There will be similar behavior of R2. To

A

avoid difficulty, do not make use of the R2 or R3 outputs. ACKNOWLEDGMENT The slip detector was built as part of an investigation performed under contract to Lockheed Missiles and Space Company, Inc. Permission to publish this description is gratefully acknowledged. The slip detector has been invaluable to an experimental study of PLD threshold.

Digital Special-Purpose Signal Processor for Two-Phase Flow Real-Time Analysis JACK-GERARD POSTAIRE AND JEAN-MICHEL FITREMANN

Abstract-In the statistical study of two-phase flow real-time control of experiments is often needed. Furthermnore, in industrial processes involving boiling units control or two-phase gas-liquid transport control, real-time signal processing is required to check the status of two-phase flows. An instrument is described for the measurement of the fundamental characteristic parameters of two-phase flows. The measurement technique employs extensively high speed digital computation. It is based on a discrimination of the probe output signal into two signals, representing, respectively, a "bubble" flow and a slug flow. The parameters are measured on averaging times ranging from 10 ms to 104 s. Measurement errors and limitations of this high-resolution measurement technique are pointed out. Its utility in evaluating the fundamental statistical parameters of two-phase flows is demonstrated.

SIGNAL OUTPUT

WAKE

BUBBLE FLOW

LOCAL VOID FRACTION PROBE "TAYLOR 8UB8LE"

FLOW

Fig. 1. Typical visual aspect of horizontal slug-flow pattern at moderate velocity and large Bond number.

1) bubble-flow regime where a swarm of small bubbles are entrained by the flow of liquid; INTRODUCTION 2) slug-flow regime where the bubble-flow alternates with the displacement of long cavities here called IN SPITE of the great level of sophistication "Taylor bubbles;" 1 in reached the design of general purpose minicompuannular-mist-flow regime where the liquid is en3) ters available on the market, special purpose machines are trained by the high-velocity gas-flow as a droplet still needed for some real-time digital signal processing. in cloud the core of the duct and as a thin wavy film Particularly, in the statistical study of two-phase gasinner boundary. its along liquid flow, real-time control of experiments or industrial It is of prime importance to study and/or control the equipment are needed. transitions between these regimes and, furthermore, the Indeed, when a mixture of gas and liquid flows through different "void fractions," i.e., the probabilities that some a duct (a situation which arises in boiling units, in twochosen the point inside is in a bubble, a "Taylor flow phase gas-liquid transport, in oil and chemical engineering, in bubble" or the are the liquid, basic dependant variables etc., ...) three regimes are likely to occur, they are as folof such flows A [1]. visual aspect of typical slug flow is lows: shown in Fig. 1. The first goal of the experimenter is to separate "bubManuscript received May 10, 1977.

J. G. Postaire was with the Department of Electrical Engineering, Faculty of Applied Sciences, University of Sherbrooke, Sherbrooke, Canada. He is now with the Centre d'Automatique, Universite des Sciences et Techniques de Lille, Lille, France. J. M. Fitremann is with the Laboratoire de Mcanique Experimentale des Fluides, Universite de Paris, 91405 Orsay, France.

sutfi-

1 This alternance of Taylor bubbles and bubble flow occurs at ciently high Bond number pgD2/o, where p is the density, D the diameter of the pipe, a the surface tension constant and g is the gravity con-

stant.

-

POSTAIRE AND FITREMANN: DIGITAL SPECIAL-PURPOSE SIGNAL PROCESSOR LEVEL GAS

. LIM ID

1i

F

S1

r

11

OUTPUT SIGNAL

.

.l

I!i

i4

255

-

j -j

j

.! .,I.hi.l

IH

iii

L

TIME

Fig. 2. Recording of the output of a local phase sensor in typical slug flow showing a "Taylor bubble" and a bubbly wake (optical void fraction probe and trigger unit from AID, Grenoble, France).

ble" signals from "Taylor bubble" signals at the output of local void fraction probe [2]. Such a probe gives a low output when its extremity dips in the liquid and a high output when it is in the gas. Therefore, a typical slug-flow signal consists of a succession of high frequency random square waves ("bubble" signal) and long duration high levels ("Taylor bubble" signal) as shown in the record of Fig. 2. The size of the smallest bubbles that can be detected depends on the probe size. Typical minimum diameter and corresponding velocity are about 0.0002 m and 10 m/s, so that 2 X 10-5 S is a likely duration of a high level of the bubble signal. A high sampling rate is thus needed if high-speed flows are analyzed with small probes. In software realizations, the time spent in memory housekeeping added to the time needed for arithmetic and logical operations are too long to develop a computer based real-time processing. The special-purpose machine that is described in this paper is designed to determine the two following data: 1) the time T1 spent by the probe in the bubbles relative to the total time T; 2) the time T2 spent by the probe in the "Taylor bubbles" relative to the total time; 3) both signals should also be recovered separately for further statistical analysis. The results of this processing are expected to be used in further sophisticated real-time treatment. Thus this machine is designed as a preprocessor for a digital comput-

Fig. 3. Basic organization of the discriminating unit.

a

er.

The

parts:

processor

is organized internally in two main

a discriminating unit which separates the short duration high levels representing "bubbles" from the long duration ones representing "Taylor bubbles" into the respective components which are available on two separate outputs; 2) a computation unit which determines the mean time spent by the probe respectively in the "bubbles" and in the "Taylor bubbles."

1)

THE DISCRIMINATING UNIT

The discriminating unit is arranged around a dynamic shift register whose length N can be set from 256 cells to 4096 cells by means of front panel switches.

Fig. 4. Flow diagram of signal discrimination.

The unit is slaved to a precision digital clock using a 1-MHz quartz crystal resonator. The period of the clock pulses is set on front-panel switches. It ranges from 1 to 100 As. The clock can also be driven by an external frequency which is limited to 25 MHz. The voltage state of the signal to be processed is repeatedly "sampled" at the clock rate. The result is a series of highs and lows that shifts through the shift register which is used as a delay line. The clock does not have to be synchronous with the transitions of the sampled signal for the delayed signal to be a reasonable replica of the input signal. Transitions in any replica will differ from the original signal by no more than the period of one clock signal cycle. That interval is the resolution limit. By appropriate selection of both the length N of the shift register and the period T of the clock, the propagation delay T = N X T provided by the shift register can be set to 35 values ranging from 250 ,is to 0.4 s. This time delay allows a time counter to test the length of the high levels of the input signal before they reach the output of the shift register (cf. Fig. 3). When high levels are shorter than T, they flow from the input towards a first output channel Si. Nevertheless, whenever the duration of a high level exceeds the output of the shift register is connected to a second channel S2. The commutation happens just when the low-to-high transition corresponding to this level reaches the last cell r,

256

IEEE TRANSACTIONS ON INSTRUMENTATION AND

ME.4SUREMENT,

VOL.

IM-26, NO. 3, SEPTEMBER 1977

INPUT SIGNAL|

"TAYLOR 8U8BBLE'' SIGNAL (S2) i l

"BUBBLE" SIGNAL (I 1

l

l

Fig. 5. Typical input and resulting delayed output signals issued by the discriminating unit. (The time lag has the same length as the discriminating time threshold.) T

of the delay line. Furthermore, the switching device remains in this state until the following high-to-low transition of the delayed signal is shifted out (cf. Fig. 4). Thus the part of the input signal composed of high levels shorter than T is sent on channel S1 while the remaining information made of high levels longer than is sent on channel S2 as shown on Fig. 5. This discrimination with an adjustable threshold time length T allows the experimenter to separate automatically the part of the signal which represents the "bubble" flow from that which represents the "Taylor bubble" flow. The "bubble" and the "Taylor bubble" signals are used as inputs of the computation unit. Furthermore, they are available as outputs of the processor. Indeed, these two signals may prove useful for further analysis to determine various statistical properties of the "bubbles" flow and the "Taylor bubbles" flow such as frequency counts, probability distribution functions of time intervals, etc. The velocities can also be measured by dual probes techniques [31 if single "bubble" and/or "Taylor bubble" signals are available separately. T

THE COMPUTATION UNIT Let T1 and T2 be the times spent by the probe, respectively, in the "bubbles" and in the "Taylor bubbles" during a time interval T. The computation unit is designed to determine the two following time fractions: at =

T1/T

a2

=

T2/T.

FLAF.

Fig. 6. Block diagram of the circuits involved in the cornputation unit.

is determined by counting clock pulses on two counters, respectively gated by the two signals issued from the discriminating unit. The results are represented in a floating-point notation. The hardware is designed so that the mantissas are affected by the same exponent as the one previously used for the representation of time interval length (cf. Fig. 6). Thus the "bubble" and the "Taylor bubble" tine fractions are determined by dividing the two computed mantissas by 104. This is automatically done at the end of the time interval when the results are displayed on the front panel. The two mantissas, which range from 0 to 104 are counted in BDC code to be displayed on two five-digit LED displays including memory for temporary storage. A decimal point is set between the first and the second digit to divide by 104. A display-time control holds and displays the results until the end of the next analyzed time interval. Simultaneously, the 2 mantissas are computed in binary code. They are available on two 14 bits output words for parallel transfer from the analyzer to any digital computer. A flag continuously records the status of the analyzer. It will go high when the analyzer has data ready at its output connector. This output ready flag is cleared during the computation time. Thus the analyzer can be used interfaced to a computer like any other peripheral device. Because not all measurements demand the accuracy of 14 bits words, analog outputs are provided for other displays. The two digital numbers are converted to the equivalent dc voltages by two 8 bits digital-to-analog converters The output signals range from 0 to +10 V and can be used for continuous recording as suitable. The operator has the possibility of counting in an automatic repetitive mode. In this mode, all counters are cleared at the end of each analysis time interval and the analyzer immediately starts a new computation. This feature allows a continuous processing of the signal on adjacent time intervals and yields a continuous recording

This analysis is performed on time intervals which are measured by counting clock pulses. A presettable counter is used for this operation. It can count 104, 105, 106, 107, or 108 clock pulses so that the length (duration) of the time intervals can be selected by front panel switches, from 10 ms to 104 s. The number of counted clock pulses can be written in a floating point representation with a fixed mantissa: 104 and a variable exponent. This property is used for the hardware design as shown later. One-shot counting is initialized by pushing an INI push button when the analysis is wanted. The total duration of of a1 and a2. the "bubble" and the "Taylor bubble" presence on the The fundamental quantities required for theoretical probe over the whole time interval submitted to analysis statistical analysis are almost readily available from the

257

POSTAIRE AND FI'I'REMANN: DIGITAL SPECIAL-PUJRPOSE SIGNAL PROCESSOR

displays or the binary outputs. Indeed, knowing a1 and a2, one can easily determine the following time and void fractions:

aBl (Bubble void fraction) = al/(1 - al) aT (Taylor bubble void fraction) = a2 aL (Liquid time fraction) = 1 - a1 - a2 aG (Gas total time fraction) = al + a2TESTS ANI) PERFORMANCES The measurement error attributable to the computation unit never exceeds 10-4. Indeed only the 5 most significant decimal digits are preserved for displays on digital outputs which range from 0.0000 to 1.0000. Thus the fundamental factor affecting the precision and accuracy of the results is the resolution limit of the discriminating unit. It is inmportant to notice that the transitions of the signals on which are performed the computations differ from an ideal delayed input signal. As pointed out earlier, each transition of the outputs of thie discriminating unit is synchronous with the clock pulse immediately following the corresponding transition of this ideal delayed signal. However, on a large number of randomly distributed bubbles, the durations which are lost at the front edges are statistically recovered at the rear edges, so that the error affecting the total duration of the bubbles is about one clock period. The length of the averaging intervals are always greater or equal to 104 clock periods. Thus, the resulting error is about 10-4. This error, added to the limitation of the computation unit yields a total error of about 2 X 10-4. The accuracy of the processor has been tested with a signal generator controlled by a high-precision digital frequency meter. As expected, the error has been shown to always be less than 10-3. CONCLUSION The processor described in this paper is planned as a basic tool for the analysis of slug flow, the study of transition from bubble flow to slug flow regime and the investigation of many statistical properties of two-phase

flows. It can be used as a basic precision digital instrument which displays high resolution information for analysis of any two-phase flow. IHowever, for more sophisticated ap-

Fig. 7. General configuration of slug flow signal processing system.

plications, such as real-time control of the flow, the processor can be integrated into a computer system. Used as a digital input peripheral unit, the computer controlled processor can be programmed to set up and make real-time measurement automatically (cf. Fig. 7). In these two configurations, digital controls give precise selection of measurement parameters. The measurement technique which employs extensive use of digital computation provides accuracy, high resolution and reliability. In view of the simplicity of the technique, the ease of measurement of a1 and a2 and of the subsequent determination of the void fractions, this processor shall prove valuable in applications which require a rapid determination of two-phase flow characteristics, for instance in industrial process control involving boiling units or twophase gas-liquid transport. ACKNOWLEDGMENT The authors would like to thank J. J. Franchault for his contribution to the realization of the processor and Professor P. Vidal for the use of the facilities of the Centre d'Automatique. We are gratefull to the Compagnie Fran~aise des Petroles, ELF-Aquitaine and the Institut Fran~ais du Petrole for their support. REFERENCES [1]

[21

[3]

J. M. Fitremann, "Methode de calcul des ecoulements diphasiques petroliers en conduite," Revue de l'IFP, vol. XXX, no. 2, pp. 303-331, Mar.-Apr. 1975. J. M. Delhaye, "Les mesures en ecoulement diphasique," BIST, CEA, no. 197, Nov. 1974. J. M. Fitremann, C. Guilpin, and J. G. Postaire, I.A.H.R. SHF 1976 on two phase-flow and cavitation in power generation systems, Lecture n°18.