ROLE AND CHARACTERIZATION OF LEAKS UNDER TRANSIENT CONDITIONS

Kai Wah Tang¹, Bruno Brunone² and Bryan Karney³

¹Graduate Student, Department of Civil Engineering, University of Toronto, Toronto, M5S 1A4, Ontario Canada; E-mail: kwtang@civ.utoronto.ca

²Assoc. Prof. of Hydr., Department of Water and Struct. Engineering, University of Perugia, via G. Duranti 93, 06125 Perugia, Italy;

E-mail: brunone@unipg.it

³Professor, Department of Civil Engineering, University of Toronto, Toronto, M5S 1A4, Ontario Canada; PH (416) 978-7776; Fax: (416) 978-7776; E-mail: karney@civ.utoronto.ca

Abstract: The significance and impact of leaks in a pipeline system creates new opportunities of leak detection.  In essence, the concept is to use the pressure response from a transient event to locate and size a leak.  Previously, Brunone (1999) determined both the location and size of a leak on the basis of the pressure trace during a transient event at a measurement section on the basis of the well-known properties of pressure waves.  More recently, formal inverse transient algorithms have been developed.  The goal in this study is to see if the genetic inverse transient procedure can correctly locate and size a leak in “blind tests” of a simple and a more complicated system.  More specifically, the pressure signal at the downstream end of the system as well as the basic pipe properties will be fed to the inverse procedure to see if the predicted existence, location and magnitude of the leak can be accurately determined.  The paper reviews the results of the blind calibration procedures as well as summarizing the key background required to understand these developments.  The significance of this study data to the later quality problem, and particularly to the danger of contamination of the pipe contents, is given special emphasis.

1    INTRODUCTION

Leaks are a serious and challenging concern to operators of water distribution systems.  Not only do they allow treated water to escape uselessly into the environment, they dissipate a considerable amount of energy as well, thereby increasing both pumping costs and the environmental footprint of a water supply project. Moreover, leaks create a two-way connection between the inside of a pipe and its immediate environment, thus introducing the possibility of drawing contaminated water into a pipe under suitable conditions.

This particular leak detection experiment utilizes a recently-developed genetic algorithm (GA) inverse calibration computer program for pipe networks. This software is a combination of a genetic algorithm processor (GAP) and a transient analysis program (TransAM) created to perform inverse calibration using data collected from transient events. This combination of computer algorithm and software lends itself readily to the task at hand. In this study, we would like to calibrate models of pipe network systems. In some senses, a pipe network is similar to a living organism. It has physical characteristics, specific needs and behaves with a degree of predictability. Thus, one can model the physical characteristics of the system such as pipe diameters, friction values, wavespeed, valve sizes (leaks), pumps and other devices as specific genes in an individual “artificial” organism.

2    GENETIC ALGORITHM

In order to carry out the calibration of the leak test system, we can populate our pipe network “biosphere” with many organisms (a large gene pool) with a large diversity of genes representing the physical characteristics of the system.  More specifically, the genetic algorithm is applied to a pipe network in 3 major steps. In step 1, gene typing, all relevant physical values calibrated for are encoded into binary numbers. The binary number system is a logical choice for gene encoding since a binary number consists of a number of bits that is either a 1 or 0. A binary 1 can represent that a particular gene is switched on and a binary 0 is a switched off gene. Therefore an individual contains a number of genes, each a binary number representing a physical characteristic value. Since each characteristic can take on a number of values, the algorithm requires a lower and an upper bound for each parameter.

In step 2, Inverse Calibration, the individual’s genes are converted to their actual physical parameter values. Since each individual contains all the physical characteristics of the system, they can be used as data for a transient analysis computer model. The current process differs from traditional techniques in that thousands of pieces of information are collected via high-speed pressure transducers. These data can are used to calibrate the numerical model.

In the final stage, before the cycle repeats, the individuals of a population are ranked in order of their ability to accurately predict the response of the system to a particular transient event. The evaluation scheme compares pressure measurements to predicted responses. In this way, we can numerically model the system with each individual in the population and compare the computer generated pressures with the measured values and calculate an error function. It is this error value that determines the superiority of the individual (minimum error represents best individual). Once the “best” individual is determined for any particular generation, the next generation is evolved. We allow the “better” individuals to reproduce and generate new additions to the gene pool. The reproduction process involves taking genes from two parents and combining them through cross-over operations. As in nature, mutation is allowed to further enhance the gene pool. The “best” individual from the previous generation is allowed to join the new generation through elitism. This ensures that all new generations will perform better or equal to past generations. The new generation is produced and the members are once again evaluated and ranked in order of superiority. After a sufficient number of generations, the best individuals have similar if not identical characteristics to the actual system.

3    LEAK TEST SYSTEMS

The first step in the leak detection procedure involved setting up a TransAM model of the test systems. The blind tests were carried out at the Hydraulics Laboratory of the University of Perugia (Italy) in a polyethylene pipe. The simple system was 352 m long, with internal diameter of 93.8 mm and wall thickness of 8.1 mm. The pipe is arranged in concentric circles (Figure 1) For the supply reservoir, an air vessel is used in which the pressure is kept automatically constant and equal to a prescribed value by varying the speed of three submerged pumps placed in the recycling reservoir. At the end section of the pipe, a hand-operated ball valve discharges to the atmosphere.

Fig. 1    The simple laboratory test apparatus for simulating leak(s)
The more complex system (Figure 2) was a “Y” configuration with a total length of 376.38 m.

Fig. 2  More complex laboratory test apparatus

The data supplied for these experiments were limited to pressure measurements at the upstream end and at the downstream valve. The discharge upstream of the leak is also recorded for the simple case. For the simple case, five leak samples numbered 2, 3, 4, 14 and 16 were supplied. In the complex case, two leak samples were supplied. Table 1 typifies the sample data collected from blind tests #3 and #4 (hr is head at upstream reservoir, Qu is flow rate upstream of leak and Hvalve is the head at downstream control valve).

Table 1  Typical data collected from simulated leak tests

4    INVERSE TRANSIENT ANALYSIS

For the simple case, a set of independent calibration runs utilizing only the data for the

Table 2  Final calibration results (simple case)

Note: Leak location is distance in meters from the downstream valve.

     Leak size is the steady state discharge from the leak indicating its size.

first pressure wave measured from each test sampled. The search for the leak will continue to focus on nodes 8 through 12. Table 2 summarizes the results of this search.

Although test#2′s result does not agree with the other tests, there appears to be a consensus on the location of the leak. The leak is independently found by test#3, test#4, test#14 and test#16 to be located at node 10 or 125 m from the downstream control valve. When the “blind fold” is taken off, the actual location of the leak was 128.5 meters from the downstream valve. Since the computer model of the test system was configured with 25 m lengths of pipes, we had expected to be a few meters off target. If the exact location of the leak is required, the computer model can be revised with shorter lengths of pipe. However, this is not critical since a difference of a few meters is not significant. Figure 3 illustrates the close approximation of this inverse leak detection procedure.

Fig. 3  Final leak detection/calibration pressure head at control valve of simple system

After analyzing test#2's calibration data, it was determined that the inverse procedure could not differentiate between locating the leak at nodes 9, 10, 11, or 12. This may be the result of test#2's low initial flow rate (1.25 l/s) which in turn generates much smaller transient pressure waves. The various uncertainties in the estimation of pipe friction values, pipe wavespeeeds, changes in flow around the leak site, and operational characteristics of the control valve can induce transients that may be more influential than the transients caused by not placing the leak at its actual location.

Fig. 4  Final leak detection pressure head at control valve of complex system

Table 3  Final calibration results (complex case)

        Note: Leak location is distance in meters downstream from the supply reservoir.

5    CONCLUSION

The genetic inverse transient calibration procedure tested in this paper can predict accurately the size and location of a leak in a simple as well as in a more complex pipeline. The independent calibrations carried out during this investigation located the leaks consistently. The genetic inverse transient calibration/leak detection also reliably estimated the size of the leak: as a matter of fact, orifice diameter for test no. 14 and 16 is larger than the one used in the other tests and discharge values reported in Table 2 confirm this. Even in the more complex Y-system the procedure did a good job in locating the leak, even though it was not obvious to the authors that this would be possible given the complexity and messiness of the recorded data.

The entire analysis required approximately two days each for the simple and complex system of computing on an entry-level personal computer with minimal user monitoring or input. Brunone(1999) has already demonstrated that measuring the travel time of the well-known properties of the transient pressure wave, one can quickly detect the location of the leak, whereas by means of GAP one needs some hours. However, the potential of the GAP to handle more complex pipe systems is significant and is demonstrated in Table 3.

However, the accuracy of predicting the exact location of the leak with the inverse method proposed here, or with most other methods, can be hampered by low flow velocities or small transient events.  Even in such cases the genetic inverse transient method determined the leak location within the vicinity of the actual site of the leak as demonstrated by test#2's results. Therefore, genetic inverse transient calibration can be a cost effective means of locating leaks in simple and complex pipe systems. In the future, the inverse leak detection method developed in this paper may be improved by exploring a variety of GA strategies.

References And Related Reading

Brunone, B. (1999), Transient Test-Based Technique for Leak Detection in Outfall Pipes.  J. of Water Resources Planning and Management, 125(5), 302-306.

Dandy, G. C., Simpson, A. R. and Murphy L. J. (1996). An improved genetic algorithm for pipe network optimization. Water Resources Research, 32(2), 449-458.

Goldberg, D. E. and Kuo, C. H. (1987). Genetic algorithms in pipeline optimization. J. Computing in Civ. Engrg., ASCE, 1(2), 128-141.

Halhal, D., Walters, G. A. and Savic, D. A. (1997). Water network rehabilitation with structured messy genetic algorithm. J. of Water Res. Planning and Management, 123(3), 137-146.

Holland, J. H. (1975). Adaptation in Natural and Artificial Systems. University of Michigan Press, Ann Arbor, Mich.

Karney, B.W., and McInnis, D. (1992). Efficient calculation of transient flow in simple pipe networks. Journal of Hydraulic Engineering, ASCE, Volume 118, No. 7, 1014-1030.

Liggett, J.A. and Chen, L-C. (1994a). Inverse transient Analysis in Pipe networks. J. Hydr. Engrg., ASCE, 120(8), 934-955.

Liggett, J.A. and Chen, L-C. (1994b). Monitoring water distribution systems: the inverse method as a tool for calibration and leak detection. In: Improving efficiency and reliability in water distribution systems, E. Cabrera and Antonio F. Vela, editors, Valencia, Spain, Kluwer Academic Publishers, 107-134.

Murphy, L. J. and Simpson, A. R. (1992). Pipe optimization using genetic algorithms. Research Report No. 93, Department of Civil Engineering, Univ. of Adelaide, Australia, June, pp. 95.

Murphy, L. J., Simpson, A. R. and Dandy, G. C. (1993). Design of a pipe network using genetic algorithms. Water, pp. 95.

Savic, D. A. and Walters, G. A. (1997). “Genetic algorithms for least-cost design of water distribution networks.” J. of Water Resources Planning and Management, 123(2), 67-77.

Simpson A. R., G. C. Dandy and L. J.Murphy (1994). Genetic algorithms compared to other techniques for pipe optimization. Journal of Water Resources Planning and Management, Vol. 120, No. 4, July/August.

Simpson, A. R., Murphy, L. J. and Dandy, G. C. (1993). Pipe network optimization using genetic algorithms. Proc., ASCE, Water Resources Planning and Management Special Conf., Seattle, Washington, May, 392-395.

Wylie, E.B., and Streeter, V.L. (1993). Fluid transients in systems. Prentice-Hall, Inc., Englewood Cliffs, N.J.