Transient flow analysis linked to fast pressure disturbance monitored in pipe systems

Transcription

1 IOP Conference Series: Earth and Environmental Science Transient flow analysis linked to fast pressure disturbance monitored in pipe systems To cite this article: J L Kueny et al 2012 IOP Conf. Ser.: Earth Environ. Sci View the article online for updates and enhancements. Related content - Simultaneous transient operation of a high head hydro power plant and a storage pumping station in the same hydraulic scheme D M Bucur, G Dunca, M J Cervantes et al. - Penstock failure detection system at the "Valsan" hydro power plant A M Georgescu, C I Cooiu, N Alboiu et al. - Numerical and field tests of hydraulic transients at Piva power plant Z Giljen This content was downloaded from IP address on 12/03/2018 at 03:34

2 Transient flow analysis linked to fast pressure disturbance monitored in pipe systems J Kueny 1, M Lourenco 1 and J Ballester 2 1 Grenoble Institute of Technology (INPG-ENSE3) School Energy, Water and Environmental Sciences Rue de la Houille Blanche, F38402 Saint Martin d Hères, France 2 EDF DPIH DTG MPSH GRENOBLE 21, Avenue de l'europe BP, GRENOBLE CEDEX 09, FRANCE kueny.jl@wanadoo.fr Abstract. EDF Hydro Division has launched the RENOUVEAU program in order to increase performance and improve plant availability through anticipation. Due to this program, a large penstocks fleet is equipped with pressure transducers linked to a special monitoring system. Any significant disturbance of the pressure is captured in a snapshot and the waveform of the signal is stored and analyzed. During these transient states, variations in flow are unknown. In order to determine the structural impact of such overpressure occurring during complex transients conditions over the entire circuit, EDF DTG has asked ENSE 3 GRENOBLE to develop a code called ACHYL CF*. The input data of ACHYL CF are circuit topology and pressure boundaries conditions. This article provide a description of the computer code developed for modeling the transient flow in a pipe network using the signals from pressure transducers as boundary conditions. Different test cases will be presented, simulating real hydro power plants for which measured pressure signals are available. 1. Introduction 1.1. A description of the pressure monitoring system in EDF s penstocks Over 12% of electric energy produced by EDF in France comes from hydropower. The penstocks associated with such schemes are a set of 550 units. The mechanical stresses withstood by the components have increased over the years due to increasing production of ancillary services (supplementary functions performed by the same equipments that produce the fundamental electricity commodity, that every power producers are constrained to participate). Moreover, the flexibility of hydraulic production is associated with many transient conditions which lead to many water-hammer phenomena. In this context, the challenge of "hydraulic safety" is a major issue for EDF, which first launched an extensive penstocks renovation program [2] and secondly, the RENOUVEAU program [2] which aims to improve monitoring of numerous systems in each hydro power plant. Particularly, RENOUVEAU includes the deployment of the penstock monitoring systems discussed in this paper. These monitoring systems are referred to as "DSPCF : dispositifs de surveillance de la pression dans les conduites forcées meaning penstock pressure monitoring system. A DSPCF consist of a transient recorder connected to a pressure transducer located on the penstock at the inlet of the machines. Optionally, other parameters are used, such as sensors measuring the intake stroke of organs Published under licence by Ltd 1

3 such as inlet valve. This system is set to record during 10 minutes at a sampling frequency of 40 Hz (see Figure 1) for each significant transient. For EDF, the penstock pressure monitoring using DSPCF has two purposes: A purpose of short-term reliability of the system: the goal is to control the mechanical safety margin of the penstock structure and of the equipments, which are connected and work under pressure. Indeed, measuring the peak value of the pressure allows controlling the mechanical loading on the penstocks during all the running situations, with a low level of uncertainty, and therefore, the envelope of the service stress related to the pressure. A purpose of long-term safety: the goal is to carry out the safety predictive maintenance of all the systems causing overpressures when running. The signal processing [4] allows an early detection of any abnormal hydraulic behaviors. In addition, the large amount of available signals allows to build the learning sets and to use novelty detection to detect these hydraulic phenomena [5]. These possibilities give powerful tools for the monitoring of any drift in the tuning parameters of the operating laws which have an impact on transient behaviors [3]. It becomes thus possible to detect automatically the signature which heralds the defects affecting the hydraulic reliability of the equipment located in the high pressure part of a hydraulic machine. Figure 1. DSPCF scheme 1.2. Enforced control safety lines operating margin The penstock structure resistance depends on the loading applied on every section along the penstock length. At the origin, the design of the penstock is based on the hypothesis that the pipe will solely be subjected to "slow" transients relative to the reflection time T r of the pipe (T r = T/2 where T : period of piping system between turbines inlet and surge tank). In this case, the pressure in the pipe is only dependent on the upstream water inertia and the pressure loading profile along the pipe is then linearly lessening from the bottom to the top of the pipe (see Figure 2). But the transient states captured by the DSPCF show a lot of cases for which the waveform of the pressure signals has a signature of fast transients. In these cases, the propagation of the water hammer along the pipe leads to a pressure loading profile which diverges from the original design profile showed in Figure 2. 2

4 Figure 2. Water hammer profile in cases of fast transients or slow transients relatively to the reflecting time Tr of the penstock A few examples of fast transient signals which have been measured are shown in Figure to 6: Figure 3. Figure 4. Figure 5. Figure 6. Figure 3, 4, 5. Circuit filling at penstock valve opening, Figure 6. No load closing. During these transient states, variations in flow are unknown. In order to determine the structural impact over the entire circuit of such overpressure occurring during complex transients conditions, it is necessary to have a numerical code able to use directly pressure times measured series as boundaries conditions. In the following, the case of B1 plant is treated to illustrate the method used to reach this goal. 2. Data filtering process As explained previously, the pressure is sampled with a time step of 0.025s. Depending on the geometry of the pipe, the program will need a time step which can be one thousand times smaller than the 0.025s time step of the pressure signal. In this case, it is necessary to interpolate the data. However, the signal is often too noisy to be interpolated without filtering it first, as shown in figure 7. The data filtering method chosen will be presented in the next part of the article. 3

5 Figure 7. Pressure signal from the hydroelectric power plant of B1, with a zoom on the first peak 2.1. The choice of the filtering process The chosen filtering method for the non-stationary and non-linear signal is the Ensemble Empirical Mode Decomposition (EEMD)[6]. Designed for non-stationary and nonlinear time series, this decomposition algorithm has theoretically the advantage of enabling the extraction of the noise and the different hydro-acoustic modes of the pipe from the signals. The EEMD method is equivalent to a filter bank, since it allows the decomposition of a signal in a set of Intrinsic Mode Function (IMF) components, each of them representing one simple oscillatory mode embedded in the signal, i.e. representing one simple physical process (among the driving mechanisms). This method represents an alternative to the adaptive Empirical Mode Decomposition (EMD) method [7], which shows its limits in this case. Indeed, the use of EMD with these signals causes the appearance of mode-mixing [6] in the decomposition, due to the intermittence of pressure signals during transient states. In order to overcome this major drawback, the EEMD method uses the statistical properties of white noise, adding a finite white noise to original data and decomposing the signal obtained with the EMD method, repeating this process on a sufficient number of iterations and canceling the noise by averaging the results. This improvement in the method induces the use of two parameters: the ratio between the standard deviation of data and the standard deviation of the added noise, and the number of iterations or trials The choice of the EEMD method s parameters The ratio between the data and noise standard deviations has a direct effect on the appearance of mode-mixing, as shown by Figure, making clear that increasing this parameter induces a shifting of mode-mixing to lower IMF components, i.e. to low-frequency components. Even if the mode-mixing cannot be completely suppressed, reducing it or shifting it to lower frequencies allows a better understanding of the driving mechanisms. The Figure showing that the mode-mixing is absent from the five first IMF-components when r 0,7, the chosen ratio is: r=0,7. Besides, both parameters have an influence on the method error, as shown by the statistical rule: sr n N where N is the number of trials, ε s is the standard deviation of data, r is the ratio between the standard deviation of data and that of added noise, and ε n is the standard deviation of error, which is defined as the difference between input data and the sum of corresponding IMF components. The influence of the number of trials on error is confirmed by Figure, showing standard deviation of error as a function of the number of iterations, with r=0,7. In order to optimize mode-mixing, error and computation time, (1) 4

6 the chosen set of parameters is r=0,7 and N=100 for start. These parameters have to be confirmed later with a robustness study. Figure 8. Aspect of the IMF components when N=100 with the signal from B1. Upper left: r=0.3, upper right: r=0.5, lower left: r=0.7, lower right: r=0.9 Figure 9. Influence of the number of iterations on standard deviation of error when ε n =0.7 with the signal from B The filtered signal Starting from the decomposition done with the chosen parameters, the filtered signal is defined as the sum of the last IMF-components and the residue, beginning with the k th IMF, the (k-1) IMFcomponents being considered as noise. Figure shows a comparison between the input signal and output signal, which is defined as the sum of the fifth and the next components. In this case, the four first IMF components are considered as noise for a start. This comparison shows the potential of the 5

7 EEMD method for the filtering process, but the selection of the IMF-components to be considered as noise has to be confirmed later, depending on the noise robustness of the ACHYL CF calculation. Figure 10. Comparison between the input signal (red) and the output signal (blue) 3. ACHYL_CF software. A new software named ACHYL_CF has been developed in order to impose the pressure signal obtained on a transient recorder as boundary condition in a pipe network model. One of the objectives of this computer code, written in C++, is to represent all kind of pipe networks with or without closed pipe loops that can be found in hydro power plants. Figure 11. ACHYL_CF network elements with details of an Inert Branch attached to a node Ni The nodes permit to connect the different branches of the network. At each node is attach different Inert Branches which concentrate classical hydro singularities like shrinkage/enlargement, elbows, valves, pumps or turbines, pipe of small length, These singularities have small lengths in comparison with the pipes in the network and are considered as Inert Elements since the time of travel of a disturbance wave through it is much smaller than the computational time interval t. An Inert Branch can be connected to a Pipe Element or another Inert Branch or can have a free extremity. In this last case, a boundary condition singularity must be present in the corresponding Branch: a Lake (a free surface with an imposed altitude z=z(t)) or a Reservoir (a free surface with a variable altitude depending on the stored volume of liquid) or a dead leg (variable pressure but flow rate Q=0) or a pressure transducer (p imposed, Q variable). Two partial differential equations (the unsteady Euler equation with the conservation of mass) provide the 1D transient fluid model considered in the Pipe Elements : 6

8 dv 1 p dz f g V V 0 dt s ds 2D (2) 2 V 1dp a 0 s dt At time t, V and p are respectively the velocity and pressure at a point located at abscissa s in the pipe (altitude z). f is the friction coefficient of the pipe and D its diameter. If K is the bulk modulus of elasticity of the liquid, E the Young s modulus of elasticity of the pipe material of thickness e, its Poisson ratio, the wave celerity a can be evaluated: K / a KD 1 ( C ) E e with C = 5/4 μ for pipe rigidly anchored at both ends, C = 1 μ 2 for buried pipeline and C = 1 for pipe with expansion joints. These equations are solved using a classical method of characteristics (see ref. [8], [9], [10]). The main advantage of ACHYL_CF is offer the possibility to analyze flow transients only in the upper part of a hydro plant, i.e. the circuit between the upper reservoir and the location where a pressure transducer is installed without taking the lower part of the network into account, which is difficult to model precisely because of the changing feature in time of turbines depending on the control condition and their state of wear. 4. Results 4.1. B1 plant description 1/2 (3) (4) Figure 12. Test case: B1 plant and model The circuit of B1 plant is represented on figure 1. It has been schematized with 5 pipes, each having different lengths, different thickness and different diameters. Therefore their corresponding celerities are different. They vary from 856m/s to 1303m/s. The penstock consists in 4 pipes and several singularities (valves, bends, widening/shrinkage, junctions ). Its length is 2160m and the mean wave celerity is about 1200m/s. A typical period of the waves in this penstock will be about T=4L/a=7.2 s. In its lower part, the penstock splits in 2 parallel branches which have a length of 530 m. 7

9 The upper circuit consists of a pipe 470m long between the reservoir of Co, where the water level can be considered as constant and the 3-branche junction. The penstock is plugged on the lateral branch (90 ). The pipe continues up to the surge tank of Bo which is modelized as an axisymetric tank which respects the evolution of the volume of water with the height (see Figure 13). Figure 13. Surge Tank model 5. Initial test: Pelton injector closing A direct calculation has first been performed in order to control that the numerical model behaves correctly. The figures 3 to 7 show the results obtained with a quasi-instantaneous closure of the injector and a linear closure of the injector in dt=20 L/a, which corresponds to about 5 periods of the penstock wave. The typical water hammer head difference is observed for both cases, as well as the characteristic period in the penstock, which is about T=7.2s, as expected. In the upper circuit, the flow is not very affected by the injector closing time. The wave starts at end of the closing time of the injector and has a period of: T D L ST P4 U 2 d P4 g Since the obtained results fit with those expected, the model is validated. (5) Figure 14. Head at different nodes in penstock and water level in the surge tank 8

10 Figure 15. Head at different nodes in penstock and water level in the surge tank (zoom) Figure 16. Head at different nodes in upper circuit and water level in the surge tank Figure 17. Head and flow rate in pipe 0 close to the Pelton injector (Dout= equivalent diameter of the injector area) 5.1. Inverse calculation using pressure sensor at penstock outlet The pressure signal used for this test is presented on figure 5 and 7. In order to interpolate it at each discretization time t i the EEMD technique presented in 2 has been applied. The flowrate estimated by ACHYL is presented on figure 19. The figures 19 to 21 show the results obtained with this signal, i.e. the head evolution calculated at different nodes of the penstock and then the stress profile along the pipe during the transient. 9

11 Figure 18. Head and flow rate in pipe 4 (upper circuit) Figure 19. Pressure signal and estimated flowrate in transducer section Figure20. Head evolution at different nodes of the circuit 10

12 Figure 21. Head evolution in pipe 3 at penstock upper part (pipe n 3) and lower part (pipe n 0) Figure 22. Head evolution in upper circuit and surge tank water level 6. Conclusions Besides allowing to model fluid transients in pipes the classic way, these first results show that the code ACHYL_CF offers the possibility to determine the mechanical stress imposed on the penstock during transient states when using the inverse calculation. This major advantage makes this software a very useful tool to control the hydro power plant safety during critical events, using the only known data, i.e. the recorded pressure signal. To achieve this objective, another useful tool has been tested, which is the EEMD method allowing to filter the data while keeping the characteristics needed for the calculation of the hydraulic stress. References [1] Kocha M, Rouard P, Vailhen O and Roulet C When information technologies become an asset for maintenance of hydro power plants [2] Caveirole P, et al Diagnosis and rehabilitation of EDF's penstocks EDF France Hydro [3] Tourte E, Ballester J L, Andreucci S and Vailhen O 2009 Early failure detection in hydro 11

13 generation EDF France Hydro [4] Mocanu D, Ballester J L and Ioana C 2010 Non-stationary signal analysis in water pipes monitoring Signals, Systems and Computers (ASILOMAR) [5] Charbonnier S, D Urso G and Ballester J L 2010 On-line monitoring of transitory variations in hydraulic power plants pressure pipes CM&MFPT 2010 [6] Wu Z and Huang N E 2009 Advances in Adaptive Data Analysis [7] Huang N E, Shen Z, Logn S R, Wu M C, Shih H H, Zheng Q, Yen N C, Tung C C and Liu H H 1998 The empirical mode decomposition and the Hilbert spectrum for nonlinear and nonstationary time series analysis Proc. R. Soc. Lond [8] Watters G Z 1984 Analysis and Control of Unsteady Flow in Pipelines (United States: Butterworth Publisher) [9] Betamio A and E.Koellle 1992 Fluid Transients in Pipe Networks (United Kindom: Computational Mechanics Publications, Elsevier Applied Science) [10] Benjamin E W and Steeter V L 1978 Fluid Transients (United States: McGraw-Hill Inc.),