Online flowrate monitoring experiences at Hydro-Québec

Transcription

1 Online flowrate monitoring experiences at Hydro-Québec J Nicolle 1, G Proulx 2, L Martell 1 1-Hydro-Québec Research s Institute, 18 boul. Lionel Boulet, Varennes, Canada, J3X 1S1 2-Hydro-Québec ESP (Essais Spéciaux de Production); 5655 de Marseille, Montréal, Canada, H1N 1J4 Abstract. This paper describes recent experiences with an online flowrate monitoring system designed for hydraulic turbines. The angular index method used in this paper was developed by Hydro-Quebec with the goal of delivering an accurate and affordable system to monitor flowrate under all circumstances. Accuracy and applicability of the method is demonstrated using code accepted experimental results for a 2 MW Francis and a 11MW propeller turbine. The last section describes in details the requirements to apply the method. Introduction For a hydroelectric utility like Hydro-Québec (HQ) who relies almost exclusively on hydropower to produce its energy, knowledge of the flowrate and efficiency of its powerhouses is essential. It allows for production optimization at various levels in the company going from short term efficiency monitoring to long term reservoir management. While many tests on efficiency and flowrate are done internally to provide an accurate picture of the machine operation at a given moment, there always arise situations where additional data would be helpful. Because they are more sensitive to head changes, low head plants seem particularly affected by a number of factors such as interactions between groups, impact of aeration, tail water level, trash rack clogging and so on... Unfortunately, many of the testing procedures concerning low head plant involve large (and costly) experimental setup. A research project was thus started to address those difficulties. The goal of this project is to obtain an accurate and affordable mean to monitor the flowrate in a hydraulic turbine. In the long run, such system should provide experimental efficiency hill and improve operational practices. By pinpointing when the turbine is operated in a non-optimal regime, an improvement in the energy output for a given amount of water is expected. While it was designed with low head turbine in mind, it should also work on higher head machines. P α Figure 1 : Overview of the angular index measurement system on a propeller turbine

2 Online flowrate monitoring experiences at Hydro-Québec 2/7 9 th IGHEM Conference, Trondheim, Norway This paper is divided into four parts. The first one provides an overview of the method. Then, the results obtained on two very different powerhouses are presented. The last section provides experimental details for those who want to reproduce the method. 1. Description of the measurement system The angular index was first presented in [1] with a combination of numerical and experimental results. It is illustrated in figure 1 for an axial propeller turbine. As most relative methods, the angular index system is based on a pressure difference but, as its name says, it also integrates the angular position of the guide vane. For this system, the pressure difference is measured between the spiral casing inlet and a location behind the guide vane. Because of the downstream probes, the pressure difference is strongly correlated to the angular position of the guide vanes. This interaction is taken care of by the modification of the index coefficient Kop in eq. 1 which is not a constant but an expression. Calibration of the coefficient Kop should normally be provided by an absolute method. The variation of the index coefficient translates the fact that the flow inside the spiral casing evolves with the opening. A simple polynomial expression is usually enough to describe the coefficient evolution with the position. As an index method combines dynamic pressure and streamlines losses, two phenomena that evolve with the velocity square, the index exponent should be fixed at exactly.5. Modification of the exponent is no longer necessary because the slight changes in the flow distribution are taken into account by coefficient variation. Q op.5 = K * P (1) The system remains accurate with head fluctuations and is almost insensitive to adjacent unit operation. This is because the measured pressure difference is sizeable and the distributor acts as a filter that smears local phenomena happening at the inlet. The angular index system provides a description of the flow. To obtain the unit efficiency, it is also necessary to monitor the power output and the head. 2. Experimental investigation of a 2 MW Francis turbine A first system was installed on a 2 MW Francis turbine. The upstream pressure was taken at the very beginning of the spiral casing while the downstream probes were located in the small space between the guide vane and the runner (figure 2). The main reason for choosing this powerhouse was that an 8 paths acoustic transit time (ATT) flowmeter, previously installed in the penstock, was available as a reference. First a calibration of the system was needed. For this, the reference data from ATT flowmeter were used to determine the coefficient curve of Kop. The calibration points are presented in figure 3 where the calibration curve appears smooth and predictable. In this case a 2 rd order polynomial equation was found accurate enough to describe the coefficient evolution. Figure 2 : Angular index installation on a Francis turbine

3 Online flowrate monitoring experiences at Hydro-Québec 3/7 9 th IGHEM Conference, Trondheim, Norway Figure 3 : Calibration of the angular index coefficient Kop Once calibrated, the evaluation of the performance of the angular index could start. Figure 4 presents 1 days of monitoring for both the ATT flowmeter and angular index method. The period shown was not included in the calibration. The two data series were not discernable one from another which is quite remarkable for two independent methods. Statistical analysis was needed to obtain an objective evaluation of the accuracy of the angular index method and is presented in table 1. While both systems provided excellent flowrate monitoring, the angular index dispersion was half the one of the reference flowmeter. 13 Normalised flowrate Angular Index Sonic Days Figure 4 : 1 days of flowrate monitoring for the sonic and angular index method Table 1. Performance of the angular index for a 2 MW Francis turbine Flowrate dispersion RMS ATT flowmeter (reference) +1 m 3 /s.5 % Angular index +.5 m 3 /s.2 %

4 Online flowrate monitoring experiences at Hydro-Québec 9th IGHEM Conference, Trondheim, Norway 4/ Normalised flowrate Normalised flowrate At one point during the measurement, a constant offset between the two methods occurred. At first, it was thought that something went wrong with one of the pressure probes of the angular index but this was discarded after observing that all the pressure probes exhibited the same behavior. The origin of the discrepancy became apparent by plotting the predicted flowrate against the guide vane opening. Then, it appeared that the offset originated from the ATT flowmeter. In figure 5, we can see the two parallel curves for the ATT flowmeter that indicates that the measured reference flow was divided in two series. Distinction between the series was based on the date. The angular index was not affected. Also noticeable in this figure is the spread of the measurement around its mean. Again, it is clear that the angular index method has a very low dispersion and thus is able to predict accurately the flowrate Sonic series 1 1 Sonic series angular index Opening [deg] Opening [deg] Figure 5 : Flowrate against angular displacement for the ATT reference and angular index 3. Experimental investigation of a 11 MW propeller turbine A second experimental set of the angular index measuring system was installed on a low head 11 MW propeller. In this case, since there was no pressure probe adequately placed at the semi-spiral casing inlet, one of the outer WK probe was used instead. Because there is ample room between the guide vane and the runner, the downstream pressure probe was installed mid-way on the upper cover as shown in figure 1. The coefficient calibration came from previous measurements that were performed some years before with current meters. 1.1 angular index 25.5 m angular index 27.5 m angular index 3 m Flowmeters 25.5 Flowmeters 27.5m Flowmeters 3 m Qnorm Opening [%] Figure 6 : Comparison of the flowmeter and angular index at various head

5 Online flowrate monitoring experiences at Hydro-Québec 5/7 9 th IGHEM Conference, Trondheim, Norway The two main interests of testing the system in this powerhouse were to see if the concept was applicable to different types of turbines and to evaluate its performance with head fluctuations. To do this, the data from a two months production period were used. They were classified by head and then compared to the measurements obtained at 3 different heads on a similar unit. In figure 6, we can see that the angular index is able to follow head fluctuations. Also, despite the more complex nature of the flow distribution in the semi-spiral casing and distributor, the flowrate dispersion remained in the same range than with the Francis turbine as shown in table 2. Table 2. Performance of the angular index for a 11 MW propeller Flowrate dispersion RMS Angular index +.95 m 3 /s.25 % Even though there was no continuous reference method that was available for this powerhouse, a simple online monitoring system was designed to follow the turbine flowrate and efficiency. It also allowed to evaluate the effectiveness of the actual automatic power frequency regulation (PFR) procedures and to compare those with optimal ones. To simulate the optimal operating mode, the operator was asked to periodically adjust the opening of the unit. A clear indication of the optimal path (in red) on the efficiency hill was provided along with a band of +/- 3 m 3 /s. A capture of the system is presented in figure 7 while the operator did a good job of keeping the group near optimal operation. +/- 3 m 3 /s Figure 7 : Flow monitoring during optimal operation (head vs flowrate) Then, the automatic system took place. It has two different operating modes. In the first one, designed to keep the operation optimal, the automatic system was able to partially correct the required power with the head variation, but was usually struggling to lay on the optimal line (figure 8, left). With the second automatic mode, having less operational constraints, the correction just did not happen (figure 8, right) which led to non optimal operation. Clearly, while it is acknowledged that efficiency is not the only issue here, there might be some room for improvements. The online monitoring system allows just that by providing a clear picture of the turbine operation that would be difficult to obtain otherwise.

6 Online flowrate monitoring experiences at Hydro-Québec 6/7 9 th IGHEM Conference, Trondheim, Norway Figure 8 : Flow monitoring during automatic operation of the turbine 4. Description of experimental setup This section is provided for those who are interested in deploying a similar system. List of material pressure transducer angular encoder manifold bleed valves steel tubing & valves data acquisition & control system 4.1. Pressure probes The upstream (or high pressure) tap should be as representative as possible of the low velocity inlet conditions but should be located behind the trash rack. Ideally, probes located at the beginning of the spiral casing should be used. If not available, WK external probes can also provide useful information. The downstream (or low pressure) taps should be linked together with a manifold to average the spiral casing flow non uniformity (figure 9). We found that at least 4 probes were necessary to maintain a realistic average of the flow behind the guide vanes. They should be disposed about 9 degrees from each other in the distributor cover ring. The exact location of the downstream probes is not so important as long as it is the same for each sector. Different positions will lead to different calibrations but the method should remain effective. However, if possible, the wakes of the guide vanes and the lip of the spiral casing should be avoided. In some situations where the flow in the distributor is thought to be strongly non-uniform, more probes could be needed. An important point to mention is that pressure holes can usually be drilled without entering the spiral casing. It is very important to install automatic bleed valves on pressure probes to make sure that no air enters the system. They should be allowed to work after each startup and on a daily basis to keep the system operational. For the pressure sensor itself, it should be as accurate as possible but keep in mind that it will measure the full head when the guide vanes are closed.

7 Online flowrate monitoring experiences at Hydro-Québec 7/7 9 th IGHEM Conference, Trondheim, Norway Figure 9 : Manifold for pressure mixing Figure 1 : Angular encoder attached to the guide vane axis 4.2. Position of the guide vane The angular encoder (figure 1) gives an accurate measurement of the guide vane position and is preferred to servomotors stroke that is easily affected by the operating ring mechanism preloading and might slightly differ from one group to the other. Since the angular index method is very sensible to the position of the guide vane, great care should be taken to obtain the best description of this quantity Calibration of the coefficient Kop The calibration of the system should be provided by a code accepted method [2] [3]. A number of points on the curve should be adequate. Based on limited experience, a second order polynomial fit is usually enough to describe the variation of the coefficient. Since the system is able to capture moderate head changes, calibration only needs to be performed at one head Acquisition of the signal It is suggested to average the signal over a period of 3 seconds and to exclude the points where the guide vane moves more than.2 degrees. Obviously, the points where the bleed valves are in function should also be excluded. Conclusion This paper intended to demonstrate the applicability and usefulness of the angular index. Being a relative method, it is complementary to absolute measurements and open the way for fine tuning hydraulic production. It allows for accurate prediction of the flowrate when head change or other phenomena occur. While being applicable to a variety of turbine, it is one of the only reliable and affordable relative methods that can be used with low head plant to investigate the flow over long period. Future plans include the design of a new, more industrial, experimental set that should be more accurate and more robust. It should be installed in a number of units for the powerhouse mentioned on this paper. The integration of the online flow data to the various HQ systems also begun. The data will be provided to the various business units who will then decide how to use it. References [1] Nicolle J, Proulx G, 21, A new method for continuous efficiency measurement of hydraulic turbines, 8 th IGHEM conference, Roorkee, India [2] IEC 641 Field acceptance tests to determine the hydraulic performance of hydraulic turbines, storage pumps and pump-turbines [3] ASME PTC , Hydraulic Turbines and Pumps-Turbines Performance Test Codes

8 POLITEHNICA UNIVERSITY OF BUCHAREST POWER ENGINEERING FACULTY Hydraulics, Hydraulic Machinery and Environmental Impact Dpt. IGHEM 212 June 27-3, 212, Trondheim, Norway EXPERIMENTAL ANALYSIS OF THE OPTIMAL CAM CHARACTERISTIC FOR A KAPLAN TURBINE Georgiana DUNCA 1, Diana Maria BUCUR 1, Constantin CĂLINOIU 2, Eugen Constantin ISBĂŞOIU 3 1 Lecturer, Power Engineering Faculty, University Politehnica of Bucharest, Romania, 313, Splaiul Independentei, Bucharest, Romania, dmbucur@yahoo.com, tel: /fax: Associate professor, Power Engineering Faculty, University Politehnica of Bucharest, Romania 3 Professor, Power Engineering Faculty, University Politehnica of Bucharest, Romania Abstract In this paper the optimal CAM characteristic is determined for a unit of 22 MW with 12.5 m designed net head and 185 m 3 /s rated discharge, in a run of river power plant. Field tests aere performed, first in the conditions of the actual CAM, then with the CAM relation broken off. The efficiency is obtained at different runner blades and guide vanes opening combinations across a range of heads. The optimum three dimensional combination between the runner blades opening and the guide vanes opening, for different head values is determined to maximize the real operational efficiency. The conclusion is that the efficiency in the conditions of the new CAM characteristic increased up to 2%, with a smaller opening of the guide vanes, which decreased from 6% to 9% of the total opening. The results are confirmed analytically, by transposing the model theoretical efficiency curves to the real operational conditions (head and discharge). The determination of the real efficiency curve and the optimal CAM characteristic is important also from the vibration point of view. Keywords: CAM characteristic, Kaplan turbine, measurements, rated discharge, rated efficiency, vibration. 1. Introduction Considering the present efforts for increasing the green energy production, one of the easiest things to do is to optimize the operation of the hydro units. For existing Kaplan turbines, index testing and optimization are the best way to assure the maximum efficiency and power output. The necessity of analyzing the optimal CAM characteristic for this particular hydro power plant comes from the real on site operation conditions, which are different from those supposed when the plant and turbines were designed. In the initial project, the position of the runner referred to the downstream level is 1.5 m to 2 m. Because the next downstream dam is not yet built, the downstream level is lower than designed and the position of the runner is above downstream level at m. This changes totally the head of the turbines, from 12.5 m to 16 m, so the best efficiency point and the optimal CAM characteristic are different from those designed. 2. Experimental setup Field tests are carried out on a hydropower plant (HPP) located on one of the internal rivers from Romania [1, 2]. The HPP is situated on the river stream. In figure 1 the reservoir and the dam are presented. Figure 2 presentes the HPP cross section. The analyzed Kaplan turbine has 22 MW at 12.5 m rated head and 185 m 3 /s rated discharge, in a run of river power plant [3]. In figures 3 and 4 the turbine runner and guide vane and the generator stator are presented.

9 Georgiana DUNCA, Diana Maria BUCUR, Constantin CĂLINOIU, Eugen Constantin ISBĂŞOIU Fig.1 The HPP reservoir and dam Fig.2 HPP cross section Fig. 3 Turbine runner Fig. 4 Generator stator and turbine guide vane 3. Measurements First, for a number of operating regimes with the actual CAM combination, the hydrounit efficiency curves (η) depending on reported discharge (Q*) and the corresponding combination of guide vanes opening (S ad ) and runner blade angle (S r ), are determined. In the second part of the tests, the CAM relation is broken and a series of operating regimes with constant runner blade inclinations reported to the maximum inclination (S r =2.5%, S r =15%, S r = 25%, S r = 37.5% and S r =5%) are tested. For each inclination the variation of the flow parameters as a function of the reported discharge are measured for: - the guide vane opening, S ad, reported to the maximum opening; - runner blades inclination, S r, reported to the maximum inclination; - the level after the upstream grill, Z 1 ; - downstream level, Z 2 ; - pressure difference indicated by the two pressure taps on the spiral case, h; - electrical power, P g. The upstream and downstream levels are measured using the existing equipment on site. The reported discharge, Q*, is measured using a differential pressure transducer with a precision of.25 % (fig. 5), connected to the two pressure taps located on the spiral case. It is proportional to the absolute discharge, according to the equation Q * = k h [m 1/2 ]. (1) 2

10 Experimental analysis of the optimal CAM characteristic for a Kaplan turbine Fig. 5 Differential pressure transducer Fig. 6 Electrical parameters monitoring system The electrical power is measured using a monitoring system, with a precision of.3%. It is presented in figure 6. The guide vane opening, S ad, and the runner blades inclination, S r, are measured using the existing equipment on site. The vibration of the turbine shaft and of the turbine cover are measured for both actual and broken-off CAM characteristic. Shaft sensing proximity probes are used to obtain relative displacement measurements of rotating or reciprocating shaft surfaces. They are mounted in a traditional manner [3], the vertical probe being located directly above the shaft at 12 o clock, and the horizontal probe on the right side of the shaft at the 3 o clock position (fig. 7). The vibrations signals are acquired using a data acquisition system, with an acquisition frequency of 1 Hz. The turbine cover vibration is measured using an accelerometer. The main advantages of this kind of vibration transducers are that they measure casing or structural absolute motion and can be easily attached to structure. It is mounted on the turbine cover as shown in figure 8. The signal is acquired using a data acquisition system, with an acquisition frequency of 1 Hz. The first analized upstream level is the highest, 16 m, due to existing conditions. Then, the entire measuring procedure is repeted for the lower upstream level of 14.8 m. In the computation stage, the results for a third medium head of 15.4 m are determined by interpolation methods. Finally, the results covere all the operational domain. Fig. 7 Vibration transducers setting on the turbine shaft Fig. 8 Accelerometer s location on the turbine cover 3

11 Georgiana DUNCA, Diana Maria BUCUR, Constantin CĂLINOIU, Eugen Constantin ISBĂŞOIU 4 CAM characteristic All the results are computed for the net head using the equations: 1 2 H nc Q c = Q, (2) H n 3 2 H nc P c = P. (3) H n The following figures are presented for the higher upstream level, which corresponds to a 16 m net head. The data obtained for the original CAM characteristic are presented in figure 9 the reported efficiency curve and in figure 1 the relation between guide vane opening, S ad, and the runner blades inclination, S r. On the basis of data analyzed, efficiency curves are drawn, depending on the discharge and on power output, for each S r (fig. 11). By drawing envelope curves around efficiency curves, optimal efficiency values are determined. The guide vanes opening (S ad ) for each runner inclination are represented as a function of reported discharge and power output (fig. 12). By connecting the points corresponding to the best efficiency points for each S r, the optimal CAM characteristic is obtained. Reported efficiency η * - [-] Reported efficiency η * - [-] Reported discharge Q * - [m.5 ] Fig. 9 Hydro unit rated efficiency for the existing CAM relation Power output P g - [MW] 7 6 Runner blades opening S R - [%] Guide vane opening S AD - [%] Fig. 1 Actual CAM relationship 4

12 Experimental analysis of the optimal CAM characteristic for a Kaplan turbine % 5% % 5% % % Reported efficiency η * - [-] % 15% Reported efficiency η * - [-] % 15% Reported discharge Q * - [m.5 ] Power output P g - [MW] Fig. 11 Rated efficiency for broke-off CAM combination with envelope curve defining optimal values % Guide vane opening S ad - [%] Guide vane opening S ad - [%] % 15% 25% 37.5% Reported discharge Q * - [m.5 ] Power output P g - [MW] Fig. 12 Guide vane opening at broke-off combination with envelope curve defining optimal values Original CAM Optimum CAM Reported efficiency η * - [-] Reported efficiency η * - [-] Reported discharge Q * - [m.5 ] 1.5 Original CAM Optimum CAM Power output P g - [MW] Fig. 13 Rated efficiency for original and optimum CAM relation 5

13 Georgiana DUNCA, Diana Maria BUCUR, Constantin CĂLINOIU, Eugen Constantin ISBĂŞOIU 7 6 Runner blade inclination S r - [%] Original CAM H nc = 14.8 m H nc = 15.4 m H nc = 16. m Guide vane opening S ad - [%] Fig. 14 Original and optimum CAM combination For the new CAM combination obtained, the efficiency is computed and compared to the original one (fig. 13). The rated discharge is smaller in case of the optimum CAM, which proves the importance of the tests. The rated best eficiency point at 16 m net head is almost the same for both CAMs, and the power output of the hydrounit is 17 MW (fig. 13). This power output is obtained in the optimal CAM combination for a guide vane opening S ad = 65.2 % and runner blades inclination S r.= 37.5 %. In figure 14 the original CAM combination and the resulted optimum CAM combination are presented, for all three net heads. For theoretical determination of the optimal operation conditions, the unitary model characteristic of the turbine (fig. 15) is transposed for the analyzed turbine. The calculus is made for turbine runner diameter of 5 m and the rotational speed of rot/min, for a large range of net heads (from 1 m to 24 m, fig. 16), using Moody relation for turbine efficiency.2.2 DM DM nm ηtb = 1 + η M. (4) D D n Considering the net heat 16 m in figure 16, the mechanical power of the turbine at the best efficiency point is 18 MW. This happens for a guide vane opening of a = 39º combined with a runner blades inclination of φ = 19º (fig. 16). For an average value of generator efficiency at 95%, results a power output of the hydro unit of 17.1 MW. This result is very well correlated with the experimental results from figure 13, where it can be seen that at the best rated efficiency point, the power output is 17 MW. The runner blades inclination varies 3º and the guide vane opening varies 5º. Considering the best eficiency point of the new CAM characteristic (S r.= 37.5 % and S ad = 65.2 %), the absolute values are φ = 11.25º for runner blades inclination and a = 32.6º for guide vane opening. This confirms the optimisation of the new CAM, because the same power output (17 MW) is obtained with a smaller opening of the runner and of the guiding vane, so with a lower discharge. All these results confirm that in the real operation conditions, with a lower downstream level, the power output of the hydro unit at the best efficiency point is reduced from 22.5 MW (designed value) to 17 MW, for 16 m net head..1 6

14 Experimental analysis of the optimal CAM characteristic for a Kaplan turbine Unit rotational speed n11 [rot/min] Unit flow q 11 [m 3 /s] Fig. 15 Unitary model characteristic of the turbine K a = 21 a = a = 18 a = 33 a = 36 a = 39 a = 27 a = Net head H - [m] φ = φ = φ = -1 φ = Turbine mechanical power P - [MW] φ = φ = a = 42 Fig. 16 Operational characteristic of KVB 22,5-12,5 turbine (a - guide vane opening, φ - runner blades inclination) 7

15 Georgiana DUNCA, Diana Maria BUCUR, Constantin CĂLINOIU, Eugen Constantin ISBĂŞOIU 5 Vibration analysis For all measured regimes, the vibrations are also analized. The results are presented for the following power output values of 8 MW, 12 MW, 15 MW and 18 MW at 16 m net head, with the actual and broken-off CAM characteristic. Machinery vibration characteristics processed in time domain and in frequency domain can be presented on several distinct types of plots. In figures 17, 19, 21, 23 the shaft orbits are presented for four power output values, corresponding to the original CAM combination. In figures 18, 2, 22, 24 are presented the orbits obtained after the CAM relation breaking, close to the optimum regimes. Theese kinds of plots are useful to identify the shaft preloads. The presence of various types of unidirectional forces acting upon the rotating mechanical system is a normal and expected characteristic of machinery [3]. In figure 25 and 26 the frequency spectrums are presented for the vibration signals acquired on the turbine cover, for power output of 15 MW, close to real best efficiency point. It can be seen that the amplitudes are lower in the operation regime close to the best efficiency point (fig. 25) than in the actual CAM at the same power output (fig. 26) x [mm] x [mm] y [mm] Fig. 17 Turbine shaft orbit, original CAM characteristic, 8 MW power output y [mm] Fig. 18 Turbine shaft orbit, broken-off CAM approx. 8 MW, S r = 2.5% and S ad = 45% x [mm] x [mm] y [mm] y [mm] Fig. 19 Turbine shaft orbit, original CAM characteristic, 12 MW power output Fig. 2 Turbine shaft orbit, broken-off CAM approx. 12 MW, S r = 15% and S ad = 55% 8

16 Experimental analysis of the optimal CAM characteristic for a Kaplan turbine x [mm] x [mm] y [mm] y [mm] Fig. 21 Turbine shaft orbit, original CAM characteristic, 15 MW power output Fig. 22 Turbine shaft orbit, broken-off CAM approx. 15 MW, S r = 37.5% and S ad = 64% x [mm] x [mm] y [mm] y [mm] Fig. 23 Turbine shaft orbit, original CAM characteristic, 18 MW power output Fig. 24 Turbine shaft orbit, broken-off CAM approx. 18 MW, S r = 5% and S ad = 67%.1.8 Amplitude [V] Frequency [Hz] Fig. 25 FFT plot of the vibration signal on the turbine cover, original CAM characteristic, 15 MW power output Amplitude [V] Frequency [Hz] Fig. 26 FFT plot of the vibration signal on the turbine cover, broken-off CAM approx. 15 MW, S r = 37.5% and S ad = 64% 9

17 Georgiana DUNCA, Diana Maria BUCUR, Constantin CĂLINOIU, Eugen Constantin ISBĂŞOIU 6. Conclusions In this study field tests are performed for a hydro unit which equips a run of river HPP, first in the conditions of the actual CAM, then with the CAM relationship broken off. The efficiency is obtained at different runner blades and guide vanes position combinations across a range of heads. In the paper the results for one head are presented. The aim is to determine optimal CAM for current conditions of operation (low downstream level) and the hydro unit behavior in operation at various regimes in terms of level of vibration. After analyzing the measurements results can be stated that the optimum CAM relation obtained by measurements is very different than the actual one. It can be seen that efficiency maximum shifts from a reported discharge of 9 m.5 to 8 m.5. The recommended optimum operation in terms of optimal CAM relation is m.5, which corresponds to a electric power of MW. The shapes of the shaft orbits show a slight preload of the shaft that forces it to have an elliptical motion in all analysed cases. The direction of the orbit displacement is coherent with the direction of the flow entering the turbine impeller. However it can be seen that the shape of the orbits obtained for the original CAM relation are tighter and irregular which indicates that in the new conditions of CAM relation the hydro unit has an improved behavior from the shaft vibration point of view. Regarding the turbine cover vibrations analysis, it can be seen that their amplitudes are decreased in the best efficiency point comparing to the values measured in the case of the original CAM characteristic. Acknowledgments Results in this paper are part of research contract with the beneficiary Hidroelectrica Company. References [1] Contract entitled Determination of the real operating performances of the hydro units equipped with Kaplan type turbines, in order to increase their operation for electric energy production [2] International Standard IEC 41/1991, Field acceptance tests to determine the hydraulic performance of hydraulic turbines, storage-pumps and pump-turbines, Bern, 1991 [3] Eisenmann, R. C. jr. and Eisenmann, R. C. sr., 1997, Machinery malfunction diagnosis and correction, Prentice Hall PTR, New Jersey. 1

18 1. Abstract IGHEM 212 The 9 th International conference on hydraulic efficiency measurements Trondheim, Norway June 27 th - 3 th, 212 Strain gauge measurements of rotating parts with telemetry Johannes Löfflad, Marco Eissner, Bernd Graf Voith Hydro Holding GmbH & Co. KG Alexanderstr. 11, Heidenheim, GERMANY XXXXXX Modern design of hydraulic turbines aims to achieve very high levels of efficiency and structural integrity in the environment of highly variable loading conditions. To combine those requirements, a profound knowledge of static and dynamic loads acting on hydraulic components is necessary. Dynamic loadings of hydro runners strongly depend on head range and required operating conditions. The knowledge of dynamic loads and stresses is often derived from strain gauge measurements. This contribution describes the equipment and procedure of strain gauge measurements at hydro turbines, including the challenges to overcome during such tests. It provides an overview of strain gauge test results for stationary as well as for transient load conditions and the arrangement of strain gauge measurement for all typical hydro turbines are shown with their respective challenges. The measurement solutions and results for Francis, Kaplan and Pelton machines are discussed in the following paper. The main focus is given to protection of measuring lines, online transmission of data, evaluation of transient operating conditions and life-time forecast by analyzing the measured data. Keywords: Strain gauge test, static and dynamic loading, stationary and transient load condition

19 2. Introduction Getting a deeper understanding of the dynamic behaviour of water turbines becomes essential with the last years. More and more it comes alight, that mainly the dynamics are driving the partial damage of turbine components like runners and generators. With the recommendation to reduce weight of components and increase efficiency of turbines, the dynamic sensitive of these machines changed completely. Within the last years the technical progresses comes up to a point, where it is possible to measure signals at rotating parts and get them online into the stationary system via telemetry. This technology is very powerful to improve the understanding of the dynamic behaviour of water turbines, because it is also possible to measure these signals time synchronous with signals from the stationary system, like pressure pulsations or vibration values. Figure 1: Typical frequency spectra of measured strains for a high head Francis runner At low outputs, a stochastic behaviour with broad band frequency content is present. At part load condition, the typical low frequency excitation caused by vortex rope phenomena (related to Rheingans frequency) appears in the frequency spectra. From part load to full load, a higher frequency component arising from RSI, the so called Gate Passing Frequency (GPF), dominates the frequency spectra of higher head Francis runners. 3. Method 3.1. Functionality For measuring the stresses on a Francis runner, a telemetry system is used. This kind of system provides a contact less transmission of sensor signals (generally strain gauges) from the rotating system to a stationary receiver. Hence the system consists of a rotating unit (emitter) and a stationary unit (receiver). The sensor signals are amplified and are converted in a first modulation (voltage frequency) to a subcarrier frequency. In a mixer the channel specific subcarrier frequencies are merged. The subcarrier mix is modulated to the main carrier frequency by HF-modulator. This signal abuts at the sending antenna and can be received by stationary antennas. The demodulator in the receiver converts the HF-signal back to a subcarrier mix and in a second step back to a voltage signal (frequency voltage). An analog amplifier provides the analog signals (± 1 V or 4-2mA) to the output of the receiver.

20 3.2. Capabilities Following table gives an overview of the capabilities of the telemetry system used in hydro power plants. Component Turbine Turbine Unit-Type Francis / Pump / Pumpturbine Francis / Pump / Pumpturbine Measurement- Type Runner strain Sensor location Data Transmission Air Water Trailing edge Leading edge Static pressure Runner cone Turbine Francis / Pump / Pumpturbine Axial shaft torque Turbine Kaplan Runner strain Turbine Kaplan Axial thrust / Shaft torque Shaft Trailing edge Leading edge Fillet, Groove Turbine shaft Turbine Pelton Runner Strain Root Generator Synchron Rotor/pole strain Pole fixation Generator Synchron / asynchron Axial and bending Shaft torque; shear forces Table 1: Overview of applications in hydro power plants Generator shaft 3.3. Configurations Typically two different applications of the system installation are practicable. For Units having a central air admission pipe, going from the runner to the top of the generator, the data transfer is realized in air. For this application, the rotating antenna is located at the top of the generator, where also the power supply for the data acquisition system is fitted. To minimize the data transfer length, the stationary antenna is located as close as possible to the rotating one. For Units without central air admission pipe, the data transfer is warranted by sending through water. For this application, the stationary antenna is located below the rotating runner at the draft tube. To achieve a stable data transfer, the rotating antenna is installed on runner band side, in between two runner blades. In order to maximize the system availability for both types of applications, an intermediary remote system is used to switch on/off the telemetry system at any time Sending through air The data transmission through air is the standard configuration for strain gauge measurements. For this application the telemetry system is located at the runner cone in a waterproof construction, which is specially adapted for the test setup. The signal wires lead the strain signal from the strain gauge position along the runner blade trailing edge to the waterproof construction. The converted and modulated signal is lead by the antenna cable through the shaft to the top of the generator where also the power supply, the remote system and the rotating antenna is located. In case of an extended test which exceeds the power supply capacity, the batteries can be changed easily. With the realized application it is not necessary to dewater the unit.

21 Figure 2: Application sending through air Sending through h water Whereas at units without a central air admission pipe the data transmission in air is not possible. In that case the data transmission from the rotating to the stationary part is accomplished in water. The method of strain gauge signal erection is the same for this configuration, but the location of the power supply, the remote system and the rotating and stationary antenna differs from the data transmission through air. All these parts are located in the waterproof construction at the runner cone, where also the telemetry system is fitted. Several stationary antennas are distributed at the draft tube, as close as possible to the rotating runner. Figure 3: Application sending through water

22 3.4. Simulation in preprocessing The strain gauge positions are derived from static Finite Element results. Based on the stress distribution for several calculated load cases the proposed strain gauge position and orientation are defined. Usually highest static and dynamic stresses on a Francis runner occur close to the trailing edge, at the transition to band and crown side. See following figure with typical stress distribution and orientation of main stresses for a Francis runner at trailing edge to band (left side) and at trailing edge to crown (right side). Arrows indicate the orientation of the principal stress, the length of the arrows is magnitude based. stress distribution at trailing edge to band stress distribution at trailing edge to crown Figure 4: stress distribution at a Francis runner trailing edge to band and to crown Practically it is not possible to measure at the maximum peak stress. This limitation is caused by the size of the strain gauge and the necessary protection layer against irruption of water Application This section introduces the definition of the strain gauge positions at the prototype, how to apply the strain gauges and how to protect them against water irruption Patterns In order to get the exact position for the strain gauge application, special tools are used. For exact position of the strain gauge sensor, steel templates are used. These are derived from the CAD-model of the Francis runner. Templates designed in CAD-System Templates used at site Figure 5: Templates used for strain gauge application

23 FARO-Arm Arm Another possibility to locate exact strain gauge positions is the usage of the FARO-Arm. This method of defining strain gauge positions on the prototype is used specially for Kaplan runners, where the usage of steel templates is not practicable. FARO-Arm used for model machines FARO-Arm used for prototype size Figure 6: strain gauge positions with FARO-Arm Strain gauge application The strain gauges are bond to the surface by using a modified alkyl cyanoacrylate compound. This bond cures at room temperature and gives a very thin, uniform layer of adhesive for optimum bond performance. A special coating procedure is required for strain gauges on water immersed surfaces. It is necessary not only to protect the strain gauge itself but also the soldering and the lead wire. A cross section of a typical water resistant strain gauge application is shown in Figure 7. Unidirectional strain gauge Local coating Aluminum foil Figure 7: Protective coating of strain gauge installation

24 Protection Layer The cables going from the strain gauges to the data acquisition unit are fixed to the surface by a double-sided adhesive tape. In order to protect the strain gauge wires against water irruption, at least one protection layer is used. The first layer is a two-component epoxy resin with excellent characteristics working under water and under pressure. For applying this layer, the surface has to be treated specially in order to improve the adhesion of the epoxy. The epoxy is shaped to minimize the effects on the hydraulic, see below figure. In case of measuring under high pressure (p > 3 bar) or under special circumstances (e.g. high sand content), a second protection layer is used. This layer is a local coating developed to resist higher pressure (water tightness) and better resistance against erosion. For this application several layers have to be applied, see below figure. First protection layer (epoxy resin) Second protection layer (epoxy resin + coating) Figure 8: protection layers for strain gauge application In case of measuring a Francis runner, the cables from the band side are applied along the trailing edge (pressure or suction side) to crown side, where the data acquisition unit is located Measurement Synchronous data acquisition With the used telemetry system and the associated online data transmission, it is possible to record all signals from the strain gauges time synchronous with other signals, like static pressure, pressure pulsations, bearing and shaft vibrations and control panel signals. Especially for transient load conditions like start up, stop, load rejections and change over operation of pump turbines a synchronous data acquisition has the following advantages compared with a data logger: Functionality check of the rotating measurement equipment Continuous optimization of the measurement program by online assessment of the measured strains Optimization of transient load conditions without dewatering (e.g. start-stop, load rejection) Estimation of Systematic measurement uncertainty 1) For the calculation of uncertainty of runner stresses the uncertainty of the young s modulus E and the uncertainty of the measured strain ε has to be considered. 2) The uncertainty of the Young s modulus is assumed to be f E=±3% 3) The uncertainty in the measured strain is based on: Shunt calibration => sensitivity k, R SG (Strain gauge), R p (Shunt) Zero instability effects due to temperature, isolation, cabling Misalignment of the strain gauge

25 3.1) Shunt Calibration 1 R P 6 ε = 1 1 µm/m k RSG + RP Thus the abs. uncertainty (e) is 2 ε ε ε + ε = ± ek + ersg ep µm/m k RSG RP e c 2 2 With ε 1 = ² k k R SG RP + R P ε R SG = k ( R + R ) SG R P P RP ε RSG + RP SG P SG =.1 = R k k( R + R P 2 ( R + R ) 6 R 6 SG p ) 2 *1 Strain gauge factor Resistance of strain gauge Resistance of Shunt f K=1% (k=2.6) f SG=.5% (R SG=35 Ohm) f P=.2% (e.g Ohm) With said values the relative uncertainty in the calibration becomes: e c=±5,4 µm/m 3.2) Zero Instabilities During the short term test zero shifts will occur and have to be considered in the estimation of uncertainty. They are measured before and after the test, for zeroing of signals, their average will be considered. These depend on environmental factors and can be estimated as follows: f c=±5.% 3.3) Misalignment Assuming that misalignment ϕ of the strain gauge will not exceed 1 to 15 deg. from the max. Stress axis the reduced stress σ can be calculated as 1 ε = ε max (1 + 2cos2ϕ ) µm/m 2 Thus the uncertainty in misalignment is assumed to f m=±3.% to 6.7% 4) Relative uncertainty of Strains Based on the assumptions made before, the relative uncertainty in the measured strains is: f = f + f ε c z m % f ε = ±5.9 % f 5) Relative uncertainty of Stress 2 f = ± fε + 2 f E ε % > f ε = ±6. 7 %

26 3.7. Data Evaluation Online Assessment for special transient load conditions Performing a strain gauge measurement by using a telemetry system offers the possibility of online assessment of recorded strain gauge data. Especially for transient load conditions, a fast and reliable online assessment of these load conditions is possible. They could be optimized in matters of time, dynamic loading and partial damage. Start up procedure The correlation between process parameter and time synchronous measured strains on the runner provides one major benefit of using a telemetry system, especially when modifying the start up sequence in order to improve the dynamic behaviour. Following figure gives an overview of several start up procedures and measured strains on a Francis runner. Start up 4 Start up 3 Start up 2 Start up 1 Time Figure 9: start up procedures and measured strains on a Francis runner for start up

27 The strain gauge signals in figure 9 are normalized to the mean stress at nominal speed. The level of mean stress after reaching nominal speed can not be influenced by the start up procedure, but as shown in figure 9 the dynamic part at the beginning is significantly reduced. Load rejection Besides the start up procedure also the behaviour during load rejection can be observed and, if necessary, be adapted. Following figure gives an overview of several stop sequences during load rejection and measured strains on a Francis runner. Strain gauge signal Time Figure 1: Stop sequence and measured strains on a Francis runner for load rejection As visible in figure 1, the static as well the dynamic part of the strain gauge signal is very sensitive to different stop sequences. Mean stress evaluation Beside the evaluation of the transient load conditions also stationary load cases can be considered. The figure below shows the trend of the mean normalized output. Mean Stress [-] CROWN measured BAND measured CROWN FEA BAND FEA Active Power [%] Figure 11: Normalized mean stress (measured and calculated) over output

28 The measured mean stresses are normalized to stresses at speed no load. The plotted lines indicate the characteristic for crown (blue) and band (red) from to 1 percent output. The dots represent the calculated values from the Finite Element Analysis. It can be seen that the trend of stresses versus output is well represented. Deviations of measured and simulated stresses cannot be avoided and may have different reasons: a) Geometrical deviations of real runner blade contour versus FEA model (manufacturing tolerances) b) Deviations in loading conditions (real flow and head conditions versus CFD assumptions) c) Deviation of real strain gauge position and assumed position in the FEA model (mismatch of location) d) Flow disturbances due to strain gauge and wire application on runner blades (necessary protection during test) The reasons (a) to (c) often result in an offset of measured stresses compared to the simulated ones. For the last reason (d), it can be observed that deviations at optimum and full load are sometimes higher compared to lower outputs. This might be caused by larger flow disturbances from strain gauge applications for well developed flow conditions. In order to improve the result quality as much as possible and to minimize the deviations, a careful preparation of the entire test including accompanying numerical simulations and a professional installation procedure is very important. Necessary actions must be coordinated between all parties. Once the mean stress comparison is done with reasonable accuracy, a closer look on dynamic stresses and strains is possible. Dynamic stress evaluation In addition to the evaluation of the mean stresses, also the dynamic parts of the signals are considered. For this a characteristic value is established. This value is called characteristic peak-to-peak, where approximately 3% of the highest peaks are neglected. Dynamic Stress [-] CROWN measured BAND measured Active Power [%] Figure 12: Normalized dynamic stress over output The measured dynamic stresses (characteristic peak-to-peak) are normalized to speed no load values. As seen in above figure the dynamic value increases for stresses at crown (blue line) and band (red line) during part load. With higher loads the characteristic value is reduced significantly.

29 Partial Damage Based on the measured strain signals on the runner and a load universe, a partial damage can be calculated in order to detect load cases, which should be avoided or the time running the unit in a special load case should be reduced in order to increase the life-time of the runner. Following chart gives an overview of partial damage for several tested load conditions. partial damage CROWN BAND Start up Shut down Load rejection SNL 16% Pmax 32% Pmax Pmax Figure 13: Normalized partial damage for different load cases The partial damage is normalized to partial damage for P max. The transient load conditions are the main fatigue contributor hence the improvement of these load cases increases the life-time of the runner most. For this example the partial damage at runner band is higher than at runner crown, depending on the runner this could also be changed. 4. Summary Performing strain gauge measurements on rotating parts using telemetry systems is state of the art. Due to the possibility of data transmission through air and water, strain gauge measurements can be realized with almost no limitations of unit application and kind of runners. Based on the database of strain gauge measurements, a calibrated procedure for predicting static and dynamic stresses has been developed and optimized. This procedure enables Voith Hydro to optimize runners with regard to static and dynamic stress as well as to fatigue behaviour. 5. References Following table gives an overview of realized and projected strain gauge measurements at Voith Hydro using a telemetry system. No. of realized and projected strain gauge measurements Francis runner Kaplan runner Pelton runner Data transmission through air 6-2 Data transmission through water 3 projected n.a.

30 Jean HERAUD NOTE TECHNIQUE Optimiz Software Features and global structure IGHEM 212 Date D4136/NT/ A Indice : A 11 Pages annexe(s) pièce(s) jointe(s) Entité émettrice : Documents associés : EDF DTG MPSH Résumé : This document is the article on Optimiz Software which will be presented at IGHEM 212. Intérêt documentaire : Oui Accessibilité : Documentation de référence : Non Libre Direction Production Ingénierie - Division Production Ingénierie Hydraulique DTG Direction 21, avenue de l'europe - B.P GRENOBLE CEDEX 9

31 Optimiz Software Page : 2 / 11 Features and global structure IGHEM 212 Réf. : D4136/NT/ A Indice : A Création - Modifications Ind. Auteur(s) Vérificateur(s) Approbateur Nom Visa Date Nom Visa Date Nom Visa Date A J HERAUD B REEB M PERSOZ Historique des modifications Indice Date Paragraphes modifiés / Objet A Avril 212 Création of presentation Diffusion Destinataire(s) pour application Nb Destinataire(s) pour information Nb IGHEM Haradl HULAAS Base Perf Essais hydrauliques Copyright EDF Ce document est la propriété d'edf. Toute communication, reproduction, publication, même partielle, est interdite sauf autorisation

32 Optimiz Software Page : 3 / 11 Features and global structure IGHEM 212 Réf. : D4136/NT/ A Indice : A OPTIMIZ software for estimation of hydraulic plants global performances in all their operating configurations Features and global structure Jean HERAUD EDF DTG France jean.heraud@edf.fr 1. INTRODUCTION EDF DTG developped in 211, a tool called "OptimizV2" which, knowing the intrinsic performances of units (directly taken from site measurements) and losses coefficients in pipes, gives the performance of the plant in all its operating configurations. (for all heads, all openings of wicket gates/injectors and for any combination of units) This allows an optimization of the use of the plant : Knowledge of the power levels associated with efficiency peaks (whatever units running, whatever head) Knowledge of power margins available for contracting system services (A power target value being proposed) Besides, it allows for major hydraulic plants, a monitoring of power. This tool version 2 already exists (macros Excel VBA) and version 3 is being at present implemented (intranet server with mathematical part on Matlab) This version has extended functionalities (monitoring for exemple or comparison of productible) This issue presents the features, uses and structure of this product. Copyright EDF Ce document est la propriété d'edf. Toute communication, reproduction, publication, même partielle, est interdite sauf autorisation

33 Optimiz Software Page : 4 / 11 Features and global structure IGHEM 212 Réf. : D4136/NT/ A Indice : A 2. PRESENTATION OF THE TOOL 2.1. MODELISATION OF A PLANT The modelisation : Describes the structure of the plant Precises the head loss coefficients Gives unit s performance data (coming from on site performance tests) OptimizV3 (being developped) : Creation / modification of a modélisation Name of modelisation : Name of the plant : Structure of the plant Head loss coefficients Group performance input data Record Cancel Structure of power plant (1/2) Number of units Number of sections 2 4 Continue Cancel Copyright EDF Ce document est la propriété d'edf. Toute communication, reproduction, publication, même partielle, est interdite sauf autorisation

34 Optimiz Software Page : 5 / 11 Features and global structure IGHEM 212 Réf. : D4136/NT/ A Indice : A Structure of power plant (2/2) For each group, select the sections involved in their operation G1 G2 Section 1 Section 2 Section 3 Section 4 Intermediate Water intake possible in headrace tunnel Indicate the section "bypassed" by this input No Record Cancel Head loss coefficients Enter head loss for each section for a turbine or pump operation (mwc/(m 3 /s) 2 ) Turbine Pump Section 1,4642 Section 2,2617 Section 3,148 Section 4,148 Record Cancel Unit performance input data Allocation of data sets (turbine) Injectors altitude (Pelton) Allocation of data sets (pump) G1 G2 St Guill-def St Guill-def No No Record Cancel Copyright EDF Ce document est la propriété d'edf. Toute communication, reproduction, publication, même partielle, est interdite sauf autorisation

35 Optimiz Software Page : 6 / 11 Features and global structure IGHEM 212 Réf. : D4136/NT/ A Indice : A Group performance dataset Series of data n 1 Name of dataset St Guill-def Net Head = 234,5 Opening 6,2 99, ,3 164,8 176,3 196,5 217,2 22 α = 1,6 β =,5 P (MW) Q (m3/s) 1,51 6,71 2,87 11,34 28,41 14,47 35,47 17,46 38,57 19,12 4,88 2,18 44,1 22,9 46,3 23,85 46,67 24,1 New operating point Series of data n 2 Net Head = 258,5 α = 1,5 β =,5 Opening P (MW) Q (m3/s) 81 19,62 9,7 1,5 26,31 12,66 114,5 31,52 14, ,1 15, ,76 16, ,6 18,51 17,5 47,82 2,9 183,5 5,45 22, ,3 25,21 New operating point Series of data n 3 Net Head = 292,25 Opening 79, 92, α = 1,4 β =,5 P (MW) Q (m3/s) 25,34 1,73 31,11 12,63 35,95 14,24 42,25 16,38 New operating point New blank series of data Type of interpolation (steady net head) Cubic spline (default) Limitations associated with the dataset : MIN limit Power Type of max limit : Opening MIN Value 15 Value : 22 Type of max limit : Power Value 45 Visualization of the data set Choice of x-axis Output Choice of y-axis Power Efficiency chart Record Cancel α and β are the coefficients for transposition into net head of power and flow rate. - In turbine, they worth by default 1.5 and.5 (these values do not need to be refined if the simulated net head is framed by net heads of input datas) - In pump, no default value are proposed. They must be defined at each time Graphic validation of input datavalidation OptimizV2 allows a visualisation of input data and of the interpolation curves associated. Il allows to check : the quality of interpolation the coherence of the data at various net head values Many types of curves are proposed among which, the one below (Electric power function of wicket gate opening.) Copyright EDF Ce document est la propriété d'edf. Toute communication, reproduction, publication, même partielle, est interdite sauf autorisation

36 Optimiz Software Page : 7 / 11 Features and global structure IGHEM 212 Réf. : D4136/NT/ A Indice : A 2.2. SIMULATOR SCREEN The screen below is taken from OptimizV2 The scalars obtained for each unit are : Opening Total head Net head Head losses Power Flowrate Output kwh/m 3 Screen of OptmizV2 Screen of OptmizV3 Copyright EDF Ce document est la propriété d'edf. Toute communication, reproduction, publication, même partielle, est interdite sauf autorisation

37 Optimiz Software Page : 8 / 11 Features and global structure IGHEM 212 Réf. : D4136/NT/ A Indice : A 2.3. GRAPHICS The tool (V2) is already able to generate curves for variations of load or head. Superposition of curves are possible (as in the Output/Power drawing below) 2.4. CALCULATION OF PRODUCTIBLE AND ENERGETICAL COEFFICIENT OptimizV3 will be able to calculate the productible of the plant or the energetical coefficient (kwh/m 3 ) for a monitored period of supervision. More precisely, with the monitored Power of each group and the upstream and downstreal levels, Optimiz will calculate Productible Energetical coefficient This calculation will be made possible by the importation of a monitoring file. Copyright EDF Ce document est la propriété d'edf. Toute communication, reproduction, publication, même partielle, est interdite sauf autorisation