Development of Endoscopic Particle Image Velocimetry For Application in High Temperature Furnace

Size: px

Start display at page:

Download "Development of Endoscopic Particle Image Velocimetry For Application in High Temperature Furnace"

Annabel Price
5 years ago
Views:

1 Development of Endoscopic Particle Image Velocimetry For Application in High Temperature Furnace D. Honoré *, C. Rottier, G. Godard, F. Corbin, A.M. Boukhalfa CORIA CNRS, Université et INSA de Rouen, Saint Etienne du Rouvray, France Abstract This paper presents the development of an endoscopic Particle Image Velocimetry system adapted to high temperature furnaces. The optical setup is designed to be able to image large fields of measurements with a collection angle of 60. This optical setup acts as a camera lens in front of the CCD camera and is set in a watercooled jacket for its thermal protection. A strong geometric barrel distortion is obtained when imaging large field of view. Distorted particle images are then corrected by a specific image processing. A first application of this endoscopic PIV system is presented in laboratory-scale furnace. This shows its ability to obtain large velocity fields even in such high temperature confined configuration. Introduction Improvement of energy efficiency and pollutant and greenhouse gases emissions of industrial combustion furnaces requires performing experimental study in pilot facilities with operating conditions close to the industrial ones. Whereas laser diagnostics are commonly used for experimental study of combustion, their application in complex combustion systems is difficult and still requires development and adaptation. In the case of high temperature furnaces, main difficulties come from the small size of optical accesses in such confined combustion chamber, the need of large field of measurements and the strong emission of high temperature refractory walls. A few optical diagnostics have been already implemented in high temperature combustion facilities [1]. For velocity measurements, Laser Doppler Velocimetry has been adapted to semi-industrial combustion facilities at the end of 1980 s. The setup consists of a small LDV probe in back scattering configuration set in a water-cooled housing with a water-cooled front window and a nitrogen purge [2-4]. This configuration has been for long the only laser velocimetry technique applicable in high temperature combustion facilities and is still used nowadays in several R&D research centres. As LDV measurement is local, complete characterisation of the flow in the combustion chamber requires repetition of measurements at different positions. This is usually done by immersing the LDV probe along a radial profile at successive longitudinal locations in the combustion chamber. Then only longitudinal and tangential velocity components can be obtained. Moreover, this induces also long durations of measurements in such costly combustion facilities. One way to reduce measurements time and obtain straight two-dimensional map of longitudinal and radial velocity components, is the use of imaging techniques such as Particle Image Velocimetry (PIV) [5]. A first attempt of application of PIV in semi-industrial combustion facility has been done in a 1MW boiler with low wall temperature [6]. For that, a commercial high temperature endoscope, initially designed for in-furnace inspection and monitoring, was used. A camera lens was required between the endoscope and the CCD camera, which limited the transmission and optical quality. The use of an endoscope is a convenient way to combine large field of view and limited optical access. Similar issues exist in internal combustion engine and have induced some developments of small endoscopes for PIV [7, 8]. This has required specific particle image and velocity fields processing as for high temperature furnaces. Specific objectives The aim of this study is to develop an endoscope adapted to high temperature combustion facilities. The optical setup is redesigned to be used directly in front of a CCD chip. Because of strong barrel distortion of large particles images, a specific image processing is developed before velocity calculation thanks to a dedicated PIV algorithm. A first test of this endoscopic PIV system is performed in a high temperature pilot furnace. The high temperature endoscope The high temperature endoscope developed for the application of PIV in furnace acts as an optical relay between the combustion chamber and the CCD camera through the width of refractory walls of the furnace. On a similar basis as high temperature LDV probe and commercial high temperature endoscope used for inspection and monitoring, it consists of a water-cooled stainless steel housing in which the optical probe is set. The useful length of the endoscope is 700 mm for an * Corresponding author: david.honore@coria.fr Proceedings of the European Combustion Meeting

2 external diameter of 78 mm (Figure 1). A continuous low nitrogen flow is injected through the optical tube and comes out in front of the first lens to ensure its thermal protection and to avoid particle deposit during furnace operation. Such endoscope can operate in combustion furnace with wall temperature up to 1600 K Different CCD cameras can be installed at the exit of the water-cooled housing. An extensible gusset is used to adjust the CCD chip on the image focal plane. For PIV measurements, the endoscope is equipped with a Hamamatsu C camera (2048 x 2048 pixels 12 bits). Figure 1. The high temperature endoscope equipped with the PIV CCD camera. In addition to the small optical access, another issue of application of PIV in a furnace comes from the strong radiation of high temperature refractory walls added to the natural emissions from the flame (chemiluminescence and soot radiation) which limit the collection of particle images. These background emissions have to be minimised on particle images to ensure sufficient signal-to-noise ratio. Electronic shutter of CCD cameras can not be used for this purpose as it does not operate on the second frame of each couple of particle images. Interferential filters centred on the laser wavelength which is usually convenient to block flame emission is no longer appropriate in high temperature furnaces. For that, a Ferroelectric Liquid Crystal (FLC) shutter is set in front of the CCD chip. The on/off contrast ratio of this shutter is 1000/1. Its opening (150 μs) is synchronised with the laser pulses. We take also advantage of the large dynamics of the CCD camera (12 bits) to have good signal-to-noise ratio. This setup permits to obtain particle images from which velocity field can be determined even in such high temperature confined environment. The optical setup in the water-cooled casing has been designed with the ray tracing software OpTaliX Pro. It had to be able to image a large field of view with a collection angle of at least 60 at 532 nm, the wavelength of the PIV Nd-YAG laser. A small aperture (φ = 1 mm) is set at the entrance of the endoscope in front of the first lens to protect it and to limit geometric aberrations. The optical setup consists of 11 converging and diverging lenses set in two parts : a first group of 3 lenses is used to construct a real image with a magnification ratio of 1/75 whereas the second part of the optical setup acts as an optical relay to carry this image in the optical tube to the CCD chip. For that, a second group of 5 lenses is used to make parallel the rays that propagate in the optical tube. Then, three achromatic lenses are set along the endoscope to keep paraxial rays and focus the final image at 1064 mm from the entrance aperture. Figure 2 presents the paths in the optical setup of a few optical rays from an object set at 250 mm to the entrance aperture to the CCD chip inside a collection half-angle of 30. Results and discussions Figure 3 presents the image of a regular squared grid obtained through the endoscope with the PIV camera. Each cross symbol on the image is separated by 10 mm with its neighbours. Objects with dimensions of 300 x 300 mm² at 250 mm can be effectively imaged, i.e. with a collection angle of 62. Large dimensions of the CCD chip (2048 x 2048 pixels) allow to keep good spatial resolution even with such large field of view. Different geometric aberrations are expected when imaging large objects. Vignetting and field curvature are limited thanks to the small entrance aperture. Drawback of the latter is the limitation of the transmitted light flux which however is sufficient for PIV measurements. A large depth of field is also obtained thanks to the small entrance pupil which enables to consider using this setup on lab-scale high temperature combustion facilities as well as on semi-industrial ones. As predicted from the ray tracing software, some astigmatism is observed from 75 mm from the centre of the image. However, this has a limited effect on PIV calculation. Indeed, in off-axis regions, even if the correlation calculation is performed on unfocused patterns rather than on well resolved particles images, result is still sufficient to detect a correlation peak and then deduce a velocity. As can be seen on the image, main aberration is the geometric barrel distortion. This is typical of optical Figure 2. Optical setup of the high temperature endoscope. The colour lines represent the paths of rays in 30 collection half-angle. 2

3 system with large collection angle as fisheye lenses in photography. In this case, the magnification is no more constant in the whole of the image and varies with the radial location of the object compared to the central optical axis. Correction of this geometric aberration is usually done by digital image processing. Figure 3. Image of a squared grid obtained with the endoscopic system. The distance between two neighboured crosses is 10 mm. constant value M = 7 pixels/mm can be considered on a central region of at least 50 x 50 mm². Same value for M is chosen for the whole of the undistorted image which is constructed step by step in each square of 70 x 70 pixels equivalent to 10 x 10 mm² on the real object plane. Figure 4 presents the methodology of construction of the undistorted image in each square (A B C D ) of 70 x 70 pixels. For each pixel (M ; N ) in the square (A B C D ) in the undistorted image, one calculates the relative positions x M and y N of the pixel in the square. Then, same values are considered on the four sides of the equivalent region (ABCD) in the raw distorted image: A'M' AM1 CM 2 x M = = = A' B' AB CD A' N' AN1 BN2 y N = = = A'C' AC BD The sub-pixel position (M ; N) of the intersection of (M 1 M 2 ) and (N 1 N 2 ) is then determined in the trapezium (ABCD). Finally, the gray level associated to the pixel (M ; N ) in the undistorted image is obtained from the four real pixels the nearest of (M ; N) by calculating the mean value of their gray levels weighted by the distances between (M ; N) and each real pixel. A specific particle image processing Methods of image processing for the correction of geometric aberrations are based on the acquisition of a calibration image consisting of a grid of regular patterns or lines, as presented Figure 3. Common unwarping corrections methods of barrel distortion use polynomial functions [7, 9]. Coefficients of the polynomial functions are determined from a set of points selected on the calibration image. In our case, the precision of the reconstruction is less than 1 mm in the 200 x 200 mm² central region but larger outside, which is not sufficient. Moreover, the coefficients of the polynomial functions depend on the position of the points selected on the calibration image. No objective and satisfactory methodology of correction can be established. For that, we have chosen to develop another correction method based on successive local corrections in image. An automatic procedure based on correlation calculation is first used to find sub-pixel positions of the crosses in the calibration image. Then the latter is divided in successive trapeziums (ABCD) corresponding to 10 x 10 mm² on the grid. As the geometric distortion is a third-order function of the value of the radius from the optical centre, the variation of the magnification ratio is important on the boundary of the image, whereas it is negligible on the centre of the image. In our configuration where the object is at 250 mm from the endoscope entrance, a Figure 4. Methodology of the construction of undistorted particle image (top) from raw distorted particle image (bottom) in each region of 10 x 10 mm². This method allows obtaining undistorted image with a very good precision as shown on Figure 5. The precision of the correction is less than 1 mm. Main limitation comes from the detection of crosses in the border of the images where astigmatism limits the resolution. 3

Standard PIV measurements have been already done in this facility [10, 11]. The image collection system consisted of a Nikkor 50mm f/1.2 lens and a CCD camera (1280 x 1024 pixels 12 bits).

4 Standard PIV measurements have been already done in this facility [10, 11]. The image collection system consisted of a Nikkor 50mm f/1.2 lens and a CCD camera (1280 x 1024 pixels 12 bits). The field of view obtained with this configuration was 120 x 90 mm². Thanks to the endoscope, a wider field of view of 300 x 250 mm² can be now obtained from the same optical access and with a better thermal confinement. Figure 5. Image of the undistorted squared grid constructed with correction processing. For each experiment, squared regions are classified from a calibration image. Then series of raw particles images are processed in an automatic batch process. Duration of the calculation of an image is 1.5 sec. on a 3 GHz PC. After this pre-processing, couples of undistorted particle images can be used to determine instantaneous velocity field whichever the PIV algorithm. Application on a lab-scale flameless mild pilot furnace This section presents the first application of our endoscopic PIV system on a combustion furnace. This has been in the framework of a national research project in collaboration with GDF SUEZ R&D Division. The facility is a laboratory-scale furnace, named FOUR (Furnace with Optical accesses and Upstream Recirculation), designed for the experimental study of flameless mild combustion regime [10]. The combustion chamber has a 500 x 500 mm² squared section and 1 meter length. Refractory walls (100 mm width) enable to operate at a maximum wall temperature of 1400 K. In-flame measurements can be done through several openings set along the combustion chamber in staggered rows. Optical diagnostics are applied on the facility by the use of squared windows (100 x 100 mm²) set on the external side of the furnace. Achievable field of view is then limited by the width of the refractory confinement which acts as an optical diaphragm. The burner is set on the hearth of the furnace. It consists of 3 parallel injections: a central air injection (25 mm inner diameter ending in a divergent shape) surrounded by 2 methane injections (3 mm inner diameter). Velocity field measurements by endoscopic PIV are performed in the vertical plane of the 3 main axes of methane and air jets. For the present endoscopic PIV experiment, a doubled pulsed Nd:Yag laser Quantel CFR200 (180 mj/pulse) is used. The laser sheet formed by a set of lenses is spreading in the furnace through a 100 x 100 mm² squared optical window by means of a periscope setup. A converging lens (f = 1000 mm) reduces the width of the laser sheet to 1 mm. A diverging lens with a short focal length (f = 19 mm) is set the closest to the entrance optical window to expand as high as possible the laser beam. In this configuration, the height of the laser sheet is limited by the width of the refractory walls. Because of the vertical spreading of the laser sheet in the combustion chamber, strong laser reflexion occurs on the hearth of the furnace. A refractory screen is then set in front of the bottom part of the endoscope to avoid saturation on particle images. The air and gas jets are independently seeded with zirconium oxide particles of 5 μm mean diameter by means of fluidised bed seeding machines. Large recirculation of combustion products associated to the mild flameless combustion regime ensures sufficient seeding of particles outside of the reactant jets. Main drawback is the progressive deposit of ZrO 2 particles on the window used for laser sheet entry. This limits the duration of the experiment and requires the shutdown of the furnace for the cleaning of the window. Figure 6 and Figure 7 present respectively a raw particle image acquired in the FOUR facility in flameless mild combustion regime and the corresponding undistorted image obtained from the unwarping procedure with a constant magnification of 7 pixels / mm. One can see that the collection system no more limits the field of measurements. The latter is rather limited by the size of the laser sheet which spreading in the combustion chamber is limited by the dimensions of the optical window and the width of the refractory walls. One can also observe that even if the FLC shutter eliminates the thermal emission from the high temperature walls, strong background emission is still present on particle images. This is due to laser reflexion on the hearth and on the refractory wall facing the entrance which illuminates the whole of the combustion chamber during laser pulses. The large dynamic range of the CCD camera (12 bits) allows to separate particle scattering signal from background emission. However, such continuous background signal present on both frames of each couple participates to the correlation 4

calculation and then induces underestimation of velocity.

The image resulting from the subtraction of the mean image by the rms one is considered as a background image as it corresponds to the part of the continuous emission on each raw image.

Resulting images have very good contrast as there is no more background signal and still sufficient signal from particles scattering for correlation calculation (Figure 8).

5 calculation and then induces underestimation of velocity. To avoid this bias, mean and root-mean-square (rms) images of particles are calculated after distortion correction for each series of acquisition. The image resulting from the subtraction of the mean image by the rms one is considered as a background image as it corresponds to the part of the continuous emission on each raw image. Then for each frame, the corresponding background image is subtracted to undistorted particle images. Resulting images have very good contrast as there is no more background signal and still sufficient signal from particles scattering for correlation calculation (Figure 8). For each couple of undistorted images with background subtraction, the instantaneous velocity field is calculated with a direct correlation algorithm enabling different shapes for the interrogation windows [12]. In the present application, the shape is adapted to the small width and the high velocity of the turbulent methane jets at the exit of the burner. So, rectangular (32 x 64 pixels) interrogation windows are chosen with an overlapping of 75%. Spurious vectors are filtered in instantaneous velocity field thanks to a global filter, based on maximum and minimum values and a local filter based on the comparison with neighbours, and then interpolated with a Gaussian distribution (3 x 3 pixels) [5]. Figure 9 presents a mean velocity field obtained by endoscopic PIV on the FOUR facility calculated from 1000 instantaneous measurements. This example shows the ability of our optical setup and image processing to obtain large field in high temperature confined combustion chamber. Figure 7. Undistorted particle image obtained with the correction algorithm from the raw particle image of figure 6. Figure 8. Undistorted particle image after background subtraction (in negative gray scale). Figure 6. Example of a raw particle image obtained with the high temperature endoscope in the FOUR facility. Figure 9. Main velocity field obtained by endoscopic PIV on the FOUR facility (colormap represents longitudinal velocity). 5

6 Conclusions In order to apply Particle Image Velocimetry in pilot furnaces, we have developed a high temperature endoscope for the collection of particle images. It consists of a set of 11 lenses, acting as a camera lens in front of a CCD camera, and installed in a water-cooled jacket (700 mm long 78 mm diameter). A nitrogen flow in front of the first lens ensures its thermal protection. This endoscope can be used in combustion furnace with wall temperature up to 1600 K, and requires only one small optical access to collect large particle images. Measurements of velocity fields by PIV in furnace have to combine the small size of the optical access and the need of large measurements areas adapted to the flame scale. With our optical setup, the collection angle is 62 : an area of 300 x 300 mm² can be imaged at a distance of 250 mm from the endoscope. However this induces a very strong barrel distortion of the images. Such effect on particle images has to be taken into account and then requires some adaptations of image processing for the calculation of velocity fields. Indeed, the magnification function is no more linear and no polynomial functions are found to have a sufficient precision to correct once the whole image. Another method has been developed. A squared grid pattern put at the location of the laser sheet is imaged with the complete setup (endoscope, FLC shutter and CCD camera). Undistorted particle images are constructed in successive regions of 70 x 70 pixels corresponding to 10 x 10 mm² in the real object plane, by considering the equivalent trapezium in the raw distorted image. Finally, PIV calculation is performed with a direct correlation algorithm which notably permits to use rectangular interrogation windows in order to combine good spatial resolution and large dynamic of the velocity measurements. A first application of this endoscopic PIV system on a furnace operating in flameless mild combustion regime is presented. It is shown that the developments of the high temperature endoscope and the associated original image processing allow the determination of large velocity fields by PIV in high temperature furnace. Acknowledgements The application of endoscopic PIV on the flameless mild combustion furnace has been done in the framework of the French National project Foyer HPE, coordinated by GDF SUEZ, with the financial support of ANR (National Research Agency) PAN-H Program References [1] D. Honoré. Advanced measurements in industrial combustion systems. In L. Vervisch, D. Veynante, J.P.A.J. Van Beeck (Eds) VKI Lecture Series "Turbulent Combustion", Von Karman Institute for Fluid Dynamics, Belgium, [2] J.M. Most, P. Joulain, L. Gourichon, Joint Meeting of the British and French Sections of the Combustion Institute, Rouen, France, [3] J.M. Most, B. Sztal, J.B. Richon, G. Flament, Proc. 5 th International Symposium on Applications of Laser Techniques to Fluid Mechanics, Lisbon, Portugal, [4] J. Dugué, R. Weber, Proc. 6 th International Symposium on Applications of Laser Techniques to Fluid Mechanics, Lisbon, Portugal, [5] M. Raffel, C.E. Willert, J. Komenhans, Particle Image Velocimetry. A practical guide. Springer, [6] D. Honoré, S. Maurel, A. Quinqueneau, Proc. 4 th International Symposium on Particle Image Velocimetry, Göttingen, Germany, [7] J. Gindele, U. Spicher, Proc. of the 9 th International Symposium on Applications of Laser Techniques to Fluid Mechanics, Lisbon, Portugal, [8] U. Dierksheide, P. Meyer, T. Hovestadt, W. Hentschel, Exp. Fluids 33 (2002) [9] J.C. Russ. The Image Processing Handbook - 3rd Edition. CRC Press & Springer Verlag GmbH, [10] C. Rottier, C. Lacour, G. Godard, B. Taupin, L. Porcheron, R. Hauguel, A.M Boukhalfa, D. Honoré, Proc. European Combustion Meeting, Chania, Crete, [11] C. Rottier, C. Lacour, G. Godard, B. Taupin, L. Porcheron, R. Hauguel, S. Carpentier, A.M Boukhalfa, D. Honoré, Proc. European Combustion Meeting, Vienna, Austria, [12] A. Susset, J.M. Most, D. Honoré. Exp. in Fluids 40 (2006)

Bias errors in PIV: the pixel locking effect revisited.

Bias errors in PIV: the pixel locking effect revisited. E.F.J. Overmars 1, N.G.W. Warncke, C. Poelma and J. Westerweel 1: Laboratory for Aero & Hydrodynamics, University of Technology, Delft, The Netherlands,