Presentation is loading. Please wait.

Presentation is loading. Please wait.

Detection and classification of snow/ice using infrared imaging

Similar presentations


Presentation on theme: "Detection and classification of snow/ice using infrared imaging"— Presentation transcript:

1 Detection and classification of snow/ice using infrared imaging
PhD student: Lavan Kumar Eppanapelli Supervisors: Mikael Sjödahl Johan Casselgren Division of Fluid and Experimental mechanics Luleå university of technology, Luleå Sweden

2 Summary Publications: Conferences: Licentiate:
Eppanapelli, L. K., Friberg, B., Casselgren, J., & Sjödahl, M. (2016). Estimation of a low-order Legendre expanded phase function of snow. Optics and Lasers in Engineering, 78, Eppanapelli, L. K., Casselgren, J., Wåhlin, J., & Sjödahl, M. (2017). Investigation of snow single scattering properties based on first order Legendre phase function. Optics and Lasers in Engineering, 91, Eppanapelli, L. K., Casselgren, J., & Wåhlin, J. (2016). Classification of snow types based on spectral reflectance characteristics in NIR region. Submitted for publication in Cold Regions Science and Technology. Eppanapelli, L. K., Lintzen, N., Casselgren, J., & Wåhlin, J. (2016). Liquid content in snow measured by spectral reflectance. Submitted for publication in Cold Regions Engineering. Conferences: Eppanapelli, L. K., Remote detection of phases of water on a wind turbine blade, presented at Winterwind 2015, Feb 2015 Eppanapelli, L. K., Characterizing different types of snow/ice on a turbine blade using multispectral imaging, presented at EAWE PhD seminar, Apr 2016 Licentiate: Eppanapelli, L. K., Classification of different types of snow using spectral and angular imaging, Luleå University of Technology, Institutionen för teknikvetenskap och matematik, ORCID-id:

3 Problem statement Cold climate often leads to
Complete production loss Reduction of power Overloading Increased fatigue Safety issues There are however techniques to prevent the icing events such as anti-icing and de-icing techniques. More details of these techniques are shown in next slide. Jasinski, William, et al. "Wind turbine performance under icing conditions." 35th Aerospace Sciences Meeting and Exhibit

4 Problem statement (cont.)
Anti-icing methods prevent ice accretion Ice phobic coatings Black paint Chemicals De-icing systems remove/melt the ice of the blades Electrical heating element Warm air and radiator Flexible pneumatic boots Most manufacturers use epoxy or polyester matrix composites reinforced with glass or carbon fibres. Current research heading towards Nano composite coatings . Parent, Olivier, and Adrian Ilinca. "Anti-icing and de-icing techniques for wind turbines: Critical review." Cold regions science and technology 65.1 (2011):

5 Available techniques Several techniques are available to address the presented problem statement. These techniques based on two methods such as direct detection and indirect detection. Direct detection sensors work based on some property change such as reflectance, where mass, conductivity and inductance etc., Indirect detection sensors work based on weather conditions and power curves of turbine. The presented technique relates to direct detection, where variations of intensity of the reflected light from surfaces is investigated. Homola, Matthew C., Per J. Nicklasson, and Per A. Sundsbø. "Ice sensors for wind turbines." Cold regions science and technology 46.2 (2006):

6 Objective Developing an experimental system that can be used to
reliably detect snow/ice be able to classify different types of snow/ice The objective was investigated using spectral and angular imaging. Further explanation regarding the technique explained in next two slides. Reliable detection of snow/ice to optimize the de-icing system Classification of snow/ice provides additional information for example to simulate/model ice load and ice accretion on the turbine blades.

7 Angular imaging Detector
10 20 30 40 50 70 80 -10 -20 -30 -40 -50 -60 -70 -80 60 Fig 1: Illustration of angular imaging of intensity of reflected light from snow/ice surface. Illumination source fixed at -45°. Measuring intensity of reflected light from surface in all the angles in hemisphere above the surface, gives an interpretation of sample surface. The plots represents the angular distribution of intensity for fresh snow and ice. Fresh snow has more uniform distribution of reflectance where ice exhibits forward scattering behavior. This is because, ice has more smoother and flatter surface.

8 Spectral imaging Fig 2: Absorption spectrums for fresh snow and ice.
Along with the angular distribution, spectral distribution of light is also considered. Spectral distribution corresponds to the selected band of wavelengths. In this case from 900 nm to 1650 nm. Spectral imaging The Figure 2 show how much light of specific wavelength is being absorbed by fresh snow and ice sample. One can observe that fresh snow exhibits higher degree of reflectance while ice exhibits rather low reflectance as ice was composed frozen water and air that increases the absorption characteristics. Fig 2: Absorption spectrums for fresh snow and ice.

9 Approach-I Here, experimental setup to characterize snow/ice according to their physical properties is investigated. Intensity of reflected light is measured at several angles and several bands of wavelengths to find optimal wavelength bands where the classification and characterization is best possible.

10 Measurements Fig 3: Experimental setup using a NIR camera and laser diodes A near-infrared camera and light source of three wavelengths (980 nm, 1310 nm and 1550 nm) were used in this setup

11 Measurements Fig 4: Experimental setup using a spectrometer and broadband light source A near-infrared spectrometer and broadband light source were used in this setup. Results shown in next few slides before Approach – II were obtained using this setup.

12 Snow types Spring snow (large grains) Granular snow (small grains)
Fresh snow Compacted snow (ice)

13 Spectral measurements
980 nm 1310 nm 1550 nm Fig 5: Absorption spectrums for all snow types Figure 5 shows how much degree of absorbance varies for each sample. These snow/ice samples were prepared and they were aged differently and compacted at certain pressure. One more interpretation was that, snow reflectance decreases as density increases. One can also observe that three bands of wavelengths 980 nm, 1310 nm and 1550 nm, appear to be optimal where the samples show distinct absorption characteristics.

14 Angular measurements Fig 6: Angular behavior of reflected light for Fresh (a), aged (b) and compacted snow (c). Light source was fixed at an angle of -45° where 0° is nadir. Figure 6(a) shows angular distribution of reflectance for fresh snow at two wavelength. It shows that fresh snow has uniform distribution over all the angles. Aged snow exhibits strong backward scattering probably due to irregularly oriented large grains reflect light in the origin direction, see Figure 6(b). In Figure 6(c), compacted snow shows strong forward scattering, as the surface is smooth and flat.

15 Analytical solver Radiative transfer equation for a plane-parallel medium where, Θ is the angle between incident and scattered light s represents the optical distance and k is absorption coefficient. P(cosΘ) determines the angular distribution of intensity scattered by snow at a given wavelength. Light of certain wavelength travels an infinitesimal depth inside snow. In that depth, it can be scattered or absorbed. One needs to know scattering phase function as a function of wavelength to solve RTE.

16 Analytical solver (cont.)
Phase function can be expanded into Legendre polynomials Legendre coefficients (ωl) are the output from the solver, and subject to constraints Only first two coefficients provide significant information. ω0 represents single scattering albedo ω1 represents the asymmetry parameter Light of certain wavelength travels an infinitesimal depth inside snow. In that depth, it can be scattered or absorbed. One needs to know scattering phase function as a function of wavelength to solve RTE.

17 Results 1550 nm 980 nm 1310 nm Angle Forward Fig 7: Classification of snow types with different physical properties Samples with individual grains are at one place, and possible to distinguish themselves too. Samples with higher density are observed to be at one place. It was also observed that there is a linear correlation between the coefficients and density . Similar snow with slightly different surface texture is also distinguishable from other samples

18 Results The approach can be used to classify snow with different physical properties sing only two coefficients Snow grain structure Density Surface texture One can make quick qualitative conclusions using only two coefficients from the solver. Intensity measurements resolved in both angular and spectral, takes about 5 minutes for a given snow sample Solver takes about 30 seconds

19 Approach-II Here, experimental setup to characterize different phases of water on a piece of wind turbine blade is presented. Laser diodes of three wavelength bands 980 nm, 1310 nm and 1550 nm, were used as a light source and the NIR camera used as a detector. The experimental setup for these measurements is similar to Figure 3.

20 Samples Moist Frozen moist Wet Ice water Ice air Clear ice

21 Images Moist Ice air Clear ice
Fig 8: Raw images of reflected light of specific wavelengths in the dark Images obtained from the camera and each spot represents a specific wavelength. One can observe the variations in intensity at different conditions. This is due to distinct absorption characteristics of each sample/condition.

22 Results Fig 8: 3D-plot of the classification of phases of water
One can observe how icing conditions separated from each other and distinguishable using the approach. A better color coded illustration of different icing conditions is shown in next slide.

23 Results (cont.,) Fig 9: Visualization of classification based on RGB concept. Intensity values at three wavelength bands were converted into a RGB matrix based on some normalizing calculations. One can observe from Figure 9 that color coded interpretation of each icing condition is possible. For example, wet condition shows black, which is actually means that all the light is being absorbed.

24 Conclusions The approach can be used to
Classify snow types with different physical properties Snow grain structure Density Surface texture Classify different phases of water on a piece of wind turbine blade An illustration of color scale shows that the technique is promising.

25 Future work A new set of measurements will be performed in spring 2017, on a larger piece of wind turbine blade using three laser diodes and NIR camera. A plan to make similar measurements on the wind turbine on roof at LTU, will be also going to be performed before summer.

26


Download ppt "Detection and classification of snow/ice using infrared imaging"

Similar presentations


Ads by Google