Coal/biomass Combustion Prediction using Image Analysis Methods. Tao Wu Edward Lester Claudio Avila. Coal Research Forum April 2008

Transcription

1 Coal/biomass Combustion Prediction using Image Analysis Methods Tao Wu Edward Lester Claudio Avila Coal Research Forum April 2008

2 History of Image Analysis Manual Methods - based on modal analysis, first applied to coal in 1934 by Glagolev, but originated by Delesse in the late 19th century. Photomultiplier systems - the flying spot microscope and photomultiplier system (Roberts and oung, 1952) was the beginning of automated systems. Image Analysis Systems - a progression in technology led to the replacement of the photomultiplier with a video camera. The image under the microscope is then fed to the image analysis system for analysis.

3 Point Counting An automatic point counter keeps a running total of the number of points out of 500 labelled for each maceral type. The movement of the stage is automated, and the distance moved each time is constant. Lineal analysis uses an automated stage which moves the block under the eyepiece using a series of micrometer spindles (Krevelen, 1961, Galehouse, 1971). Each maceral type has its own spindle which is used to pass the cross hairs over the particle.

4 Advantages of photomultiplier systems Early systems could analyse 106 points in 20 minutes (Denton 1967). The system from the Nippon Steel company was developed ten years later and compromised speed for increased accuracy (Kojima., 1976). 20,000 points required 30 minutes and so it is still preferable to manual methods.

5 Advent of IA Image analysis systems are fast reliable and do not require a skilled operator to run on a day to day basis. Each captured image can be manipulated to correct halo effects and uneven illumination. Image analysis systems can perform the same functions as a photomultiplier, as well as being able to attempt microlithotype analysis (Chao et al., 1982a, Crelling, 1982) liberation analysis (Finch & Gomez, 1989) and association analysis (Vleeskens et al., 1984).

6 IA processing Image Capture Image Digitisation Image Manipulation Image Segmentation Image Quantification

7

8

9 IA processing Image Capture Image Digitisation Image Manipulation Image Segmentation Image Quantification

10 Finding particles Problem with finding liptinite Morphological associations Fluorescence Colour Image analysis on coloured resins Other methods

11 Red resin

12 Green resin

13 Image at 320ms Image at 750 ms Sharpened image

14 Particle Mask Clean particle mask Clean non touching particle mask

15 Separated particles mask individual Particle mask Separated particle

16 IA processing Image Capture Image Digitisation Image Manipulation Image Segmentation Image Quantification

17 Frequency (pixels) Liptinite Vitrinite Inertinite Grey Scale

18 Particulate analysis Isn t this established already? What s the difference between particulate analysis and bulk analysis?

19

20

21 Char morphology & Classification Porosity >40% Unfused <25% Ash <50% Porosity <40% Unfused 25-75% Unfused >75% Tenuisphere Porosity >60% Porosity>40% 5%<Porosity<40% Porosity <5% Ash >50% Crassisphere Mixed-Porous Mixed-Dense Inertoid Fusinoid/Solid Tenuinetwork Porosity <40% Mineroid Crassinetwork

22 Char image analysis procedure

23 Burnout Modelling of Biomass/Coal Blends

24

25 Burnout History ChBa Cumulative Carbon Conversion Time Elapsed /s Kotinskaya Kyrgaiskaya Prokopijevskiy Tugnuisky Erunakovsky Taldinsky Skolovskaya 7thNovember

26 COAL with BIOMASS

27 Different Proportions of biomass CCP/Daw Mill coal Rate of Weight Loss (dw/dt) Dawmill CCP 5% CCP 10% CCP 15% CCP 20% -1.6 Temperature (ºC)

28 Proportion Proportion predicted and difference, % PKE Sawdust Olive cake CCP 5% 6 (+1) 7 (+2) 6 (+1) 7 (+2) 10% 10 (0) 11 (+1) 8 (-2) 11 (+1) 15% 13 (-2) 16 (+1) 12 (-3) 14 (-1) 20% 16 (-4) 20 (0) 15 (-5) 16 (-4)

29 Cumulative Wrongness Index for finding the best fit 250 Absolute Variation from Actual Sawdust Olive Cake Cereal PKE Prediction Percentage (%)

30 Typical images of coal and biomass chars Olive cake char 2-3mm PKE char 2-3mm

31 Characterisation of Char Morphology Average porosity of pure biomass and coal samples Size fractions (mm) Average porosity (%) CCP OC PKE Daw Mill

32 1.0 CCP + Daw Mill 0.8 Carbon Burnout Time Elapsed /s 0% Daw Mill 5% CCP 10% CCP 20% CCP 30% CCP

33 1.0 Olive Cake + Daw Mill 0.8 Carbon Burnout Time Elapsed /s 0% Daw Mill 5% Olive Cake 10% Olive Cake 20% Olive Cake 30% Olive Cake

34 ADVANCED COMBUSTION MODELLING Virtually modelling free!

35

36 Formation of Different Chars Vitrinite + <5%Inertinite Vitrinite + <50%Inertinite Char Types Tenuisphere Inertoid Tenuinetwork Solid

37 How to Predict Char Morphology? N N N N N Inertinite > 0.70? Inertinite > 0.50? Inertinite > 0.05? Peak Position >90? Size <50microns? Porosity >20% or noise on DT of particle significant? Peak position >190? Mixed-Dense Inertoid Solid Inertinite Fragment N >2? Peak position >110? Inertinite fragments>3? Fragments close to perimeter? Size >60microns? Unreactives>0.5 Unreactives>0.25? Mixed-Porous Crassisphere Crassinetwork Crassinetwork Crassisphere Tenuinetwork Mixed-Porous Peak position >110? Inertinite < 0.25? Peak position >90? Unreactives < 0.20? Unreactives < 0.25? Unreactives < 0.25? Crassisphere Crassinetwork Crassisphere Crassinetwork Crassisphere Tenuinetwork Crassisphere Mixed-Porous Tenuisphere Tenuinetwork Tenuisphere Unreactives<0.05? Unreactives<0.25? Size <60microns? Size <80microns? N Size <80microns? Mixed-Porous Crassinetwork Crassinetwork N N N N N N N N N N N N N Crassinetwork N N N N N Crassinetwork

38 REAL AND FAKE?

39 Predicted Char Morphology

40 Predicted Char Burnout History 25% 50% 75% 95%

41 Char Burnout History Carbon conversion % Burnout Stage /iteration Particle 1 Particle 2 Particle 3 Particle 4

42 Conclusions Image analysis plays a key role in Coal characterisation Char Characterisation Combustion Modelling IA essentially is quantitative rather than qualitative IA is not perfect but can be based around rules from operators (expert systems) IA is more consistent than manual operators

43 acknowledgements BCURA B77