Detection of empty grate regions in firing processes using infrared cameras

Transcription

1 11 th International Conferene on Quantitative InfraRed Thermography Detetion of empty grate regions in firing proesses using infrared ameras by J. Matthes*, P. Waibel* and H.B. Keller* *Institute for Applied Computer Siene, Karlsruhe Institute of Tehnology, Hermann-von-Helmholtz-Platz 1, Eggenstein-Leopoldshafen, Germany, joerg.matthes patrik.waibel Abstrat Reently, infrared ameras are used in grate firing proesses for waste and biomass inineration in order to optimize the proess ontrol. These infrared (IR) ameras with a bandpass filter at 3.8 to 4.0 µm allow for a detailed measurement of the solid bed's temperature distribution. Additionally, the urrent load ondition of the grate is of interest for the proess ontrol, whih annot be measured so far. In this paper a new image proessing algorithm is presented, whih extrats detailed information about the load ondition for eah grate zone from the IR-images. The new algorithm is based on a orrelation method using prior domain knowledge (template) in order to inrease the reliability of the alulated parameters. 1. Introdution 1.1 Priniple of amera aided ontrol of grate firing proesses Figure 1 depits the onsidered type of a grate firing ombustion plant whih runs typially with hanging fuels like biomass or waste. Image proessing system Proess ontrol system Fig. 1. Grate firing plant with amera aided ombustion ontrol sheme The fuel enters the grate and passes a drying, pyrolysis, and ignition phase before the burning down begins from the fuel bed surfae. The grate is subdivided into ontrollable zones and lanes. The primary air flows through the grate from below. Sine the primary air does not ensure omplete gas burning out, seondary air is injeted above the fuel bed. The hot flue gas then passes a boiler to produe steam. One of the main ontrol objetives is a onstant steam prodution, but hanging and heterogeneous fuel properties lead to a loal different and non-stationary ombustion. Using additional proess information from speial infrared ameras with a bandpass filter installed at positions 2 or 3 (Fig. 1) ombined with an image proessing system like "ISPECT pro ontrol", the ombustion ontrol an be improved signifiantly [1, 2]. 1.2 Camera tehnologies for monitoring firing proesses Absolute temperatures of the ombustion proess on the grate (primary proess of thermal waste treatment) an only be measured by IR ameras using a bandpass filter in the range of 3.8 to 4.0 μm. Video ameras are subjet to major restritions, as they annot see through flames. Both soot radiation and emissions of gas omponents are nearly negligible in the range of 3.9 μm. Figure 2 exposes the emissivities of the relevant gas omponents in a typial burning hamber atmosphere. Furthermore the influene of soot in the gas phase is depited. Due to the dereasing influene of the soot in the flame, in the IR-range the flames are almost transparent (see Figure 2 right).

2 11 th International Conferene on Quantitative InfraRed Thermography, June 2012, aples Italy VIS µm IR µm Fig. 2. Emissivity of gas omponents, soot with transpareny window (left), VIS and IR image from position 2 (right) [2]. IR ameras with bandpass filter at 3.9 µm are available with miro-bolometer hips. These alibrated ameras an measure absolute temperatures in a range from 400 to 2000 C. A 24/7 operation of the IR amera is ahieved by using a water-ooled borosope and additional purge air whih prevents the lens from being ontaminated. Furthermore, a rejetion unit allows for the amera system to move in and out automatially e.g. in ase of a breakdown of the plant air or water supply. 2. Appliations of amera systems in grate firing proesses 2.1 VIS amera based ontrol of flue gas burnout Fig. 3. VIS amera based detetion of soot streaks in the flue gas Emission of inomplete ombustion produts (e.g. CO and hydroarbons) is limited by legislation. Espeially heterogeneous fuels, suh as domesti waste, may ause loal ombustion onditions to vary onsiderably. The flue gases entering the flue gas burnout zone exhibit signifiant loal variations of the alorifi value and O 2 onentration over the ross setion (i.e. flue gas streaks). Globally ontrolled and homogeneous supply of seondary air over the ross setion of the flue gas burnout zone requires relatively large ombustion hambers to ensure effiient flue gas burnout. IR and/or video amera-based online detetion of the loal burnout status of the flue gas over the entire ross setion of the flue gas burnout zone enables the loal ombustion status to be aptured. Fig. 3 shows the view with a VIS amera from position 2 (Fig. 1) onto the flue gas burning. In the entre, a bright flame indiates a good burnout of the flue gas. In the lower right orner a strong soot streak an be observed, whih indiates an oxygen defiit in this area. In the upper left orner of the image old material on the grate an be seen. Using speial developed image proessing algorithms, the position and size of soot streaks (regions with oxygen defiit) an be deteted online. These harateristi quantities are used for fast ontrol of loal demand-oriented seondary air supply. In this way, optimum flue gas burnout an even be ahieved in relatively small ombustion hambers. In partiular, formation of inompletely burned flue gas streaks under nonsteady-state ombustion onditions an be prevented largely. CO-peaks and elevated onentrations of soot partiles are redued effetively. At low onentrations of soot partiles in the ash deposits on the boiler surfae, formation of PCDD/F (dioxin) is learly redued.

3 11 th International Conferene on Quantitative InfraRed Thermography, June 2012, aples Italy 2.2 VIS amera based ontrol of fixed bed burnout To observe the fixed bed burnout behaviour a VIS amera at the end of the firing grate (position 3, Fig. 1) is installed. Using an online image proessing system, information about length, intensity and area of the burning region an be alulated. This allows for an adaptation of the primary air supply and the transport veloity of the grate to guarantee an effiient arbon burnout of the remaining slag. Figure 4 shows the VIS image and the alulated information by image proessing. The green bordered region indiates the deteted ombustion zone of the fixed bed burnout with the horizontal red line marking the end of the fire. The entre of gravity (red ross) of this region learly outlines disadvantageous fire positions. Additional information about the flame harateristis like shape and dynamis an be extrated by analyzing the blue bordered flame region. Fig. 4. Monitoring of fixed bed burnout with VIS amera 2.3 IR amera based ontrol of main ombustion Employing an IR amera at position 2 or 3 (Fig. 1) with a bandpass filter at 3.9 µm the behaviour of the main ombustion zone in the fuel bed an be observed. After a digital image filtering, aiming at the minimization of the influene of dust partiles and soot in the images, a number of different proess information an be extrated out of the filtered image. This information is transferred to the proess ontrol system and is used for an online adaptation of proess parameters in order to maximize the throughput of the plant as well as to redue the deviations in steam prodution and emissions of pollutants. Typial image based proess information are the mean temperatures of different grate zones, the position and size of the main burning zone in the fuel bed (Fig. 5), the degree of overage as the ratio between freshly injeted and ignited fuel, the transport speed and the residene time of the ombustion material and the dustemitting areas per zone and lane. The IR amera based system for the optimization of the ontrol of the main ombustion is the basis for the new image based detetion of empty grate regions presented in setion 3. Fig. 5 : Grate segmentation and harateristis of the main ombustion zone based on IR amera measurement Other appliations for a amera based optimization an be found in the field of metal reyling [3] or waste inineration in rotary kilns [4], in the ement prodution or online burner flame analysis [5].

4 11 th International Conferene on Quantitative InfraRed Thermography, June 2012, aples Italy 3. Image based detetion of empty grate regions 3.1 Drop of steam prodution due to empty grate regions A onstant steam prodution - even with varying fuel properties - is managed by adapting the feeding. If the urrent steam prodution drops below its set value, the feeding must be inreased and vie versa. Unfortunately, there is a big delay between feeding and steam prodution. Additionally, an exessive inrease of the fuel feeding leads to a overing of the ombustion zone with fresh (and old) fuel and thus to a further drop of the steam prodution. At the moment there is no reliable information available whether there is enough fuel on the grate or not. For this reason and to avoid overing states, the ombustion ontrol works very onservatively and frequently does not inrease the feeding suffiiently, even if parts of the firing grate are empty. This leads to frequent drops of the steam prodution. Figures 6 and 7 show this effet. Already half an hour before the steam prodution starts to drop at 22:15 the IR images show parts of the empty firing grate (espeially the division bars between the lanes). At 22:15 large parts of the grate are not suffiiently overed by fuel whih leads to an signifiant drop of the steam prodution at 22:30. If these situations an be deteted automatially by an image proessing system, these drops in steam prodution ould be avoided. Fig. 6. Evolution of the steam prodution (set point 91 t/h) from 21:30 to 23:30 21:45 22:00 22:15 22: Image proessing for empty grate detetion Fig. 7. IR-Images from 21:45 to 22:30 (saled from 600 C to 1400 C) Here a new image proessing algorithm is presented, whih allows the reliable detetion of empty grate regions. Based on that, parameters are derived that an be used in the ombustion ontrol system. The main idea of the new algorithm is to use a simplified image of an empty grate as template. By orrelating the template with the urrent image in a defined neighbourhood, a new image an be derived, whih ontains the similarity of the image with the template for eah pixel. Figure 8 (upper, left) shows the IR image g whih was geometrially transformed to get a virtual retangular view of the grate (negleting geometri distortion). Additionally, the simplified grate template h is depited (Figure 8, upper right). For the alulation of the orrelation image r for eah pixel (x, y) the orresponding position in the IR image g and the template image h is regarded. The alulation of orrelation is performed using a window around (x, y) with a size of (2M+1) rows and (2+1) olumns (red windows in Figure 8). For the alulation of eah r(x, y) this window is ropped out of the IR image as well as out of the template image. The ropped parts subtrating their mean values and dividing by their standard deviations: g and h are normalized by g g(x, y)-mean(g ) (x, y)= stdev(g ) and h h(x, y)-mean(h ) (x, y)= stdev(h ). (1)

5 11 th International Conferene on Quantitative InfraRed Thermography, June 2012, aples Italy With the normalized ropped parts of the IR image position (x, y) is defined by g and the template image h the orrelation r(x, y) at 1 r (x, y)= (2M 1)(2 1) M m M n g (x n, y m) h (x n, y m). (2) The orrelation value r(x, y) is alulated for all positions and registered into a orrelation image. In this orrelation image high values indiate regions with a strong similarity between the urrent IR image and the empty grate (Figure 8, lower left). ow, the empty grate region ould be deteted using a orrelation threshold. evertheless, in order to integrate the empty grate information into a proess ontrol system it is more reasonable to determine signals that desribe the state of the grate load ontinuously for different grate regions. Therefore the mean value of the orrelation is alulated for user defined regions of the IR image. Conveniently, these user defined regions oinide with separately ontrollable zones of the grate (Figure 8, lower right). filtered and transformed IR image g grate template image h orrelation image r zone 2 r(x,y) zone 3 Fig. 8. Image proessing steps for empty grate detetion 3.3 Results In Figure 9 the values of the empty grate orrelation for zone 2 and zone 3 are depited over time. Additionally the evolution of the steam prodution is plotted. Already half an hour before the steam prodution begins to drop the orrelation for zone 3 is slightly growing. At about 22:05 a signifiant inrease of the orrelation for zone 3 an be deteted. Thus already 15 minutes before the steam prodution starts to derease, the new empty grate orrelation signal for zone 3 signifiantly points at a lak of fuel. Using this information an optimized proess ontrol system an inrease the fuel feeding early and thus prevent the steam prodution from dropping. At 22:20, when the steam

6 mean orrelation with empty grate in % 11 th International Conferene on Quantitative InfraRed Thermography, June 2012, aples Italy prodution begins to derease, also the empty grate orrelation for zone 2 rises rapidly. Even though this signal gives no signifiant advantage onerning temporal aspets, it learly shows that the reason is the lak of fuel zone 2 zone Conlusion Fig. 9. Empty grate orrelation over time for zone 2 and 3 ompared to steam prodution In firing proesses for waste or biomass inineration amera systems an be used for different measurement tasks. Cameras in the VIS range are applied for fixed bed or flue gas burnout monitoring. IR ameras with a bandpass filter at 3.9 µm are used for the measurement of solid bed temperature distribution, beause at this spetral range flames are almost transparent. Using an image proessing system like ISPECT pro ontrol, new signals an be derived online from the amera images and transmitted to the ontrol system for a proess optimization. With this amera based proess ontrol CO-peaks ould be redued by 80% and the CO-ontent by 50%. Additionally the margin of deviation of the steam prodution ould be redued by 10-30%. As a result the degree of effiieny of the boiler ould be inreased by 2% [2]. The proess ontrol an be further improved by using additional amera based information on the load state of the grate. In this paper a new image proessing algorithm is presented, whih alulates new signals from the IR images indiating a low load state of different regions of the grate. These signals provide a good predition for drops of the steam prodution. Utilizing these signals for a better adaptation of the fuel feeding, the deviation of the steam prodution an be further redued. REFERECES [1] Zipser S., Gommlih A., Matthes J., Keller H.B., Combustion plant monitoring and ontrol using infrared and video ameras. In: Pro., IFAC Symposium on power plants and power systems, Kananaskis, Canada [2] Keller H.B., Matthes J., Zipser S., Shreiner R., Gohlke O., Horn J., Shöneker H., "Cameras for ombustion ontrol of highly flutuating fuel ompositions". VGB PowerTeh, 3/2007. [3] Matthes J., Waibel P., Keller H.B., "A new infrared amera-based tehnology for the optimization of the Waelz proess for zin reyling." Minerals Engineering, 24(2011) P [4] Waibel P., Vogelbaher M., Matthes J., Keller H.B., "Infrared amera-based detetion and analysis of barrels in rotary kilns for waste inineration." In: Pro. Quantitative Infrared Thermography 2012, aples (to appear). [5] Gehrmann H.J., Kolb T., Seifert H., Waibel P., Matthes J., Keller H.B. "Co-ombustion of low rank fuels in power plants." 30th Internat. Conf. on Thermal Treatment Tehnologies (IT3) and Hazardous Waste Combustors, Jaksonville, Fla., May 10-13, 2011