WO2008052542A1 - Système et procédé pour prédire l'émission de nox et/ou la concentration en chaux libre dans un four à ciment - Google Patents

Système et procédé pour prédire l'émission de nox et/ou la concentration en chaux libre dans un four à ciment Download PDF

Info

Publication number
WO2008052542A1
WO2008052542A1 PCT/DK2007/000180 DK2007000180W WO2008052542A1 WO 2008052542 A1 WO2008052542 A1 WO 2008052542A1 DK 2007000180 W DK2007000180 W DK 2007000180W WO 2008052542 A1 WO2008052542 A1 WO 2008052542A1
Authority
WO
WIPO (PCT)
Prior art keywords
measurement
kiln
data
image data
cement kiln
Prior art date
Application number
PCT/DK2007/000180
Other languages
English (en)
Inventor
Bao Lin
Bodil Recke
Torben Mønsted SCHMIDT
Jørgen K. H. KNUDSEN
Sten Bay JØRGENSEN
Original Assignee
Fls Automation A/S
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Fls Automation A/S filed Critical Fls Automation A/S
Publication of WO2008052542A1 publication Critical patent/WO2008052542A1/fr

Links

Classifications

    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
    • G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
    • G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
    • G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance

Definitions

  • a system and a method for prediction of NOx emission and/or free lime concentration in a cement kiln A system and a method for prediction of NOx emission and/or free lime concentration in a cement kiln.
  • the present invention relates to a method for determining one or more cement kiln process parameters, the method comprising the steps of obtaining at least one process measurement representing at least one process parameter and obtaining at least one item of image data representing a burning zone.
  • the invention also relates to a corresponding device. Further, the invention relates to a computer program, a computer readable medium and a rotary cement kiln system.
  • Another method to obtain measurements on chemical processes that are not easily performed using conventional process measurement methods is to use soft sensors and/or data fusion.
  • a soft sensor is a common name for software in which several measurements of one or more process parameters performed by one or more physical sensors may be processed together to calculate (predict) the value of one or more new process parameters. The new quantity thus need not be measured directly.
  • Multi-sensor data fusion In data fusion, sensory data from disparate sensors may be combined such that the resulting information provide a more accurate, more complete and/or more dependable result than would have been possible from using the sensors individually.
  • Multi-sensor data fusion is known from Varshney, P. K., “Multisensor Data Fusion", Electronics and Communication Engineering Journal 9(6): 245-253, 1997, and has been used in various fields such as robotics, automatic sensing and automatic pattern recognition.
  • Various techniques for implementing data fusion are available such as digital signal processing (DSP), fuzzy logic, artificial intelligence and statistical estimation.
  • Soft sensors can generally be classified into two categories: Model-based and data-driven soft sensors.
  • a soft sensor may also be known as a virtual sensor.
  • Model-based soft sensors may use artificial intelligence to perform e.g. pattern recognition and may therefore require high levels of computational intensity.
  • data-driven soft sensors advantage is taken of modern measurement techniques enabling a large amount of operating data to be collected, stored and analyzed.
  • data-driven soft sensors high dimensionality of data may show strong collinearity and therefore multivariate analysis may be used to extract process information that is statistically significant.
  • process data may come from a variety of sensors in a variety of formats which may turn the use of soft sensors into a problem. Further, process measurements may be contaminated with outlying observations making the derivation of an inferential regression model problematic.
  • the present invention solves, among other things, at least one of the abovementioned objects by a method for determining one or more cement kiln process parameters, wherein the method comprises the step of performing a partial least squares regression on said obtained at least one process measurement and said obtained at least one item of image data to calculate a soft sensor output indicative of one or more process parameters.
  • the method projects the process measurements into a smaller number of orthogonal latent variables thereby enabling the method to provide a determination of one or more difficult accessible cement kiln process parameters and to overcome the problem of high-dimensionality and/or collinearity of the combined obtained data/measurements and image data.
  • said partial least squares regression is calculated dynamically.
  • a dynamic calculation of PLS (DPLS) regression is performed by augmenting the process measurement data block with time lagged variables.
  • the method comprises a step of pre-processing said one or more process measurements.
  • a pre-processing step may for example be performed by removing outlying data points from the process measurements.
  • the method is able calculate a soft sensor output of a process parameter and thus to determine a process parameter in spite of process measurements comprising contaminated data e.g. outlying data points.
  • the method comprises a step of filtering the at least one item of image data in a time-domain.
  • the time domain filtering may be performed by a low-pass filter whereby image features with a suitable long time constant may pass the filter.
  • the time domain filtering may be performed by a band-pass filter whereby image features with a time constant in a certain pass band may pass the filter.
  • the time domain filtering may be performed by a moving average filter (simple, weighted, exponential, etc.).
  • the filtering may be performed by any type of filter or a combination of filters.
  • the method comprises a step of performing multivariate image analysis to extract image features for use in the partial least squares regression.
  • Multivariate images may be used to represent, for example, complex systems that have spatial, intensity, spectral and time (temporal) resolutions.
  • Multivariate statistical analysis approaches may efficiently compress original highly correlated data.
  • principal components analysis is a multivariate technique that projects an original image data onto a reduced dimensional space through linear combinations of the original variables. Performing multivariate image analysis enables, for example, the method to handle image data showing strong collinearity.
  • the method comprises the step of performing multivariate image analysis comprising using at least one higher than first order moment in said multivariate image analysis.
  • Using one or more higher than first order moments in the multivariate image analysis enables the method to handle e.g. colour features of image data.
  • the one or more process parameters determined by the soft sensor output comprises cement kiln NOx emission and/or free lime concentration.
  • the at least one process measurement comprises one or more items chosen from the group of flame temperature measurement, kiln drive current measurement, kiln feed measurement, fuel to calciner measurement and fuel to kiln measurement.
  • Embodiments of the present invention also relate to a device for determining one or more cement kiln process parameters and embodiments thereof corresponding to the method and embodiments thereof comprising means for obtaining at least one process measurement representing at least one process parameter and means for obtaining at least one item of image data representing a burning zone.
  • the invention relates to a device comprising means for performing a partial least squares regression on said at least one obtained process measurement and said at least one item of image data to calculate a soft sensor output indicative of one or more process parameters.
  • the invention also relates to a computer program comprising program code means adapted to cause a data processing system to perform the steps of the method according to the invention when said program code means are executed by said data processing system.
  • the invention relates to a computer readable medium having stored thereon a computer program comprising program code means adapted to cause a data processing system to perform the steps of the method according to the invention when said program code means are executed by said data processing system.
  • the invention relates to a rotary cement kiln system adapted to perform the steps of the method according to the invention in order to provide a soft sensor output indicative of one or more process parameters of said rotary cement kiln system.
  • Figure 1 is a schematic drawing representing a rotary cement kiln comprising a temperature sensor and a camera.
  • Figure 2 is a flowchart representing the steps of a multivariate image analysis method for extraction of feature in an image.
  • Figure 3 is a schematic drawing representing a system for statistical data fusion in a soft sensor.
  • Figure 4 is a schematic drawing representing a device for combining one or more measurements from one or more sensors and/or a device for performing multivariate image analysis and/or a device for performing a partial least squares regression.
  • Figure 5 is a schematic drawing of a rotary cement kiln comprising a soft sensor for determining a process parameter.
  • Figure 6 is a flowchart representing the steps of a partial least squares regression.
  • Figure 7 represents a graph of values of the PRESS parameter versus the order of a dynamic partial least squares regression.
  • Figure 7 b) represents a graph of a validation curve for a CaO/lime soft sensor not combining process measurements with image features.
  • Figure 8 a) represents a graph of a modelling curve for a CaO/lime soft sensor combining process measurements with image features.
  • Figure 8 b) represents a graph of a validation curve for a CaO/lime soft sensor combining process measurements with image features.
  • Figure 9, a) represents a graph of a validation curve for a NOx soft sensor not combining process measurements with image features.
  • Figure 9, b) represents a graph of a validation curve for a NOx soft sensor combining process measurements with image features.
  • a rotary cement kiln may comprise a cylindrical steel shell with a diameter d and a height h, (110), a feed in, (120), a burner pipe, (130), providing fuel to a flame, (140); and a cement clinker output, (150).
  • the steel shell, (110) may be formed as a tube sloping slightly (approximately 1 - 4°) rotating at approximately 30 to 250 revolutions per hour.
  • Material is input into the kiln feed at (120).
  • the input material may comprise a mixture of limestone and clay. Due to the slope and rotation of the kiln, the input material will move towards the flame, (140).
  • the flame, (140) is provided by a burner pipe, (130).
  • the burner pipe may provide a fuel such as oil, coal, or any other combustible fuel to the flame. (180) represent a rotational direction of the rotary cement kiln.
  • the rotary cement kiln may further comprise at least one camera, (160), and/or at least one temperature sensor, (170).
  • the at least one camera, (160) may be situated at the base of the burner pipe, (130), in the steel shell, (110), as indicated in the figure. In this position the camera may be in a cold position relative to the temperature of the rotary cement kiln and/or be in a close position to the flame and/or burning zone. Thereby, the camera may record one or more images of the flame and/or burning zone and provide image data of the flame to at least one other piece of equipment e.g. through a data connection and/or an optical connection and/or a Bluetooth connection and/or an infrared connection. Alternatively or additionally, the at least one camera may be situated near the centre of the cylindrical height h of the steel shell in order to observe and record images of both input/feed material and the flame.
  • the at least one camera may be situated anywhere in the kiln.
  • the at least one camera, (160) may be any type of camera such as a RGB camera e.g. based on a CCD photosensitive chip or a CMOS photosensitive chip.
  • the camera may be a CMYK camera e.g. based on a CCD or CMOS photosensitive chip.
  • the camera, (160) may be an infrared sensitive camera.
  • the camera, (160) may be a ultraviolet sensitive camera.
  • the at least one camera may be a video-camera.
  • the camera may be any type of camera and/or video-camera.
  • the rotary cement kiln may comprise, for example, a RGB camera and an infrared camera, said two cameras may, for example, be positioned at the base of the burner pipe.
  • the at least one temperature sensor, (170) may be situated in close proximity to the flame, (140), as indicated in Figure 1. Alternatively or additionally, the at least one temperature sensor may be situated in close proximity to e.g. the camera, (160). Alternatively or additionally, the at least one camera may be positioned at the opposite end of the camera e.g. at the position where material is input into the rotary cement kiln, (120). Alternatively or additionally, the at least one camera may be positioned in a gas stream of the cement kiln. Alternatively or additionally, the at least one camera may be positioned anywhere in the rotary cement kiln.
  • the at least one temperature sensor, (170) may be any type of temperature sensor, such as for example an electrical resistance thermometer, an infrared thermometer, a liquid crystal thermometer, a silicon bandgap temperature sensor, a thermistor, a thermocouple, or any other type of temperature sensor.
  • the temperature sensor, (170) may provide one or more readings to at least one other piece of equipment for example through a data connection and/or an optical connection and/or a Bluetooth connection and/or an IR connection.
  • Data from the at least one camera and/or the at least one temperature sensor may for example be transmitted to a computer (for example a PC, a mainframe, a parallel processing unit and/or any other digital processing unit), (190), via a data connection, an optical connection, an IR connection and/or a Bluetooth connection between the abovementioned devices.
  • Said computer may perform data analysis on said received data.
  • the computer may perform multivariate image analysis (MIA) to extract features from image data produced by the at least one camera, (160).
  • MIA multivariate image analysis
  • the image data from e.g. a RGB camera may show strong collinearity which may be an additional reason to use MIA on e.g. RGB image data of the flame.
  • the computer may perform a PLS regression on said image data and said temperature data.
  • a MIA method for extraction of features in a RGB image performed e.g. in a computer (e.g. a PC), a digital signal processor (DSP), a mainframe and/or any other device capable of performing MIA, is illustrated.
  • the MIA framework has been described in e.g. Paul Geladi and Hans Grahn, "Multivariate Image Analysis", John Wiley & Sons Ltd, 1997; M. H. Bharati and J. F. MacGregor “Multivariate Image Analysis for Real-Time Process Monitoring and Control", Industrial and Engineering Chemistry Research, 37, 4715 - 4724, 1997.
  • a MIA method may comprise the steps of:
  • step (210) the computer receives, e.g. through a framegrabber and/or a digital video recorder, RGB image data from the camera, (160), said image data comprising data representing said flame, (140).
  • step (210) the computer receives, e.g. through a framegrabber and/or a digital video recorder, RGB image data from the camera, (160), said image data comprising data representing said flame, (140).
  • step (220) the computer transforms the RGB triplet [R G B] n n
  • step (230) the computer applies a kernel principal component analysis (PCA) to a 3 by 3 covariance matrix of R, G and B vectors to construct a plane using a plurality of scores (principal components), for example the first 2 or more scores. Pixels with nearly identical scores, (t ⁇ t 2 ) values, overlap each other in a plot.
  • PCA kernel principal component analysis
  • step (240) the computer constructs a 256 by 256 histogram (normal 24-bit RGB images have a resolution between 0 and 255 for each channel). At given operational conditions of the cement kiln, the histogram remains stable.
  • step (250) the computer defines one or more masks (e.g. areas) on the score space (e.g. (V 1 t 2 )) for selection of specific features, such as for example color features, in order to detect regions of interest, such as the flame, the luminous and the background areas.
  • masks e.g. areas
  • V 1 t 2 the score space
  • specific features such as for example color features
  • step (260) the computer calculates features related to the color and luminous area, including the average color of the luminous area and the whole image, the size, the brightness and the uniformity of the luminous area.
  • step (265) the method ends.
  • Each feature among a plurality of features may have an individual magnitude. Therefore, the magnitude of a feature may require a normalization i.e. the original value of said feature may have to be normalized.
  • variations may be observed from image to image due to the possibly turbulent nature of the flame. Since the variation of one or more process conditions may occur at a larger timescale than said variations in the images, filtering may be required in order for an image feature to be used as data in a soft sensor. For example, a time-domain filter may be used to extract features in e.g. a time series. The filtering window length may be used to determine the relative weighting factor of an image feature to a conventional process measurement such as for example temperature, pressure, kiln drive torque, etc..
  • the multivariate image analysis of image data representing, e.g. a flame in a kiln, color features may comprise higher order moments.
  • the second-, third- and fourth-order moments of a RGB image may be used to characterize each color channel (Red, Green, Blue in the case of an RGB image) of an image.
  • Equation (1) represents the first order moment
  • equation (2) the second order moment
  • equation (3) the third order moment
  • equation (4) represents the fourth order moment of a RGB image.
  • RGB Red-Green-Blue
  • a Hue-Saturation-Intensity (H, S, I) model may decompose color and intensity, thereby enabling color features to be extracted reliably.
  • the transformation between a RGB model and a HSI models may be described as:
  • I is a parameter representing the intensity
  • S is a parameter representing the saturation
  • H is a parameter representing the hue.
  • the 1 st to 4 th order moments disclosed above under equations 1) - 4) may be used to determine moments for the components (H, S, I) of the HSI model by exchanging k e ⁇ R , G , B] with k e ⁇ H,S,l ⁇ .
  • the 1 st to 4 th order moments of the HSI components may be included in a set of image features similarly to the RGB components disclosed above.
  • HSI color data may be used as image data input into the PLS and/or DPLS model for determining one or more cement kiln process parameters.
  • a RGB image and/or a HIS image may suffer from saturation of one or more pixels i.e. a value of 255 in one or more of the color channels in an 8 bit image.
  • the variation of kiln temperature may be captured through non-saturated pixels for example surrounding one or more saturated pixels.
  • a camera with a larger color depth may be used to provide additional valuable information regarding the flame.
  • a data-driven soft sensor is an inferential model developed from process observations that are available. Multivariate regression techniques may be employed to develop an empirical model. The combined image features and process measurements commonly result in a high-dimensional input matrix that is strongly collinear. Multivariate linear regression (MLR) suffers from numerical problems as well as degraded models when applied to a collinear data set.
  • MLR Multivariate linear regression
  • PLS partial least squares
  • PLS is a multivariate statistical approach for relating an input data matrix (representing independent variables), X , to a dependent data block (representing dependent variables), Y .
  • one column in the matrix X may for example represent the kiln temperature
  • another column may represent the kiln drive current, etc.
  • One column in the matrix Y may for example represent free lime in the clinker and another may for example represent NOx emission from the cement kiln.
  • Y may, for example, represent dependent variables being sampled at the same frequency.
  • one column in the matrix Y may represent free lime (CaO) and another may represent tricalcium silicate (C 3 S), both for example being sampled approximately once every 2 hours.
  • one column in the matrix Y may represent NOx and another may represent O 2 , both for example being sampled at a frequency of approximately 10 seconds.
  • PLS regression establishes a model relating a set of dependent variables, Y , to a set of independent variables, X .
  • the model may be used to, for example, predict values, ⁇ , of the dependent variables, Y , based on (new) values of the independent variables, X .
  • PLS can be thought of as a simultaneous PCA decomposition of X and Y as:
  • T , P and U , Q are the so-called score and loading matrices of X and Y , respectively.
  • the columns of T are generally known as latent variables (LV) and may be calculated to maximize the covariance between dependent variables (Y) and regressors (X).
  • NPALS Non-iterative Partial Least Squares
  • (6) is the regression vector, and W is a weighting matrix maintaining orthogonal scores.
  • DPLS dynamic PLS
  • N may represent a number of samples used to derive the DPLS model
  • p may represent a number of lagged terms
  • the regression relationship may then, for example, be:
  • f DPLS (k) ⁇ c u - ⁇ ⁇ k- J) + c 2J - ⁇ 2 (k - j) + - + c nJ . ⁇ n (k- j)
  • the prediction of the dependent variable may be a weighted summation of the current and previous process variables x(k) and thus, the issue of auto- correlations in process data may be addressed.
  • a soft sensor based on DPLS is able to provide a smooth estimation compared to a similar estimation obtained using a static regression model.
  • the number of lagged terms to use may be determined by minimizing the PRESS variable, as defined below, of a sequence of DPLS models.
  • a DPLS model of a system may provide a smoother estimation than a PLS model for the same system, such as for example a cement kiln system.
  • Predictive significance may be used as a test for choosing the number of LVs and/or lagged terms. It is determined by the minimum PRESS (one-step- ahead prediction residual sum of squared errors), which evaluates the difference between a PLS and/or DPLS model and a process measurement on validation data:
  • N r is the total number of samples in a validation data set
  • y is the predicted value
  • y m is the measured value. The smaller the PRESS value, the better prediction.
  • Process measurements may be contaminated with data points that deviate from real values for example due to instrument failure. Since outlying observations may degrade the estimation of a regression model, data preprocessing may be used to remedy/remove such contaminated data points.
  • the PLS and/or DPLS analysis may, for example, be performed on pre- processed (optional) cement kiln process measurement, e.g. the kiln flame temperature, and on image features extracted using MIA on the kiln flame as disclosed above.
  • the PLS and/or DPLS regression may provide a soft sensor whose output may be used to determine one or more process parameters of the cement kiln.
  • the process parameters determined by said soft sensor may for example be a process parameter not easily assessed.
  • a system for statistical data fusion in a soft sensor comprises one or more sensors, (310), (320) and (330).
  • One or more sensors may for example be a temperature sensor.
  • One or more other sensors may for example be a RGB camera.
  • one or more sensors may for example be a kiln drive current sensor.
  • one or more sensors may for example be chosen from the group of sensors comprising a kiln feed sensor and a calcinations fuel sensor.
  • Module (340) represents means for preprocessing the data output from the one or more sensors, for example the sensor represented by (310).
  • the means for preprocessing may be omitted if for example no contaminated data points will occur in the data output from the sensor, (310).
  • the means for preprocessing may be used to remove/remedy contaminated data points in a data series from a sensor, (310).
  • the means for preprocessing may be used for any sort of preprocessing.
  • Means for preprocessing (350) may preprocess data from the sensor represented by (320) and likewise means for preprocessing (360) may preprocess data from the sensor represented by (330).
  • a single means for preprocessing may preprocess data from a plurality of sensors.
  • the one or more means for pre-processing may for example be a computer, such as a personal computer, a DSP, a mainframe, etc.
  • the means for preprocessing may, for example, identify said contaminated data points in a data series using a PCA method, followed by a down-weighting of said contaminated data points.
  • the measurements from the one or more sensors may be stored.
  • the measurements from the one or more sensors may be combined in module (380) e.g. into forming a variable representing the combined measurements.
  • Module (380) may for example be a hard-disk, an optical disk, flash memory or any other type of non-volatile memory.
  • module (380) may be a random access memory, static or dynamic, or any other type of volatile memory.
  • module (380) may be a computer in which the processing part of the computer performs the combination of the measurements from the one or more sensors.
  • Module (380) may for example be a device as represented by figure
  • Module (370) may represent a block of memory in which dependent variable data, which is to be regressed/correlated/related to a measurement/combined measurements from one or more sensors, is stored.
  • the dependent variable data stored in module (370) and the combination of measurements from one or more sensors stored in module (380) are input into module (390).
  • Module (390) may be the same device as used to combine the measurement in module (380) and which may be represented by a device as in figure 4.
  • module (390) is another device of the type represented by figure 4.
  • module (390) is a computer, such as a personal computer, a DSP, and/or a mainframe.
  • module (390) PLS and/or DPLS regression is performed on the dependent variable from module (370) and the combined measurements from the one or more sensors from module (380).
  • the regression may for example be performed as disclosed above.
  • the output from module (390) is a regression model that may be used to, for example, predict operation conditions of e.g. a kiln from measurements taken on the kiln, e.g. temperature measurements and images.
  • the regression model from module (390) may be used to control one or more processes in e.g. a cement kiln.
  • the device (400) comprises one or more micro-processors (401) connected with a main memory (402) and e.g. a storage device (406) via an internal data/address bus (404) or the like.
  • the device (400) may also be connected to or comprise a display (407) and/or communication means (403) for communication with one or more remote systems via a network.
  • the memory (402) and/or storage device (406) are used to store and retrieve relevant data together with executable computer code for providing the functionality according to the invention.
  • the micro-processor(s) (401) is responsible for generating, handling, processing, calculating, etc. the relevant parameters according to the present invention.
  • the storage device (406) which may be optional, comprises one or more storage devices capable of reading and possibly writing blocks of data, e.g. a USB-slot for memory cards, DVD, CD, optical disc, PVR, etc. player/recorder and/or a hard disk (SCSI, IDE, ATA, etc), floppy disk, smart card, PCMCIA card, etc.
  • the device (400) may receive data from e.g. (370) and/or (380), and store the received data in the main memory (402) and/or the storage device (406).
  • the one or more processors (401) of the device may retrieve the data from the main memory and/or the storage device (406) and using a piece of software, e.g. a suitable computer program, correlate the measurement data and/or image data with the dependent variable, (370).
  • FIG 5 a schematic drawing of a rotary cement kiln, (500), is presented.
  • the cement kiln comprises a rotary kiln, (510), a burner pipe, (520), providing a flame, (530), access to fuel.
  • the kiln comprises a plurality of preheaters, (540), supplying feed material to the rotary kiln.
  • preheaters Alternatively, there may be a single preheater. Alternatively, there may be no preheaters in which case the feed material is input directly into the rotary kiln, (510).
  • the feed material in the rotary kiln moves in the direction of the arrow indicating "solid" in the figure.
  • the feed material may comprise one or more chemical compositions and/or materials used to make cement and/or cement clinker, (560).
  • the kiln further comprises one or more sensors for monitoring process parameters such as for example temperature of the flame, temperature of the feed material, kiln drive current, kiln feed and fuel to the calciner, (550).
  • the sensor output is transmitted to a database, (551), in a control system, (552).
  • the sensor output may further comprise measurements on the lime content of the cement clinker.
  • the kiln comprises a video-camera, (570), for recording images of the flame, (530).
  • the video-signal from the video- camera is output to a framegrabber and/or a digital video-recorder, (580).
  • the framegrabber and/or digital video-recorder may be used to digitize the video-signal from the video-camera.
  • the framegrabber and/or digital video- recorder may be part of e.g. a computer, (581), or other device comprising a processor for processing data such as for example the device presented in figure 4.
  • the framegrabber and/or digital video-recorder may be a stand alone piece of equipments connected to e.g. a computer via Bluetooth, IR, data and/or optical connections.
  • the camera, (570) may be a digital camera and/or a digital video-camera, connected directly to the computer.
  • Images of the flame may be recorded using the video-camera, digitized by the framegrabber and/or digital video-recorder and/or transmitted to the computer via IR, Bluetooth, data and/or optical connection.
  • the computer may extract features of the image using a method described above, (582). Additionally, the process measurement data may be transmitted from the database, (551), in the control system, (552), to the computer, (583). The data may for example be transmitted from the database to the computer via IR, Bluetooth, an optical cable, a data cable, etc. Based on the process measurement data, (583), and the image features, (582), the computer may determine a PLS and/or DPLS regression model representing the kiln, (584).
  • the regression model may provide soft sensor output to be stored in the control system, (552).
  • the soft sensor output may be transmitted from the computer, (581), to the control system, (552), via for example IR, Bluetooth, an optical cable, an electrical cable, a data cable, etc.
  • the soft sensor output may for example be used to control the processes of the kiln and/or to predict process conditions of the kiln.
  • FIG 6 a flowchart of a method for providing a soft sensor for sensing a cement kiln process condition is presented, (600).
  • the method comprises the steps of: • The method starts in step (610);
  • step (620) process measurements of one or more process parameters, e.g. flame temperature, obtained by one or more measuring devices, e.g. a thermometer, are provided to the control system (552); • In step (630), one or more image data files of e.g. the flame, (530), obtained by said camera, (570), are provided to said computer, (581);
  • process parameters e.g. flame temperature
  • measuring devices e.g. a thermometer
  • step (640) image features from said image data files are extracted by performing MIA on said image data files in said computer; • In step (650), said image features are filtered using e.g. a time-domain filter in said computer;
  • step (660) said image data features are analysed using higher than first order moments, e.g. second to fourth order moments, in said computer;
  • step (670) said one or more process measurements are pre- processed in order to remove/remedy possible contaminated data points;
  • step (680) PLS and/or DPLS regression is performed on said pre- processed process measurements and said analysed image data to develop a soft sensor indicative of one or more process parameters in said cement kiln;
  • step (690) the method ends.
  • Said filtering step (650) and said pre-processing step (670) may be optional.
  • a set of process data and flame images with sufficient variation are selected for modeling with the expectation that these data provide statistically significant dependence between flame frames and process variables as indicators of the amount of the free lime in the clinker and the amount of NOx.
  • One or more score process measurements are logged every minute by the cement kiln control system, (552). The measurements include kiln drive current, kiln feed, fuels to calciner and kiln, plus several temperature measurements. Kiln images are digitalized and stored once every minute considering the storage requirement and CPU load for analysis.
  • a data block of 8000 samples and an equal number of images are selected in this study: 4000 samples are used to develop the model and 4000 samples for validation of the model.
  • the quality of a kiln product is indicated by the amount of free lime in the clinker, (560), which is traditionally measured in a laboratory once every 2 or 4 hours.
  • Two major problems are related to the offline measurement: Firstly, the sampling and analysis result is obtained in a time delay of about an hour. Secondly, it is difficult to take a representative spot sample of particular material with a wide range of particle sizes. To take a representative sample of a particulate material with particles as large as 100 mm, a minimum sample of 1000kg is required, which in practice is difficult. Instead, a smaller clinker-sample is taken. It is implicitly assumed that the clinker is homogeneous and that the small clinker-sample has the same composition (including free lime) as the rest of the clinker in the cement kiln. This assumption does not always hold.
  • quality may indirectly be controlled by a secondary variable, the burning zone temperature. Since the secondary variable might not always be a reliable indication of product quality, real-time estimation with a free lime soft sensor is desirable for effective quality control.
  • the PRESS variable of a CaO/lime soft sensor with process measurements using a DPLS model with orders varying from 0 to 8 is shown in (700) of Figure 7 a).
  • DPLS model of order 4 achieves a minimum PRESS of 0.456. Including further time lagged terms may introduce additional noise into the model possibly leading to an increased PRESS value.
  • (800) of Figure 7 b) validation of the CaO soft sensor (solid line) on lab analysis (star) reveals differences, especially in the first 1500 samples.
  • a second CaO soft sensor is developed by combining image features with conventional process measurements.
  • a moving average filtering is performed on image features.
  • the following filtering window lengths are investigated: 10, 30, 60 and 90 images.
  • a CaO soft sensor with a 4 th order DPLS model derived from data filtered with a window length of 90 images reaches the minimum PRESS value of 0.214, which is over 50% less than the estimation based on filtered conventional process measurements.
  • Comparison between the developed CaO soft sensor and lab measurement during the modelling and validation periods is shown in (900) and (1000) of Figure 8 a) and b), respectively.
  • Reference (1000) of figure 8 b) reveals improved predictive ability compared to reference (800) of Figure 7 b).
  • image features provide complementary information to that of traditional/conventional process measurements thereby enhancing the performance of the CaO soft sensor.
  • the prediction of the 4 th order DPLS soft sensor may yield a small noise component, which may be introduced by time lagged process measurements. Since no pre-filtering of the process measurements has occurred before combination with filtered image features, a low pass filtering of process measurements may lead to a smoother estimation.
  • NOx emissions from a combustion process is of environmental relevance due to its contribution to ambient ozone formation. Cement kilns represent a source of NOx emissions.
  • Traditional continuous emission monitoring uses analytical sensors, which are expensive and costly to maintain, McA voy, T., "Intelligent "Control” Applications in the Process Industries", Annual Reviews in Control 26(1): 75-86, 2002. Regular maintenance is necessary for online gas analyzers which results in an extended period without NOx measurement.
  • Development of a NOx soft sensor is investigated on the abovementioned modelling data set. 1800 samples are used for validation. Using traditional/conventional process measurements as inputs, the minimum PRESS (6.03*10 7 ) is obtained with a NOx soft sensor of 60th order DPLS model using process measurements.
  • (1100) in Figure 9 compares the soft sensor estimation with NOx gas analyzer data, which indicates that the soft sensor follows the trend well.
  • the minimum PRESS value can be reduced to 4.18*10 7 by the NOx soft sensor with a 30 order DPLS model.
  • the NOx soft sensor derived by statistically fusing traditional process measurements and image features provides an improvement over (1100) of Figure 9.
  • the soft sensor derived through statistical data fusion of process measurements and images features estimate NOx with satisfactory accuracy to be used for process control and/or prediction. Since the high order of the soft sensor may increase the load of an online application, pre-filtering of regressors may be usefull.
  • both case studies reveal that image features enhance the performance of soft sensors for cement kiln system, which indicate that image features may capture variations in the burning zone. Since the image system may be a direct observation of the kiln flame, relevant information may be obtained there from. For example, both the NOx formation and free lime content in clinker may depend on the flame temperature, which may be captured through the intensity feature of an image. Other image features relating to the size and position of the flame may also provide useful information.

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Automation & Control Theory (AREA)
  • Radiation Pyrometers (AREA)
  • Curing Cements, Concrete, And Artificial Stone (AREA)

Abstract

L'invention porte sur un procédé (et un capteur logiciel correspondant) pour déterminer un ou plusieurs paramètres de traitement de four à ciment, le procédé comprenant les étapes consistant à recueillir au moins une mesure de traitement représentant au moins un paramètre de traitement et recueillir au moins une donnée d'image représentant une zone de brûlage. Le procédé comprend en outre l'étape consistant à réaliser une régression partielle par la méthode des moindres carrés sur ladite ou lesdites mesures de traitement recueillies et ladite ou lesdites données d'image recueillies pour calculer une sortie de capteur logiciel. Par la présente invention, on obtient, entre autres choses, la détermination de paramètres de traitement de four à ciment difficilement accessibles.
PCT/DK2007/000180 2006-11-02 2007-04-12 Système et procédé pour prédire l'émission de nox et/ou la concentration en chaux libre dans un four à ciment WO2008052542A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US85603006P 2006-11-02 2006-11-02
US60/856,030 2006-11-02

Publications (1)

Publication Number Publication Date
WO2008052542A1 true WO2008052542A1 (fr) 2008-05-08

Family

ID=38042996

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/DK2007/000180 WO2008052542A1 (fr) 2006-11-02 2007-04-12 Système et procédé pour prédire l'émission de nox et/ou la concentration en chaux libre dans un four à ciment

Country Status (1)

Country Link
WO (1) WO2008052542A1 (fr)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103336107A (zh) * 2013-05-30 2013-10-02 中国科学院沈阳自动化研究所 一种水泥熟料f-CaO含量软测量方法
CN106528992A (zh) * 2016-10-27 2017-03-22 贵州理工学院 一种铝用碳素阳极煅烧窑煅烧带软测量判定方法
WO2020188412A1 (fr) 2019-03-15 2020-09-24 Gpcp Ip Holdings Llc Commande en boucle fermée avec détection par caméra de la taille de morceaux de particules de chaux pour améliorer la sonnerie du four à chaux et améliorer le temps de fonctionnement et l'efficacité de fonctionnement
CN112699600A (zh) * 2020-12-23 2021-04-23 中国大唐集团科学技术研究院有限公司火力发电技术研究院 火电运行参数与NOx排放浓度间的偏回分析方法
CN113406313A (zh) * 2021-06-28 2021-09-17 浙江邦业科技股份有限公司 基于全自动游离氧化钙分析仪数据实时预测熟料f-CaO的方法
WO2022099532A1 (fr) * 2020-11-12 2022-05-19 天津水泥工业设计研究院有限公司 Système de four à ciment et procédé de préparation de scorie de ciment
CN112699600B (zh) * 2020-12-23 2024-06-07 中国大唐集团科学技术研究院有限公司火力发电技术研究院 火电运行参数与NOx排放浓度间的偏回分析方法

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH07196347A (ja) * 1993-12-28 1995-08-01 Chichibu Onoda Cement Corp セメント焼成キルンにおける焼成製品の品質管理方法及びその装置
EP1014240A1 (fr) * 1998-12-17 2000-06-28 Siemens Aktiengesellschaft Système, procédé et appareil de déduction fondé sur des cas pour la prédiction d'un capteur dans une installation technique, en particulier dans un four à ciment

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH07196347A (ja) * 1993-12-28 1995-08-01 Chichibu Onoda Cement Corp セメント焼成キルンにおける焼成製品の品質管理方法及びその装置
EP1014240A1 (fr) * 1998-12-17 2000-06-28 Siemens Aktiengesellschaft Système, procédé et appareil de déduction fondé sur des cas pour la prédiction d'un capteur dans une installation technique, en particulier dans un four à ciment

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103336107A (zh) * 2013-05-30 2013-10-02 中国科学院沈阳自动化研究所 一种水泥熟料f-CaO含量软测量方法
CN106528992A (zh) * 2016-10-27 2017-03-22 贵州理工学院 一种铝用碳素阳极煅烧窑煅烧带软测量判定方法
CN106528992B (zh) * 2016-10-27 2019-06-28 贵州理工学院 一种铝用碳素阳极煅烧窑煅烧带软测量判定方法
WO2020188412A1 (fr) 2019-03-15 2020-09-24 Gpcp Ip Holdings Llc Commande en boucle fermée avec détection par caméra de la taille de morceaux de particules de chaux pour améliorer la sonnerie du four à chaux et améliorer le temps de fonctionnement et l'efficacité de fonctionnement
CN113167530A (zh) * 2019-03-15 2021-07-23 Gpcp知识产权控股有限责任公司 用摄像机检测石灰粒子的砾石尺寸来控制闭环以改善石灰窑结圈并改善正常运行时间和操作效率
EP3938724A4 (fr) * 2019-03-15 2022-09-14 GPCP IP Holdings LLC Commande en boucle fermée avec détection par caméra de la taille de morceaux de particules de chaux pour améliorer la sonnerie du four à chaux et améliorer le temps de fonctionnement et l'efficacité de fonctionnement
US11718559B2 (en) 2019-03-15 2023-08-08 Gpcp Ip Holdings Llc Closed loop control with camera detection of pebble size of lime particles to ameliorate lime kiln ringing and improve uptime and operating efficiency
CN113167530B (zh) * 2019-03-15 2023-10-20 Gpcp知识产权控股有限责任公司 用于再循环石灰泥浆的系统和碳酸钙浆液加工方法
WO2022099532A1 (fr) * 2020-11-12 2022-05-19 天津水泥工业设计研究院有限公司 Système de four à ciment et procédé de préparation de scorie de ciment
CN112699600A (zh) * 2020-12-23 2021-04-23 中国大唐集团科学技术研究院有限公司火力发电技术研究院 火电运行参数与NOx排放浓度间的偏回分析方法
CN112699600B (zh) * 2020-12-23 2024-06-07 中国大唐集团科学技术研究院有限公司火力发电技术研究院 火电运行参数与NOx排放浓度间的偏回分析方法
CN113406313A (zh) * 2021-06-28 2021-09-17 浙江邦业科技股份有限公司 基于全自动游离氧化钙分析仪数据实时预测熟料f-CaO的方法

Similar Documents

Publication Publication Date Title
JP4645422B2 (ja) 判定装置、判定装置の制御プログラム、および判定装置の制御プログラムを記録した記録媒体
WO2008052542A1 (fr) Système et procédé pour prédire l'émission de nox et/ou la concentration en chaux libre dans un four à ciment
Cécillon et al. Assessment and monitoring of soil quality using near‐infrared reflectance spectroscopy (NIRS)
US20100211333A1 (en) Leak Detection and Identification System
Steffen et al. A comparison of data quality control protocols for atmospheric mercury speciation measurements
Pimstein et al. Performance of three identical spectrometers in retrieving soil reflectance under laboratory conditions
US10393711B2 (en) System and method for identifying and distinguishing materials, method for identifying or distinguishing materials, and measuring device for recording material properties of materials
US10656134B2 (en) Isothermal analysis system and method to determine at least one type of carbon compounds in drill cuttings
Richter et al. Free iron oxide determination in Mediterranean soils using diffuse reflectance spectroscopy
US6792357B2 (en) Optical corrosion measurement system
CN105956591B (zh) 一种在线高温零件红外图像光谱抽样检测方法
US20130311136A1 (en) Rule-Based Sample Verification and Chemical Monitoring Methodology
CN114730184A (zh) 用于估算逃逸性排放上的变化的计算机系统及方法
CN111504927A (zh) 用于食品检测的高效稳定的原子吸收光谱仪及其控制方法
Taguem et al. Quantile regression with a metal oxide sensors array for methane prediction over a municipal solid waste treatment plant
JPH07318508A (ja) 画像判定装置および画像判定方法
Bootsmaa et al. Online air monitoring with enoses at the tata steel plant in the Netherlands
CN116612836B (zh) 三氟甲烷生产的尾气量预测方法和系统
JP6879276B2 (ja) 気中分散微粒子の発生判定方法及び装置並びに塊状物質の性状測定方法及び装置
RU2582296C1 (ru) Оптико-электронное устройство для контроля качества моторного масла
US20160290919A1 (en) Spectral analysis through model switching
CN108519383B (zh) 一种全波段生物显色实验的检测方法
JP2010164446A (ja) 検査対象品の外観検査方法及びプログラム
Zamzami et al. Response of Semiconductor Sensor in An Electronic Nose with Multi-Sample Delivery System
Magare et al. Milk Powder Analysis for Protein Content Estimation Using Fieldspec4 Spectroradiometer.

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 07722559

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 07722559

Country of ref document: EP

Kind code of ref document: A1