CN101698896A - System and method for steel-making online end-point control through furnace mouth radiation information fusion - Google Patents
System and method for steel-making online end-point control through furnace mouth radiation information fusion Download PDFInfo
- Publication number
- CN101698896A CN101698896A CN 200910035884 CN200910035884A CN101698896A CN 101698896 A CN101698896 A CN 101698896A CN 200910035884 CN200910035884 CN 200910035884 CN 200910035884 A CN200910035884 A CN 200910035884A CN 101698896 A CN101698896 A CN 101698896A
- Authority
- CN
- China
- Prior art keywords
- sub
- information
- furnace mouth
- video
- model
- Prior art date
- Legal status (The legal status 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 status listed.)
- Granted
Links
Images
Landscapes
- Radiation Pyrometers (AREA)
Abstract
The invention discloses a system and a method for steel-making online end-point control through furnace mouth radiation information fusion. The system comprises a furnace mouth radiation acquisition subsystem, an optical fiber spectrum division multiplexing subsystem, a multi-spectral combined detection subsystem, a furnace mouth flame video acquisition subsystem and a comprehensive determination subsystem, wherein the furnace mouth radiation acquisition subsystem is connected with the optical fiber spectrum division multiplexing subsystem; the optical fiber spectrum division multiplexing subsystem is connected with the multi-spectral combined detection subsystem; the multi-spectral combined detection subsystem is connected with the comprehensive determination subsystem through a serial port; and the furnace mouth flame video acquisition subsystem comprises a CCD camera provided with an optical lens and a video capture board, the CCD camera is arranged on the furnace mouth radiation acquisition subsystem and is connected with the video capture board, and the video capture board is connected with the comprehensive determination subsystem. The system and the method achieve a long-range detection (20 meters), so the system can work normally under a severe environment of steel-making without using a fussy cooling device and a dust collector.
Description
Technical field
The invention belongs to the terminal point control techniques that is used for converter (Basic Oxygen Furnace, be called for short BOF) steel-making converting process, particularly a kind of furnace mouth radiation information merges and is used to make steel the system and the method thereof of online terminal point control.
Background technology
The traditional method of judging about converter terminal mainly contains the method for STEADYSTATE CONTROL MODEL and dynamic control model now.Static model carry out quantitative Analysis according to material balance principle and heat balance principle to the starting condition of converting process, still, owing to do not consider the information of converting process, can not carry out on-line tracing and revise in real time.If dynamic-control system sublance equipment, sublance can by sublance records in the converting process of converter result with after the blowing target of being scheduled to is compared, revise the blowing operation according to both deviations, but, be subjected to the size and the high restriction of sublance cost of investment of sublance, now only use, also can't be generalized to middle-size and small-size converter shed at large-scale converter steel factory.
Thereby at present, the middle-size and small-size converter steel factory that accounts for main body in the domestic steel-making still mainly uses based on the single static model, is the terminal point determining method of foundation with demonstration of operator's console instrument and artificial experience.In actually operating, the thermometric of falling the stove is decided to carry out next step operation again behind the carbon, but generally speaking, be subjected to the influence of the complexity of converting process and workman's the mental status, the hit rate of a catch carbon is often lower, the operation that usually needs to re-blow, and this just reduces productivity, influence steel quality, strengthened working strength of workers.
Because traditional method or inaccurate to endpoint, perhaps the high suitability of cost is wideless, has engendered some to the novel end-point control method that converter terminal is judged, wherein a kind of optical means is authorized in the Chinese patent 96197034 of husky profit A Luoke.
This patent has designed a kind of optical sensor metering facility with cooling system and dust-collecting air brush, this equipment is gathered the light intensity of the 560nm wavelength period of converter mouth visible light under the closer distance of 30 feet of distance 290 ton large-scale converters, calculate by giving special-purpose programmable logic controller the blowing oxygen quantity that records the maximum light intensity difference of the process of bessemerizing and synchronization as the model input variable, what obtain molten steel this moment is blowing carbon content.This patent is thought that this technology is applicable to and is contained 0.06% or the prediction of the carbon content of the carbon content of soft steel still less and a catch carbon.
But the related device of this patent itself has caused the raising of cost, each sampling time in early stage is 60s, and each sampling time in later stage is 4s, and this has all brought certain influence to the continuity of data sample, and can lose valid data, especially at the crucial later stage of blowing.
In this patent, method therefor only be applicable at present oxygen supply stable, the above mammoth conveter of 200t, carbon content be lower than the control of the molten steel endpoint carbon content in 0.05% scope, and in the device loaded down with trivial details refrigerating unit and cleaning apparatus are arranged, when judging the medium and high carbon steel converter, use this patent that bigger error is arranged.
Summary of the invention
The object of the present invention is to provide a kind of be used for the converter steelmaking scene accurately, low-cost, noncontact is remote, easy to operate, the system of service-strong BOF Steelmaking Endpoint on-line measurement control and be used for the model and the method for converter terminal prediction, is difficult to the accurate problem of predicting thereby can solve current converter steelmaking blow end point.
The technical solution that realizes the object of the invention is: a kind of furnace mouth radiation information merges and is used to make steel the system of online terminal point control, comprises that furnace mouth radiation obtains sub-system, optical fiber spectrum division multiplexing sub-system, multispectral complex probe sub-system, fire door flame video and obtains sub-system and synthetic determination sub-system; Furnace mouth radiation obtains sub-system and connects optical fiber spectrum division multiplexing sub-system, and this optical fiber spectrum division multiplexing sub-system connects multispectral complex probe sub-system, and this multispectral complex probe sub-system is connected with the synthetic determination sub-system by serial ports; Fire door flame video obtains sub-system and comprises ccd video camera and the video frequency collection card that is being equipped with optical lens, and ccd video camera is erected at furnace mouth radiation and obtains on the sub-system, and this pick up camera connects video frequency collection card, this video acquisition card connection synthetic determination sub-system;
Furnace mouth radiation obtains sub-system the spectral energy information of converter mouth flame radiaton is collected, and be imaged onto on the focal plane, the focal plane coupling that one end of optical fiber spectrum division multiplexing sub-system and furnace mouth radiation obtain sub-system, the other end is divided into four the road or four the tunnel to the spectral energy information that collects with upper channel, and be transferred to respectively be equipped with the different wave length interference filter with the corresponding photodetector of passage in go, these photodetectors are positioned at the receiving end of multispectral complex probe sub-system, by this multispectral complex probe sub-system optical signal is transformed into electrical signal and sends the serial ports receiving software of synthetic determination sub-system to by serial ports, carry out the reception of the spectrum intensity signal of furnace mouth radiation, storage and mapping operation; Fire door flame video is obtained the ccd video camera adjustment of sub-system and collection vision that furnace mouth radiation obtains sub-system is coincide, after this ccd video camera collects the RGB color model information of converter mouth flame video image, be transferred to video frequency collection card by video inputs, this video frequency collection card utilizes the image acquisition and processing software of synthetic determination sub-system [5], this RGB color model information by the storage of the chrominance information in the hsv color model after the color model conversion, this chrominance information that obtains is carried out mapping operation as the characteristic information data of fire door flame image according to these data; Finally, the BP neural network terminal point predictive model utilization of synthetic determination sub-system obtains the online terminal point predictive control that spectrum intensity signal and these two portions data of fire door flame image information are carried out converter steelmaking.
A kind of furnace mouth radiation information merges and is used to make steel the method for online terminal point control, and step is as follows:
The first step, obtain the spectrum intensity signal, the spectral energy information of converter mouth flame radiaton is collected, by optical fiber spectrum division multiplexing sub-system the spectral energy information that collects is divided into four the road or four the tunnel with upper channel, by multispectral complex probe sub-system optical signal is transformed into electrical signal and passes through the reception of serial ports receiving software, obtain the spectrum light intensity information data at furnace mouth radiation place, carry out mapping operation according to these data;
Second step, obtain fire door flame image information, the RGB color model information of the converter mouth flame video image of gathering, be transferred to video frequency collection card by video inputs, this video frequency collection card utilizes image acquisition and processing software, by the storage of the chrominance information in the hsv color model after the color model conversion, this chrominance information that obtains is carried out mapping operation as the characteristic information data of fire door flame image according to these data this RGB color model information;
The 3rd step, spectrum intensity signal and fire door flame image information are carried out the benchmark adjustment, the magnification of less relatively spectrum light intensity value is adjusted, promptly the corresponding pairing spectrum light intensity value constantly of blowing image maximum in latter stage is adjusted into onesize image value, with the adjusted basis of this magnification that obtains as global variable; Power central slide filtering methods such as utilization simultaneously carry out filtering to above-mentioned data and smoothly reach denoising processing, spectrum light intensity and characteristics of image change curve in finally obtaining bessemerizing;
The 4th step, the BP neural network model that makes up carries out quantitative judgement constantly to terminal point, at first select input variable in the BP neural network prediction model, suppose that selected whole variable numbers are m, the neuron number of hidden layer is h, and z is network output, and then hr is the input of each node of hidden layer, Hr is the output of each node of hidden layer, definition G
Rs=Cov (h
r, x
s) w
Rs/ Var (h
r) for importing the influence value of node s to latent node r, g
r=Cov (H
r, z) w
r/ Var (H
r) for concealing the influence value of node r to output, Q
s=∑ G
Rsg
rBe the influence value of input variable to overall output; If the sum that influences of n variable in the overall variable accounts for more than 90% of general impacts value, then choosing this n variable is the input variable of network model;
Secondly, finally the input and output parameter of selected model is:
(1) work output of BP network model: consider that whole model is the prediction that realizes the steel-making terminal time, so be chosen as the output of converter steelmaking blow end point time T (unit: second) as model;
(2) input of BP network model: because with of the output of converter steelmaking blow end point time T as model, the value of considering the whole process middle and later periods is relatively stable, so after the comparative analysis, spectrum light intensity average, image average, the ratio between the two of selected 400-500 time period second, the blowing pairing time of image value maximum point in latter stage, the image value of this maximum point correspondence, spectrum light intensity value and between the two above seven variablees of ratio as the input parameter;
At last, set up BP neural network terminal point predictive model based on the input/output variable of spectrum light intensity and characteristics of image change curve and model, utilize a large amount of converter steelmaking blowing data of collection in worksite, after the training of BP neural network terminal point predictive model, the terminal point of scene blowing heat is carried out the real-time estimate of network model, utilize the terminal point predictor that this model obtains and the on-the-spot comparing result of terminal point actual value, realize the terminal point predictive control of converter.
The present invention compared with prior art, its remarkable advantage:
(1) by using telescopic system to combine, realized long-range detection (20m) with optical fiber spectrum division multiplexing system, so just realized need not be loaded down with trivial details refrigerating unit and cleaning apparatus just can be under the severe environment of making steel works better;
(2) proposed a kind of optical fiber spectrum division multiplexing technology, can extract the multispectral pattern of converter mouth optical radiation comprehensively, thereby more constitutionally obtains the inner link of radiation information, also improve measuring accuracy and reliability simultaneously.
(3) proposed extraction with fire door flame video image characteristic information and be placed on method in the information analysis relevant, and, analyzed the converter optical radiation blowing change curve that has obtained blowing reaction process in the reflection stove in conjunction with the spectrum intensity signal with visual color.Test data in this optical radiation blowing change curve is obtained easy, feasible, accurate, real-time.
(4) the BP neural network terminal point predictive model of Gou Jianing, the result shows the time of response less than 2s, the precision of prediction in 8s surpasses 85%, satisfies the requirement of online quick judgement of converter terminal and precision of prediction.
(5) said system and method, its characteristic is that measurement control method online in real time, device are simple and with low cost, to improving the quality of products, reducing cost, promote the tooling modernization that obvious effects will be arranged, can provide reference frame to the control of the converter terminal in the present stage actual production process, have broad application prospects.
Below in conjunction with accompanying drawing the present invention is described in further detail.
Description of drawings
Fig. 1 is that furnace mouth radiation information of the present invention merges and is used to make steel the one-piece construction figure of online terminal point Controlling System.
Fig. 2 is the structure iron of multispectral radiation detection sub-system.
Fig. 3 is the three-dimensional plot of chromatic(ity)component.
Fig. 4 is the change curve of spectrum light intensity and image feature value.
Fig. 5 is the BP neural network terminal point predictive model structure iron that makes up.
Fig. 6 is terminal point actual value and the BP neural network comparison diagram to the terminal point predictor.
Embodiment
The present invention has made up a kind of furnace mouth radiation information theoretical based on spectral radiance and that experience steel-making is theoretical and has merged the system that is used for the online terminal point control of converter (BOF) steel-making, comprise that furnace mouth radiation obtains sub-system 1, optical fiber spectrum division multiplexing sub-system 3, multispectral complex probe sub-system 4, fire door flame video and obtains sub-system 2, synthetic determination sub-system 5, wherein synthetic determination sub-system 5 comprises the serial ports receiving software, image acquisition and processing software, BP neural network terminal point predictive model three parts.The one-piece construction figure of this system as shown in Figure 1.
Furnace mouth radiation obtains sub-system 1 and connects optical fiber spectrum division multiplexing sub-system 3, and this optical fiber spectrum division multiplexing sub-system 3 connects multispectral complex probe sub-system 4, and this multispectral complex probe sub-system 4 is connected with synthetic determination sub-system 5 by serial ports; Fire door flame video obtains sub-system 2 and comprises ccd video camera and the video frequency collection card that is being equipped with optical lens, and ccd video camera is erected at furnace mouth radiation and obtains on the sub-system 1, and this pick up camera connects video frequency collection card, this video acquisition card connection synthetic determination sub-system 5.
Workflow is as follows: furnace mouth radiation obtains sub-system 1 spectral energy information of converter mouth flame radiaton is collected, and be imaged onto on the focal plane, one end of optical fiber spectrum division multiplexing sub-system 3 and furnace mouth radiation obtain the focal plane coupling of sub-system 1, the other end is divided into four the road or four the tunnel to the spectral energy information that collects with upper channel, and be transferred to respectively be equipped with the different wave length interference filter with the corresponding photodetector of passage in go, these photodetectors are positioned at the receiving end of multispectral complex probe sub-system 4, by this multispectral complex probe sub-system 4 optical signal is transformed into electrical signal and sends the serial ports receiving software of synthetic determination sub-system 5 to by serial ports, carry out the reception of the spectrum intensity signal of furnace mouth radiation, storage and mapping operation; Fire door flame video is obtained the ccd video camera adjustment of sub-system 2 and collection vision that furnace mouth radiation obtains sub-system 1 is coincide, after this ccd video camera collects the RGB color model information of converter mouth flame video image, be transferred to video frequency collection card by video inputs, this video frequency collection card utilizes the image acquisition and processing software of synthetic determination sub-system 5, this RGB color model information by the storage of the chrominance information in the hsv color model after the color model conversion, this chrominance information that obtains is carried out mapping operation as the characteristic information data of fire door flame image according to these data; Finally, the BP neural network terminal point predictive model utilization of synthetic determination sub-system 5 obtains the online terminal point predictive control that spectrum intensity signal and these two portions data of fire door flame image information are carried out converter steelmaking.
Consider that from security standpoint framework plays system in the operation room of the on-the-spot about 20m of steel-making.The operation steps of system's key instrument is: set up trivet; Furnace mouth radiation is obtained sub-system and ccd video camera is installed on the trivet; Obtain the focal plane of sub-system at furnace mouth radiation focalizer is installed; Regulate orientation and pitching that furnace mouth radiation obtains sub-system, flame kernel is overlapped with the center of focalizer; Regulate focusing handwheel, make imaging combustion clear; The gun sight or the ccd video camera of focalizer are aimed at flame, flame kernel is overlapped with the gun sight of focalizer or the center, visual field of ccd video camera; Take out focusing lens, load onto the jumbo fiber bundle of optical fiber spectrum division multiplexing sub-system; If in use trivet moves, the gun sight or the ccd video camera that need only with focalizer are that benchmark is regulated the instrument host position, needn't use focalizer again; Aforesaid operations is a benchmark adjustment instrument host with the gun sight of focalizer preferably.After adjusting furnace mouth radiation and obtaining sub-system, can use that the image in the video window carries out fine tuning to ccd video camera on the nut of support of ccd video camera and the observer computer.
1.1 furnace mouth radiation obtains the severe environment of sub-system 1 at the converter steelmaking scene, in order to incite somebody to action fire door imaging combustion at a distance to the focal plane, selected the heavy caliber camera lens in two not cemented objectives for use, obtained the structural parameter table (seeing Table 1) that furnace mouth radiation obtains sub-system by calculation of parameter, this furnace mouth radiation obtains sub-system and uses trivet to fix, support, height and commentaries on classics degree by adjusting the tripod pallet obtain to gather preferably the vision orientation.This furnace mouth radiation obtains sub-system and comprises by the designed heavy caliber camera lens of table 1, be positioned at furnace mouth radiation obtain the sub-system focal plane outside, can before and after stretch and be used to regulate the focusing handwheel of imaging combustion sharpness, and external, can be installed in furnace mouth radiation and obtain the focalizer that being used on the sub-system focal plane aim at flame kernel.This furnace mouth radiation obtains sub-system in the face of converter mouth, to gather converter mouth radiating spectral energy information; Obtain focusing handwheel on the sub-system by regulating this furnace mouth radiation, spectral energy information can clearly be imaged onto on the focal plane that this furnace mouth radiation obtains sub-system.The design this furnace mouth radiation obtain sub-system have bore unrestricted, can utilize advantages such as airspace correct residual sum of errors relative aperture can increase, can reach the visual effect of artificial low coverage, high temperature and pollution infringement can be kept away from again, the collecting work of remote (about 20m) condition can be satisfied detection system.
Table 1 furnace mouth radiation obtains the structural parameter (mm) of sub-system
1.2 it is many part that optical fiber spectrum division multiplexing sub-system 3 can be divided into after receiving spectral energy equably, and be transferred to respectively in the ten tunnel different photodetectors that are being equipped with different spectral filters and go.The present invention selects the ladder index fiber of jumbo fiber core diameter for use, the correlation theory and the technology of utilization optical fiber " wavelength-division multiplex ", design optical fiber spectrum sub-system realizes " optical fiber spectrum division multiplexing ", adopt the corresponding biography light of subchannel fibre bundles such as bundle number and spectral energy information, these performances that pass the light fibre bundle are identical, diameter is the same, from an end of every bundle, take out an optical fiber, with the synthetic unit spot of the optical fiber that takes out, then all unit spot random alignment are formed a jumbo fiber bundle, behind this jumbo fiber bundle interface gummed, cut out smooth even curface, form the entrance pupil face of optical fiber spectrum division multiplexing sub-system 3; All pass the other end of light fibre bundle and thereby the interference filter of different wave length is close to the effect that has realized that spectrum is divided.
1.3 multispectral complex probe sub-system 4 receives the spectral energy information that optical fiber spectrum division multiplexing sub-system transmits, selecting Fa Buli for use---uncle Luo Gan relates to spectral filter light is filtered into arrowband monochromatic ray information, this the multispectral complex probe sub-system conversion of monochromatic ray information and warp is output as the numerary signal of reciprocity passage then, sends to computer by serial ports.The interference filter wavelength parameter of design is shown in Table 2.
Table 2 spectral filter Specifeca tion speeification
This multispectral complex probe sub-system 4 will satisfy quick response, handle large information capacity and requirement such as work in real time under severe environment such as high temperature, flue dust, main being responsible for becomes ten road spectral energy signals that the transmission of optical fiber spectrum division multiplexing sub-system comes into ten way word signals, and send to the task of computer, thereby selected for use being that the Circuits System of core is formed (but the suitable device of the similar data of any energy measurement all can use) by photodetector with the micro-chip in native system.
The schematic circuit of this multispectral complex probe sub-system 4 as shown in Figure 2, its chief component is photodetector, filtering circuit, amplifier, sampling hold circuit, A/D change-over circuit, micro-chip, clock circuit, communicating circuit etc.Send the furnace mouth radiation spectral energy of ten road five equilibriums of multispectral complex probe sub-system to through optical fiber spectrum division multiplexing sub-system, conversion output ten way word signals in this multispectral complex probe sub-system, this ten ways word signal is represented the radiation light intensity of different wave length respectively, according to there are differences between different wave length flashlight yield of radiation, through needing after the filtering by the different amplifying circuit of gain, reach and enter the A/D change-over circuit behind the requirement voltage amplitude and change, micro-chip collection conversion back data send to computer by serial ports after treatment, and utilize the serial ports receiving software that the spectrum intensity signal is received, storage and mapping operation.
This serial ports receiving software utilizes the MSComm control of VC software platform, the frequency parameter of having set baud rate, data bit, stop bit, parity checking and the received signal of serial ports transmission is 10ms/ time, multispectral complex probe sub-system 4 every transmission one secondary data, this serial ports receiving software just receives and storage work, again the light intensity data of the 546nm wavelength respective channel that obtains is carried out mapping operation.
Here descend explanation, the light intensity that collects is a relative light intensity, as long as light intensity data unit of measure self-consistentency, rule that just can the reaction converter converting process, unit choose to this variation tendency without any influence, any unit of measure arbitrarily or fixed, can be used for measuring the spectrum intensity signal of furnace mouth radiation in native system.
1.4 fire door flame video obtains the ccd video camera that is being equipped with optical lens of sub-system 2 and is erected at furnace mouth radiation and obtains on the sub-system, and the collection vision that adjustment and furnace mouth radiation obtain sub-system is coincide.
After this ccd video camera collects the RGB color model information of converter mouth flame video image, be transferred to video frequency collection card by video inputs, this video frequency collection card utilizes the image acquisition and processing software of synthetic determination sub-system, this RGB color model information by the storage of the chrominance information in the hsv color model after the color model conversion, this chrominance information that obtains is carried out mapping operation as the characteristic information data of fire door flame image according to these data.This image acquisition and processing software workflow of Processing tasks while gathering of realizing fire door flame video image is:
(1) creates the window of catching of fire door flame image information: call the capCreateCaptureWindow function, generate video capture window.
(2) gather fire door flame image video information to buffer zone and carry out analyzing and processing, it is the core content of system's real time implementation, need utilize callback mechanism, three call back functions that need Accreditation System to use: mistake call back function capSetCallbackOnError, state call back function capSetCallbackOnStatus and video flowing call back function capSetCallbackOnVideoStream, setting by the video flowing call back function and the structure processing mode of tabling look-up, make the RGB color model information of the fire door flame image that collects, through obtaining the chrominance information in the corresponding hsv color model after the color model conversion, promptly earlier the pairing data sheet of the change over condition of two kinds of color model is built, in call back function, add the software code that the RGB color model is converted to the hsv color model; Then each two field picture of gathering is according to pixels extracted, and search position in the pairing HSV table of this rgb value; Then the essential numerical value to all pixels of this two field picture superposes, and obtains and draw the three-dimensional plot of chromatic(ity)component, as shown in Figure 3; Last finding in this three-dimensional plot in the converting process when the reaching home of certain colourimetric number correspondence changes tangible curve, and draws.What time following note simultaneously: get rid of the more violent variation that occurs once in a while, this is to disturb the possibility that causes bigger; Select the number of pixels value to change metastable that colourity rank; Should not select background more colourity rank to occur; Destination county is bessemerized in corresponding closing under this colourity rank has than obvious variation, changes obviously relatively being beneficial to endpoint.What adopt here is that look-up method can reach the purpose of handling while gathering for speed up processing.
(3) call function capCaptureGetSetup is provided with the correlation parameter of acquisition window, and the default per second is gathered 15 frames.
(4) window and video capture device are caught in call function capDriveConnect connection, and use the capDriverGetCaps function to return the function of sampler.
(5) for the better differentiation situation of observation flame, system call function capPreview, reach good preview effect with frame number faster.
(6) do not carry out the preservation of video flowing in the above step, if some other requirement is arranged, can preserve by call function capFileSetCaptureFile, but the video file that does not have compression is very big, can use video compress manager (Video Compression Mananger) to compress the back and preserve.
(7) call function capCaptureAbort finishes video.
This important evidence that different integral color variations is arranged with reference to the different blowing fire door flame constantly of experience steel-making, system becomes the hsv color model data to the image transitions of the rgb format that collects, obtained chromatic(ity)component three-dimensional plot as shown in Figure 3, X-axis is the colourity rank among the figure, Y-axis is a single-frame images number, and the Z axle is corresponding number of pixels.Because the histogram of single-frame images can not reflect the Changing Pattern of entire image video, thereby from the 3-D view that all HSV image histograms are formed, find colourity H comprising more potential information 17 times, when closing on terminal point with compare value before and obvious variation occurred.
1.5 being spectrum light intensity and image feature values that a spectrum intensity signal that collects and fire door flame image information differ bigger, synthetic determination sub-system 5 carries out the benchmark adjustment, the principle of benchmark adjustment is that the magnification of less relatively spectrum light intensity value is adjusted, promptly the corresponding pairing spectrum light intensity value constantly of blowing image maximum in latter stage is adjusted into onesize image value, with the adjusted basis of this magnification that obtains as global variable; Power central slide filtering methods such as utilization simultaneously carry out filtering to above-mentioned data and smoothly reach the denoising processing, and its mathematical expression form is
K=n+1, n+2, Λ, N-n has finally obtained the spectrum light intensity in the bessemerizing as shown in Figure 4 and the change curve of image feature value.
In this change curve, can find, in this polynary heterogenetic high-temperature physics chemical reaction process of converter steelmaking, the spectrum light intensity value progressively becomes big with the converter steelmaking converting process, the blowing mid-term since the reaction violent interference increase the phenomenon that fluctuation is arranged, light intensity curve slowly descends during near blow end point; Meanwhile image feature value slowly rose at the initial stage of blowing, existed violent vibration mid-term, significantly improved during near terminal point latter stage and arrive, and near blow end point the time, image curve has the appearance of a maximum value.These variation characteristic real time reactions the variation of the spectrum picture characteristic quantity in the steelmaking process.But because the difference of factors such as starting material that different heat adds and blowing condition, caused the terminal point of each heat to have bigger fluctuation constantly, thereby this description qualitatively can not be judged terminal point exactly, on this basis, made up BP neural network terminal point predictive model terminal point has been carried out quantitative judgement constantly.
The model construction of the BP neural network prediction model that the present invention makes up as shown in Figure 5.Owing to now the selection of input variable in the BP neural network prediction model also there is not a kind of method of comparison system, all be to choose generally according to planner's experience.The method that the present invention selects to be adopted to BP neural network prediction model input variable is: the selected whole variable numbers of supposition are m, the neuron number of hidden layer is h, and z is network output, and then hr is the input of each node of hidden layer, Hr is the output of each node of hidden layer, definition G
Rs=Cov (h
r, x
s) w
Rs/ Var (h
r) for importing the influence value of node s to latent node r, g
r=Cov (H
r, z) w
r/ Var (H
r) for concealing the influence value of node r to output, Q
s=∑ G
Rsg
rBe the influence value of input variable to overall output; If the sum that influences of n variable in the overall variable accounts for more than 90% of general impacts value, then choosing this n variable is the input variable of network model;
Finally the input and output parameter of selected model is:
(1) work output of BP network model: consider that whole model is the prediction that realizes the steel-making terminal time, so be chosen as the output of converter steelmaking blow end point time T (unit: second) as model.
(2) input of BP network model: because with of the output of converter steelmaking blow end point time T as model, the value of considering the whole process middle and later periods is relatively stable, so after the comparative analysis, spectrum light intensity average, image average, the ratio between the two of selected 400-500 time period second, the blowing pairing time of image value maximum point in latter stage, the image value of this maximum point correspondence, spectrum light intensity value and between the two above seven variablees of ratio as the input parameter.
Set up BP neural network terminal point predictive model based on the input/output variable of spectrum light intensity and characteristics of image change curve and model.Utilize a large amount of blowing data of collection in worksite, after the training of network model, the terminal point of on-the-spot 49 groups of heats is carried out the real-time estimate of network model, the model prediction time of response is 1.2s, and Fig. 6 has provided current experimental result.From the experimental result comparison diagram that utilizes terminal point predictor that this model obtains and terminal point actual value as can be seen, the predicated error that has 42 groups is in 8s, that is to say that this neural network model has reached 85.7% at the precision of prediction of 8s, satisfy the requirement of online quick judgement of converter terminal and precision of prediction.
Claims (6)
1. a furnace mouth radiation information merges and is used to make steel the system of online terminal point control, it is characterized in that comprising that furnace mouth radiation obtains sub-system [1], optical fiber spectrum division multiplexing sub-system [3], multispectral complex probe sub-system [4], fire door flame video and obtains sub-system [2] and synthetic determination sub-system [5]; Furnace mouth radiation obtains sub-system [1] and connects optical fiber spectrum division multiplexing sub-system [3], and this optical fiber spectrum division multiplexing sub-system [3] connects multispectral complex probe sub-system [4], and this multispectral complex probe sub-system [4] is connected with synthetic determination sub-system [5] by serial ports; Fire door flame video obtains sub-system [2] and comprises ccd video camera and the video frequency collection card that is being equipped with optical lens, ccd video camera is erected at furnace mouth radiation and obtains on the sub-system [1], this pick up camera connects video frequency collection card, this video acquisition card connection synthetic determination sub-system [5];
Furnace mouth radiation obtains sub-system [1] spectral energy information of converter mouth flame radiaton is collected, and be imaged onto on the focal plane, one end of optical fiber spectrum division multiplexing sub-system [3] and furnace mouth radiation obtain the focal plane coupling of sub-system [1], the other end is divided into four the road or four the tunnel to the spectral energy information that collects with upper channel, and be transferred to respectively be equipped with the different wave length interference filter with the corresponding photodetector of passage in go, these photodetectors are positioned at the receiving end of multispectral complex probe sub-system [4], by this multispectral complex probe sub-system [4] optical signal is transformed into electrical signal and sends the serial ports receiving software of synthetic determination sub-system [5] to by serial ports, carry out the reception of the spectrum intensity signal of furnace mouth radiation, storage and mapping operation; Fire door flame video is obtained the ccd video camera adjustment of sub-system [2] and collection vision that furnace mouth radiation obtains sub-system [1] is coincide, after this ccd video camera collects the RGB color model information of converter mouth flame video image, be transferred to video frequency collection card by video inputs, this video frequency collection card utilizes the image acquisition and processing software of synthetic determination sub-system [5], this RGB color model information by the storage of the chrominance information in the hsv color model after the color model conversion, this chrominance information that obtains is carried out mapping operation as the characteristic information data of fire door flame image according to these data; Finally, the BP neural network terminal point predictive model utilization of synthetic determination sub-system [5] obtains the online terminal point predictive control that spectrum intensity signal and these two portions data of fire door flame image information are carried out converter steelmaking.
2. furnace mouth radiation information according to claim 1 merges and is used to make steel the system of online terminal point control, it is characterized in that obtaining in the sub-system [1] at furnace mouth radiation, select the heavy caliber camera lens in two not cemented objectives for use and calculate suitable structural parameter table and make up furnace mouth radiation and obtain sub-system, this furnace mouth radiation obtains sub-system in the face of converter mouth, to gather converter mouth radiating spectral energy information, obtain focusing handwheel on the sub-system by regulating this furnace mouth radiation, spectral energy information clearly is imaged onto on the focal plane that this furnace mouth radiation obtains sub-system, simultaneously, this furnace mouth radiation obtains sub-system and uses trivet to fix, support, height and commentaries on classics degree by adjusting the tripod pallet obtain to gather preferably the vision orientation.
3. furnace mouth radiation information according to claim 1 merges and is used to make steel the system of online terminal point control, it is characterized in that in optical fiber spectrum division multiplexing sub-system [3], adopt the corresponding biography light of subchannel fibre bundles such as bundle number and spectral energy information, from an end of every bundle, take out an optical fiber, with the synthetic unit spot of the optical fiber that takes out, then all unit spot random alignment are formed a jumbo fiber bundle, behind this jumbo fiber bundle interface gummed, cut out smooth even curface, form the entrance pupil face of optical fiber spectrum division multiplexing sub-system; All pass the other end of light fibre bundle and the interference filter of different wave length is close to.
4. furnace mouth radiation information according to claim 1 merges and is used to make steel the system of online terminal point control, it is characterized in that in multispectral complex probe sub-system [4], at first the spectral energy information that sends of subchannel such as spectral energy information is through Fa Buli---and uncle Luo Gan relates to spectral filter light is filtered into arrowband monochromatic ray information, this the multispectral complex probe sub-system conversion of monochromatic ray information and warp is output as the numerary signal of reciprocity passage then, sends to computer by serial ports.
5. a furnace mouth radiation information merges and is used to make steel the method for online terminal point control, it is characterized in that:
The first step, obtain the spectrum intensity signal, the spectral energy information of converter mouth flame radiaton is collected, by optical fiber spectrum division multiplexing sub-system [3] spectral energy information that collects is divided into four the road or four the tunnel with upper channel, by multispectral complex probe sub-system [4] optical signal is transformed into electrical signal and passes through the reception of serial ports receiving software, obtain the spectrum light intensity information data at furnace mouth radiation place, carry out mapping operation according to these data;
Second step, obtain fire door flame image information, the RGB color model information of the converter mouth flame video image of gathering, be transferred to video frequency collection card by video inputs, this video frequency collection card utilizes image acquisition and processing software, by the storage of the chrominance information in the hsv color model after the color model conversion, this chrominance information that obtains is carried out mapping operation as the characteristic information data of fire door flame image according to these data this RGB color model information;
The 3rd step, spectrum intensity signal and fire door flame image information are carried out the benchmark adjustment, the magnification of less relatively spectrum light intensity value is adjusted, promptly the corresponding pairing spectrum light intensity value constantly of blowing image maximum in latter stage is adjusted into onesize image value, with the adjusted basis of this magnification that obtains as global variable; Power central slide filtering methods such as utilization simultaneously carry out filtering to above-mentioned data and smoothly reach denoising processing, spectrum light intensity and characteristics of image change curve in finally obtaining bessemerizing;
The 4th step, the BP neural network model that makes up carries out quantitative judgement constantly to terminal point, at first select input variable in the BP neural network prediction model, suppose that selected whole variable numbers are m, the neuron number of hidden layer is h, and z is network output, and then hr is the input of each node of hidden layer, Hr is the output of each node of hidden layer, definition G
Rs=Cov (h
r, x
s) w
Rs/ Var (h
r) for importing the influence value of node s to latent node r, g
r=Cov (H
r, z) w
r/ Var (H
r) for concealing the influence value of node r to output, Q
s=∑ G
Rsg
rBe the influence value of input variable to overall output; If the sum that influences of n variable in the overall variable accounts for more than 90% of general impacts value, then choosing this n variable is the input variable of network model;
Secondly, finally the input and output parameter of selected model is:
(1) work output of BP network model: consider that whole model is the prediction that realizes the steel-making terminal time, so be chosen as the output of converter steelmaking blow end point time T (unit: second) as model;
(2) input of BP network model: because with of the output of converter steelmaking blow end point time T as model, the value of considering the whole process middle and later periods is relatively stable, so after the comparative analysis, spectrum light intensity average, image average, the ratio between the two of selected 400-500 time period second, the blowing pairing time of image value maximum point in latter stage, the image value of this maximum point correspondence, spectrum light intensity value and between the two above seven variablees of ratio as the input parameter;
At last, set up BP neural network terminal point predictive model based on the input/output variable of spectrum light intensity and characteristics of image change curve and model, utilize a large amount of converter steelmaking blowing data of collection in worksite, after the training of BP neural network terminal point predictive model, the terminal point of scene blowing heat is carried out the real-time estimate of network model, utilize the terminal point predictor that this model obtains and the on-the-spot comparing result of terminal point actual value, realize the terminal point predictive control of converter.
6. furnace mouth radiation information according to claim 5 merges and is used to make steel the method for online terminal point control, and the image acquisition and processing software in second step is characterized in that:
(1) creates the window of catching of fire door flame image information: call the capCreateCaptureWindow function, generate video capture window;
(2) three call back functions using of Accreditation System: mistake call back function capSetCallbackOnError, state call back function capSetCallbackOnStatus and video flowing call back function capSetCallbackOnVideoStream, setting by the video flowing call back function, make the RGB color model information of the fire door flame image that collects, through obtaining the chrominance information in the corresponding hsv color model after the color model conversion, promptly earlier the pairing data sheet of the change over condition of two kinds of color model is built, in call back function, add the software code that the RGB color model is converted to the hsv color model; Then each two field picture of gathering is according to pixels extracted, and search position in the pairing HSV table of this rgb value; Then the essential numerical value to all pixels of this two field picture superposes, and obtains and draw the three-dimensional plot of chromatic(ity)component; Last finding in this three-dimensional plot in the converting process when the reaching home of certain colourimetric number correspondence changes tangible curve, and draws; What time following note simultaneously: get rid of the more violent variation that occurs once in a while, this is to disturb the possibility that causes bigger; Select the number of pixels value to change metastable that colourity rank; Should not select background more colourity rank to occur; Destination county is bessemerized in corresponding closing under this colourity rank has than obvious variation, changes obviously relatively being beneficial to endpoint;
(3) call function capCaptureGetSetup is provided with the frame number that acquisition window and per second are gathered;
(4) window and video capture device are caught in call function capDriveConnect connection, and use the capDriverGetCaps function to return the function of sampler;
(5) call function capCaptureAbort finishes video.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200910035884 CN101698896B (en) | 2009-09-28 | 2009-09-28 | System and method for steel-making online end-point control through furnace mouth radiation information fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200910035884 CN101698896B (en) | 2009-09-28 | 2009-09-28 | System and method for steel-making online end-point control through furnace mouth radiation information fusion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101698896A true CN101698896A (en) | 2010-04-28 |
| CN101698896B CN101698896B (en) | 2013-01-30 |
Family
ID=42147361
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 200910035884 Expired - Fee Related CN101698896B (en) | 2009-09-28 | 2009-09-28 | System and method for steel-making online end-point control through furnace mouth radiation information fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101698896B (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102175178A (en) * | 2011-02-18 | 2011-09-07 | 华南理工大学 | System and method for measuring diffusion flame frontal surface three-dimensional structure of motion fire source |
| CN102392095A (en) * | 2011-10-21 | 2012-03-28 | 湖南镭目科技有限公司 | Termination point prediction method and system for converter steelmaking |
| CN102906281A (en) * | 2010-06-02 | 2013-01-30 | 西门子Vai金属科技有限责任公司 | Method for determining the time of ignition in the top-blowing process |
| CN104531936A (en) * | 2014-12-01 | 2015-04-22 | 南华大学 | On-line measure method for carbon content of molten steel in revolving furnace based on flame image characteristic |
| CN105678332A (en) * | 2016-01-08 | 2016-06-15 | 昆明理工大学 | Converter steel-making endpoint determination method and system based on flame image CNN recognizing and modeling process |
| CN106153551A (en) * | 2015-04-10 | 2016-11-23 | 南京理工大学 | Converter steel-smelting molten steel carbon content based on SVM online Real-time and Dynamic Detection system |
| CN106153556A (en) * | 2015-04-10 | 2016-11-23 | 南京理工大学 | Pneumatic steelmaking carbon content dynamic detection system |
| CN106148637A (en) * | 2015-04-10 | 2016-11-23 | 南京理工大学 | The pneumatic steelmaking carbon content dynamic detection system of high stable |
| CN106153552A (en) * | 2015-04-10 | 2016-11-23 | 南京理工大学 | Converter steel-smelting molten steel carbon content online Real-time and Dynamic Detection system |
| CN106153550A (en) * | 2015-04-10 | 2016-11-23 | 南京理工大学 | Converter steel-smelting molten steel carbon content based on SVM online Real-time and Dynamic Detection method |
| CN106153553A (en) * | 2015-04-10 | 2016-11-23 | 南京理工技术转移中心有限公司 | Converter steel-smelting molten steel carbon content online Real-time and Dynamic Detection system |
| CN109711004A (en) * | 2018-12-11 | 2019-05-03 | 重庆邮电大学 | A kind of optical fibre refractivity big data prediction technique |
| CN112734722A (en) * | 2021-01-08 | 2021-04-30 | 昆明理工大学 | Flame endpoint carbon content prediction method based on improved complete local binary pattern |
| CN113718082A (en) * | 2021-08-17 | 2021-11-30 | 河南省冶金研究所有限责任公司 | Prediction system and method for judging steelmaking end point temperature of converter based on flame image |
| CN114058778A (en) * | 2021-11-18 | 2022-02-18 | 中国安全生产科学研究院 | Steelmaking equipment temperature acquisition safety monitoring system |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1588346A (en) * | 2004-08-30 | 2005-03-02 | 邢台钢铁有限责任公司 | Method for predicting converter terminal point using artificial nurve network technology |
| TW200628536A (en) * | 2004-11-30 | 2006-08-16 | Ajinomoto Kk | Curable resin composition |
-
2009
- 2009-09-28 CN CN 200910035884 patent/CN101698896B/en not_active Expired - Fee Related
Cited By (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102906281A (en) * | 2010-06-02 | 2013-01-30 | 西门子Vai金属科技有限责任公司 | Method for determining the time of ignition in the top-blowing process |
| CN102906281B (en) * | 2010-06-02 | 2014-06-04 | 西门子Vai金属科技有限责任公司 | Method for determining the time of ignition in the top-blowing process |
| CN102175178B (en) * | 2011-02-18 | 2012-10-31 | 华南理工大学 | System and method for measuring diffusion flame frontal surface three-dimensional structure of motion fire source |
| CN102175178A (en) * | 2011-02-18 | 2011-09-07 | 华南理工大学 | System and method for measuring diffusion flame frontal surface three-dimensional structure of motion fire source |
| CN102392095A (en) * | 2011-10-21 | 2012-03-28 | 湖南镭目科技有限公司 | Termination point prediction method and system for converter steelmaking |
| CN102392095B (en) * | 2011-10-21 | 2013-09-11 | 湖南镭目科技有限公司 | Termination point prediction method and system for converter steelmaking |
| CN104531936B (en) * | 2014-12-01 | 2016-08-31 | 南华大学 | Converter molten steel carbon content On-line Measuring Method based on Flame Image Characteristics |
| CN104531936A (en) * | 2014-12-01 | 2015-04-22 | 南华大学 | On-line measure method for carbon content of molten steel in revolving furnace based on flame image characteristic |
| CN106153552A (en) * | 2015-04-10 | 2016-11-23 | 南京理工大学 | Converter steel-smelting molten steel carbon content online Real-time and Dynamic Detection system |
| CN106153556A (en) * | 2015-04-10 | 2016-11-23 | 南京理工大学 | Pneumatic steelmaking carbon content dynamic detection system |
| CN106148637A (en) * | 2015-04-10 | 2016-11-23 | 南京理工大学 | The pneumatic steelmaking carbon content dynamic detection system of high stable |
| CN106153550A (en) * | 2015-04-10 | 2016-11-23 | 南京理工大学 | Converter steel-smelting molten steel carbon content based on SVM online Real-time and Dynamic Detection method |
| CN106153553A (en) * | 2015-04-10 | 2016-11-23 | 南京理工技术转移中心有限公司 | Converter steel-smelting molten steel carbon content online Real-time and Dynamic Detection system |
| CN106153551A (en) * | 2015-04-10 | 2016-11-23 | 南京理工大学 | Converter steel-smelting molten steel carbon content based on SVM online Real-time and Dynamic Detection system |
| CN106153553B (en) * | 2015-04-10 | 2019-11-15 | 南京理工技术转移中心有限公司 | The online Real-time and Dynamic Detection system of converter steel-smelting molten steel carbon content |
| CN105678332B (en) * | 2016-01-08 | 2020-01-10 | 昆明理工大学 | Converter steelmaking end point judgment method and system based on flame image CNN recognition modeling |
| CN105678332A (en) * | 2016-01-08 | 2016-06-15 | 昆明理工大学 | Converter steel-making endpoint determination method and system based on flame image CNN recognizing and modeling process |
| CN109711004A (en) * | 2018-12-11 | 2019-05-03 | 重庆邮电大学 | A kind of optical fibre refractivity big data prediction technique |
| CN109711004B (en) * | 2018-12-11 | 2023-03-28 | 重庆邮电大学 | Optical fiber refractive index big data prediction method |
| CN112734722A (en) * | 2021-01-08 | 2021-04-30 | 昆明理工大学 | Flame endpoint carbon content prediction method based on improved complete local binary pattern |
| CN112734722B (en) * | 2021-01-08 | 2022-09-13 | 昆明理工大学 | Flame endpoint carbon content prediction method based on improved complete local binary pattern |
| CN113718082A (en) * | 2021-08-17 | 2021-11-30 | 河南省冶金研究所有限责任公司 | Prediction system and method for judging steelmaking end point temperature of converter based on flame image |
| CN114058778A (en) * | 2021-11-18 | 2022-02-18 | 中国安全生产科学研究院 | Steelmaking equipment temperature acquisition safety monitoring system |
| CN114058778B (en) * | 2021-11-18 | 2022-09-20 | 中国安全生产科学研究院 | Steelmaking equipment temperature acquisition safety monitoring system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101698896B (en) | 2013-01-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101698896B (en) | System and method for steel-making online end-point control through furnace mouth radiation information fusion | |
| CN102206727A (en) | Converter steelmaking endpoint determination method and system, control method and control system | |
| CN110438284B (en) | Intelligent tapping device of converter and control method | |
| CN102876838B (en) | Carbon content and system for detecting temperature in a kind of converter | |
| CN103954334B (en) | Fully automatic image pickup type water meter verification system and operating method thereof | |
| CN111593151B (en) | On-line detection method for depth of blast furnace tap hole | |
| CN109068073A (en) | A kind of thermal infrared imager autofocus system and method with temperature-compensating | |
| CN106153550A (en) | Converter steel-smelting molten steel carbon content based on SVM online Real-time and Dynamic Detection method | |
| CN101726491A (en) | Method and device for automatically acquiring dynamic images of tail cross section of sintering machine | |
| CN109612512A (en) | A kind of multi-modal integrated testing platform of space base electro-optical system and test method | |
| CN109829973A (en) | A kind of three-dimension visible sysem based on light application | |
| CN106755683A (en) | A kind of blast-furnace roasting band temperature field detection device based on colorimetric method | |
| CN202193799U (en) | Converter steelmaking endpoint judging system and control system thereof | |
| CN111784803A (en) | Automatic acquisition system and method for drill core correlation data | |
| CN117237369B (en) | Blast furnace iron notch opening depth measurement method based on computer vision | |
| CN108885265A (en) | Range image processing unit, range image acquisition device and range image processing method | |
| CN101846623A (en) | Method for detecting reflectivity of coal or organic rock vitrinite and special equipment thereof | |
| CN106148637B (en) | The pneumatic steelmaking carbon content dynamic detection system of high stable | |
| CN102968644A (en) | Method for predicting smelting finishing point of argon-oxygen refined iron alloy | |
| CN106153551A (en) | Converter steel-smelting molten steel carbon content based on SVM online Real-time and Dynamic Detection system | |
| CN111598874A (en) | Mangrove canopy density survey method based on intelligent mobile terminal | |
| CN107489420A (en) | Sublevel caving method without sill pillar Rock fragmentation is distributed remote real time monitoring system and method | |
| CN216524095U (en) | Thermal state monitoring system for empty ladle and tapping process of steel ladle | |
| CN116405638A (en) | Intelligent inspection device, inspection system and inspection method based on multielement sensor | |
| CN202274996U (en) | Performance test system for tank stabilizer |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20130130 Termination date: 20180928 |