CN1668920A - Method for online measurement of molten phases - Google Patents

Abstract

A method for identifying and quantifying information about molten phases, including slags, fluxes, metal and matte using a multivariate image analysis approach. Using this procedure, the properties of molten phases such as disruption of slag, the size of bare metal, partial solidification of slag, and temperature of slag can be accurately determined within a reasonable computation time. Moreover, this method can be implemented as an online measurement tool of molten phases.

Description

The method of the on-line measurement of molten condition

Technical field

The present invention relates to discerning from the information of molten condition (comprising slag (slag), flux (flux), metal and matte (matte)) and quantizing.Use is based on the method for the principal component analysis (PCA) of the view data of taking from surface of molten phases.

Prior art

The multivariate Flame Image Process provides from the reliable method of image data extraction information.This method successfully has been applied in the application of several Flame Image Process, for example satellite image data and medical domain.But also this method is not applied to the on-line measurement of molten condition in existing the application.

To measuring in real time reliably of processing procedure is the key factor of any control system of exploitation.Handle (for example steel-making) for high temperature fused state, because extreme condition, realizing measuring in real time is difficulty and expensive.At present, have several method can collect the information of molten condition, for example detect the relevant surface areas of molten condition and whether above-mentioned state is in the assessment of fusing fully, these all are based on operator's artificial visual observation.Therefore, need the on-line measurement of more reliable molten condition.

The objective of the invention is in rational computing time, to record and narrate in detail and quantize online information about molten condition with the relevant surface areas that detects molten condition, the state that determines whether to be in fusing fully and the temperature of estimating these states.Because computing time is quite fast, so this method can be used as the on-line measurement device and is integrated in the control system.

Disclosure of the Invention

According to the present invention, the principal component analysis (PCA) of view data that the surface that a kind of use takes from molten condition is provided is with the method for molten condition characterization.This method comprises: (a) exploitation standard; And (b) use this standard to determine and quantize online view data.In order to develop described standard, the process of exploitation comprises the following steps: that (i) obtains the digital picture of surface of molten phases; (ii) carry out the principal component analysis (PCA) of this image; (iii) judge the standard value of described major component on the basis of the knowledge of molten condition attribute, it will be used for determining online attributes of images.Use identification and quantizing in the standard of online view data, carrying out the following step: the digital picture of (a) obtaining surface of molten phases; (b) carry out the principal component analysis (PCA) of this image; (c) standard value that will analyze with major component compares to determine above-mentioned attributes of images; And (d) quantize above-mentioned attributes of images.

Brief Description Of Drawings

Fig. 1 is the schematic diagram of the on-line measurement of molten condition.Basically, this system comprises three major parts, and promptly measured molten condition is used to obtain the digital camera of view data, and the computing machine that is used to handle this view data;

Fig. 2 represents to take from the example of the RGB image of molten condition;

Fig. 3 is the schematic diagram of principal component analysis (PCA) process;

Fig. 4 is preceding two major components drawing (plot) (t of the image among Fig. 2 ₁To t ₂) example;

Fig. 5 is the drawing that is associated with the naked metallic area of prediction, and it is with the inert gas flow that injects from container bottom, as the function of gas injection time;

Fig. 6 is and the temperature of molten bath (bath) and the average Second principal component, t of slag attribute ₂Relevant drawing.

Preferred forms of the present invention

The schematic diagram of the on-line measurement system of molten condition is represented by the numeral among Fig. 1 20 usually.As shown in the figure, system 20 is used for measuring the molten condition at container 22, comprises being used to the computing machine 26 that obtains the digital camera 24 of view data and be used to handle this view data.

The first step of measuring molten condition attribute (local solidification of for example division of the top of the slag, slag phase or the temperature of slag) is to use digital camera 24 to catch the view data on this slag surface with RGB (Red-Green-Blue, R-G-B) form.Rgb format is the mode that is generally used for representing high-resolution colour picture, and wherein each pixel is represented red, green, blue (RGB) component of this pixel color respectively with three numeric representations one.In the coloured image of Fig. 2, the white portion of this image is corresponding to naked metal (baremetal), yellow area is corresponding to thin slag (thin slag), and brown areas is corresponding to fluid slag (fluidslag), and black region is corresponding to solidified slag (solidified slag).Such image can be schematically represented as the heap of three n * m pixel image.On the mathematics angle, this image can be regarded matrix I as _m, its size is n * m * 3, as shown in Figure 3.The image that is taken from steel-making bucket (ladle) surface like this is visual in Fig. 2.Digital Image Data is sent in the process computer 26 so that determine the attribute of molten condition on the basis of the information that is obtained by view data.

In the process of the view data of handling the molten condition that obtains, use principal component analysis (PCA) or be called PCA.PCA is the major component (perhaps score vector (score vector)) of multivariate statistics process (these variablees are highly related) to disclose it that is applied to one group of variable.These major components are the linear combination of original variable, and these variablees are separate, and can obtain the main information [Jackson, 1991] in the original variable in its first few major component.

Multivariable Statistical Methods, for example principal component analysis (PCA) (PCA) and partial least squares (PLS) have been successfully used to multivariate graphical analysis [Esbensen etc., 1989; Geladi etc., 1989; Grahn etc., 1989; Bharati and MacGegor, 1998].Use these methods, one group of higher-dimension and data height correlation can be one group by projection and have the incoherent data that dimension reduces.The PCA method is used to assess the image of molten condition in the present invention.

In order to simplify this problem, the three-dimensional matrice I of Fig. 3 _{M (m * n * 3)}Be unfolded two-dimensional matrix X for expansion _{((n, m) * 3)}, as shown in Figure 3.

The image array X that launches is by carrying out principal component analysis (PCA) be decomposed [Jackson, 1991].Relation between original matrix and its major component provides by following equation:

$X=\underset{i}{\mathrm{\Σ}}{t}_i{p}_i^T+E=T{P}^T+E---\left(2\right)$

Wherein, X is I _mThe expansion form; T be sub matrix (score matrix); P is loading matrix (loading matrix); E is a residual matrix.

Suppose that all information in the image all are retained in preceding two major components, i.e. t ₁And t ₂, the X approximate matrix is so:

$\hat{X}=\underset{i=1}{\overset{2}{\mathrm{\Σ}}}{t}_i{p}_i^T---\left(3\right)$

Score vector t _iBe the linear combination of explaining the variable (row) among the data matrix X of the maximum deviation in the multivariate data.These vectors have mutually orthogonal character.Load vector p _iBe variance-covariance (variance-covariance) structure (X in this data matrix ^TX) proper vector (with descending).It (is P that these vectors have mutually orthogonal character ^TP=I; Wherein I is a unit matrix).Based on the character of score vector and load vector, the value of score matrix T can take advantage of P to obtain [Geladi etc., 1989] by X:

T＝XP????????????????????????????????????(4)

All information all are retained in preceding two major components preceding two score vector (t in the hypothesis image below ₁And t ₂) combination equate [Bharati and Macgregor, 1998] substantially with these pixels, shown in equation (3) arithmetic.Therefore, the combination of these major components can be used for information extraction from above-mentioned image (perhaps distinguishing the material of above-mentioned image).In addition, the mean value of the pixel intensity of each wavelength is by t ₁Expression, and contrast between the pixel intensity of different wave length and difference are by t ₂Expression [Bharati and Macgregor, 1998].According to the present invention, t ₁Or t ₂Mean value can be used for the attribute of token image, for example determine temperature.

The view data of the image of representing among Fig. 2 launches to obtain matrix X by the processing procedure that use Fig. 3 provides.The major component of the standard operation analysis matrix X of use PCA for example [Jackson, 1991] provides load vector p _iValue and table 1 in the expression eigenwert.All of this report are calculated with high-level [computer and are carried out, i.e. MATLAB ^TMVersion 6 and MATLAB ^TMFlame Image Process tool box (ImageProcessing Toolbox) version 3.

The load vector sum eigenwert of the image of representing among table 1. Fig. 3

Score	????1	????2	????3
				Load vector eigenwert population variance %	????0.7002 ????0.6189 ????0.3558 ????0.2458 ????84	????-0.5738 ????0.1915 ????0.7963 ????0.0387 ????13.23	????-0.4247 ????0.7617 ????-0.4893 ????0.0081 ????2.77

As shown in table 1, the population variance of the accumulation of preceding two major components is 97.23% (being respectively 84.00% and 13.23%).Therefore, suppose that it is rational that main information in the image is retained in preceding two major components; The combination of these major components can be used for information extraction from image (the perhaps material in the difference image), only carries out subsequently analysis with preceding two major components then.The load vector of these two major components is:

$p_1^T=\left[\mathrm{0.70020.61890.3558}\right]$ With $p_2^T=[-\mathrm{0.57380.19150.7963}]$

That represent in Fig. 4 is preceding two score vector (t ₁To t ₂) scatter-plot (scatter plot).This figure has 3110400 score combinations of being drawn, and each represents 2160 * 1440 locations of pixels in the original image.Should be noted that the overlapping that several points are arranged in this figure, this is because a large amount of pixel is painted among the figure into and the similar features in the original image produces similar score vector combination.

By preceding two major component (t with pixel ₁And t ₂) value project corresponding image, can discern by t ₁And t ₂The information of the original image explained of combined value.The result of this step can be used for describing pixel class.Use t ₁And t ₂Combined value, and unite information by the Regional Representative of a pixel, the area of object can be determined in the image.The result of this step can be used for the pixel class describing to provide in the table 2.By making in this way, if the area of the representative of a known pixel, the total area in considering so can be multiplied by by the area of a pixel in Fig. 4 same group point quantity and determine.For example, the value that the area of observed slotted eyes in the steel-making bucket that calculates in this way in Fig. 2 (spout eye) or naked metal is obtained is 1.764 square metres.

Table 2. is mapped to information in the original image with preceding two major components

??t 1	????t 2	Original image
			1.1475 to 1.2634 0.6138 to 1.1475 0.0790 to 0.6138	0.2995 to 0.5322-0.2245 to 0.2995-0.3356 to-0.1998	Thin slag (yellow) fluid slag of eye (white) and bucket wall (brown)

Fig. 5 has represented the example of the naked metallic area of prediction, in the figure, engages inert gas flow together as the function of gas injection time.Shown in clear among the figure, the area of naked metal is the function of inert gas flow.Clear from the discussion of front, the method according to this invention can be used for describing surface properties (for example local solidification of the division of slag or naked metal and slag) and is used to quantize surface properties about its area.

Because Second principal component, t ₂Be illustrated in contrast or difference [Bharati and MacGregor, 1998] between the pixel intensity of different wave length, so use the mean value of Second principal component, to quantize the temperature in molten bath.Relation between temperature and the brightness also is the function of the reflecting attribute of material, and it is the function of ladle sample composition (ladle chemistry) to a certain extent.

Fig. 6 represents bath temperature and average Second principal component, t ₂Between relation be used to represent various slag grades.As shown in Figure 6, therefrom can obtain the temperature in molten bath can be by Second principal component, t ₂Mean value representative.Therefore, can reach a conclusion: the temperature of molten condition (comprising slag, flux, metal and matte) can be used t ₂Mean value determine.

For with the image processing process result as real-time measuring data, the very important point is reasonably to handle this image in the time.In present work, the processing time of measuring naked metallic area is several seconds.Therefore, can draw a conclusion, computing velocity is enough for on-line measurement system.Aforementioned calculation is at IBM ^TMPentium III/800MHz, 250MHz RAM, at Window ^TMUse MATLAB in the personal computer that 2000 environment move down ^TMVersion 6 and MATLAB ^TMFlame Image Process tool box (Image Processing Toolbox) version 3 carries out.

Claims (8)

1. an identification and quantize method from the information of the molten condition product with exposed surface zone, described method comprises the following steps:

A) exploitation is used for the standard of the online evaluation of digital picture; And

B) carry out described online evaluation, wherein use the following step to develop described standard:

I) obtain the digital picture in the exposed surface zone of a molten condition product, to produce standard image data;

Ii) described standard image data is carried out principal component analysis (PCA), to confirm score vector t as described standard image data feature ₁And t ₂

Iii) with described score vector t ₁And t ₂Value and the characteristic attribute of described molten condition product carry out related, with definition t ₁And t ₂Standard value;

And use the following step to carry out described assessment:

The digital picture in exposed surface zone of iv) obtaining a molten condition product is to be created on the line image data;

V) described online view data is carried out principal component analysis (PCA), with the score vector t of definition as described online view data feature ₁And t ₂

Vi) according to described t ₁And t ₂Standard value be that a characteristic attribute is specified in the zone of described online view data;

Thereby vii) set up the output identification and the quantification state of described characteristic attribute.

2. the method for claim 1, wherein said molten condition comprises any one in slag, flux, metal, matte and the glass.

3. the method for claim 1, wherein said digital picture obtains in visible spectrum.

4. the method for claim 1, wherein said digital picture are included in the brightness value pixel element matrix of measuring at least three wavelength coverages.

5. method as claimed in claim 4, the pixel element of wherein said digital picture have the brightness of the red, green, blue color of variation.

6. the method for claim 1 is wherein used described score vector t ₁And t ₂The described characteristic attribute of proofreading and correct is selected from following group: the state recognition of molten condition product; The surface area that the state of each identification takies; The temperature of the state of each identification.

7. steel-making bucket with high temperature fused state of a monitoring has naked metal, is coated with the method in the zone of the naked metal of slag and fluid slag with difference, and described method comprises the following steps:

A) exploitation is used for the standard of the online evaluation of digital picture; And

B) carry out described online evaluation, wherein use the following step to develop described standard:

I) obtain the digital picture in the exposed surface zone of steel-making bucket, to generate standard image data;

Ii) described standard image data is carried out principal component analysis (PCA), with the score vector t of definition as described standard image data attribute ₁And t ₂

Iii) with described score vector t ₁And t ₂Value be associated with the characteristic attribute of described molten condition product, with the definition t ₁And t ₂Standard value;

And use the following step to carry out described assessment:

The digital picture in exposed surface zone of iv) obtaining the molten condition product is to be created on the line image data;

V) described online view data is carried out principal component analysis (PCA), with the score vector t of definition as described online view data attribute ₁And t ₂

Vi) according to described t ₁And t ₂Standard value be the regional specific characteristic attribute of described online view data;

Thereby vii) set up the output identification and the quantification state of described characteristic attribute.

8. method as claimed in claim 7 is wherein used described score vector t ₁And t ₂The described characteristic attribute of proofreading and correct is selected from following group: state recognition; The surface area that the state of each identification takies; The temperature of the state of each identification.

Images