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

On-line measurement method of molten state

technical field

The present invention relates to the identification and quantification of information from molten states including slag, flux, metal and matte. A method based on principal component analysis based on image data taken from a molten state surface is used.

current technology

Multivariate image processing provides robust methods for extracting information from image data. This approach has been successfully used in several image processing applications, such as satellite image data and medicine. However, this method has not been applied to the online measurement of the molten state in existing applications.

Reliable real-time measurement of the process is an important factor in the development of any control system. For high temperature molten state processing (such as steelmaking), it is difficult and expensive to achieve real-time measurements because of the extreme conditions. Currently, there are several methods for gathering information on molten state, such as detection of relevant surface areas in molten state and assessment of whether said state is in complete melting or not, all based on operator's human visual observation. Therefore, there is a need for more reliable online measurement of molten state.

It is an object of the present invention to detail and quantify online information on molten states to detect relevant surface areas in molten states, determine whether they are in fully molten states, and predict the temperatures of these states within reasonable computational time. Because the calculation time is rather fast, the method can be used as an online measuring device and integrated into a control system.

invention disclosure

According to the present invention, there is provided a method of characterizing a molten state using principal component analysis of image data taken from a surface in the molten state. The method includes: (a) developing a standard; and (b) using the standard to identify and quantify online image data. To develop the criteria, the development process included the following steps: (i) acquiring a digital image of the surface in the molten state; (ii) performing a principal component analysis of the image; (iii) judging the The standard values of the principal components, which will be used to determine the properties of the online image. In using the standard for identifying and quantifying online image data, the following steps are performed: (a) acquire a digital image of the surface in the molten state; (b) perform a principal component analysis of the image; (c) compare the analysis to the standard values of the principal components making a comparison to determine an attribute of said image; and (d) quantifying an attribute of said image.

Brief description of the drawings

Figure 1 is a schematic diagram of on-line measurement of molten state. Basically, the system consists of three main parts, the molten state being measured, a digital camera for acquiring image data, and a computer for processing this image data;

Figure 2 shows an example of an RGB image taken from a molten state;

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

Figure 4 is an example of the first two principal component plots (t ₁ vs t ₂ ) of the image in Figure 2;

Figure 5 is a plot associated with the predicted bare metal area, together with the inert gas flow injected from the bottom of the vessel, as a function of gas injection time;

Figure 6 is a plot relating the temperature of the bath and the average second principal component _t2 of the slag properties.

BEST MODE FOR CARRYING OUT THE INVENTION

The schematic diagram of the on-line measurement system of the molten state is generally indicated by the numeral 20 in FIG. 1 . As shown, a system 20 is used to measure the molten state in a vessel 22 and includes a digital camera 24 for acquiring image data and a computer 26 for processing the image data.

The first step in measuring molten state properties (such as splitting of the slag surface, localized solidification of the slag phase, or temperature of the slag) is to capture the molten slag in RGB (Red-Green-Blue) format using a digital camera 24. Image data of the slag surface. The RGB format is commonly used to represent high-resolution color images, where each pixel is represented by three numerical values—one each representing the red, green, and blue (RGB) components of the pixel's color. In the color image of Figure 2, the white areas of the image correspond to bare metal, the yellow areas correspond to thin slag, the brown areas correspond to fluid slag, and the black areas correspond to solidified Slag (solidified slag). Such an image can be represented schematically as a stack of three nxm pixel images. From a mathematical point of view, the image can be regarded as a matrix _Im with a size of n×m×3, as shown in FIG. 3 . Such an image taken from the surface of a ladle is visualized in FIG. 2 . The digital image data is transferred to a process control computer 26 for determining the properties of the molten state based on the information obtained from the image data.

Principal component analysis, or PCA, is used in processing the obtained image data of the molten state. PCA is a multivariate statistical procedure applied to a set of variables (which are highly correlated) to reveal its principal components (or score vectors). These principal components are linear combinations of original variables, these variables are independent of each other, and the main information in the original variables can be obtained in its first few principal components [Jackson, 1991].

Multivariate statistical methods, such as principal component analysis (PCA) and local least squares (PLS), have been successfully used in multivariate image analysis [Esbensen et al., 1989; Geladi et al., 1989; Grahn et al., 1989; Bharati and MacGegor , 1998]. Using these methods, a set of high-dimensional and highly correlated data can be projected into a set of uncorrelated data with dimensionality reduction. In the present invention the PCA method is used to evaluate the image of the molten state.

In order to simplify this problem, the three-dimensional matrix Im _(m×n×3) in FIG. 3 is expanded into an expanded two-dimensional matrix X _{((n, m)×3)} , as shown in FIG. 3 .

The unfolded image matrix X is decomposed by performing principal component analysis [Jackson, 1991]. The relationship between the original matrix and its principal components is given by the following equation:

$\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; T</span>}{\mathrm{<span\; class="notranslate">\; P</span>}}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

Among them, X is the expanded form of I _m ; T is the score matrix (score matrix); P is the loading matrix (loading matrix); E is the residual matrix.

Assuming that all information in the image is preserved in the first two principal components, namely _t1 and _t2 , then the X matrix is approximated as:

$\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$

The score vector t _i is the linear combination of the variables (columns) in the data matrix X that account for the largest variance in the multivariate data. These vectors have the property of being orthogonal to each other. The loading vector p _i is the eigenvector (in descending order) of the variance-covariance structure (X ^T X ) in this data matrix. These vectors have the property of being mutually orthogonal (ie P ^T P = I; where I is the identity matrix). Based on the properties of the score vector and loading vector, the value of the score matrix T can be obtained by multiplying P by X [Geladi et al., 1989]:

T=XP (4)

Assuming below that all information in the image is retained in the first two principal components, the combination of the first two score vectors (t ₁ and t ₂ ) is basically equal to these pixels [Bharati and Macgregor, 1998], as in equation (3) arithmetic shown. Combinations of these principal components can thus be used to extract information from (or to distinguish the material of) the aforementioned images. Furthermore, the average value of pixel brightness for each wavelength is denoted by _t1 , while the contrast and difference between pixel luminances of different wavelengths is denoted by _t2 [Bharati and Macgregor, 1998]. According to the invention, the average value of _t1 or _t2 can be used to characterize the properties of the image, for example to determine the temperature.

The image data of the image represented in FIG. 2 is developed to obtain a matrix X by using the processing procedure given in FIG. 3 . Analyzing the principal components of the matrix X using standard procedures of PCA such as [Jackson, 1991] gives the values of the loading vector p _i and the eigenvalues indicated in Table 1. All calculations in this report were performed in a high-level computer language, MATLAB ^™ version 6 and MATLAB ^™ Image Processing Toolbox (Image Processing Toolbox) version 3.

Table 1. Loading vectors and eigenvalues for the images represented in Fig. 3 Score 1 2 3 % total variance of loading vector eigenvalues 0.70020.61890.35580.245884 -0.57380.19150.79630.038713.23 -0.42470.7617 -0.48930.00812.77

As shown in Table 1, the cumulative total variance of the first two principal components is 97.23% (84.00% and 13.23%, respectively). Therefore, it is reasonable to assume that the main information in the image is retained in the first two principal components; a combination of these principal components can be used to extract information from the image (or to distinguish the material in the image), and then use only the first two principal components to perform subsequent analysis. The loading vectors for these two principal components are:

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

Shown in FIG. 4 is a scatter plot of the first two score vectors (t ₁ vs. t ₂ ). The graph has 3110400 score combinations plotted, each representing a 2160x1440 pixel location in the original image. It should be noted that there is an overlap of several points in this figure because a large number of pixels are drawn into the figure and similar features in the original image yield similar combinations of score vectors.

By projecting the values of the first two principal components ( _t1 and _t2 ) of a pixel to the corresponding image, the information of the original image explained by the combined value of _t1 and _t2 can be identified. The results of this step can be used to delineate the pixel level. Using the combined values of _t1 and _t2 , together with the information represented by the area of one pixel, the area of the object in the image can be determined. The results of this step can be used to delineate the pixel levels given in Table 2. By using this method, if the area represented by a pixel is known, the total area under consideration can be determined by multiplying the area of a pixel by the number of points of the same group in FIG. 4 . For example, using this method to calculate the area of the spout eye or bare metal observed in the ladle in Figure 2 yields a value of 1.764 square meters.

Table 2. Information for mapping the first two principal components into the original image t ₁ t ₂ The original image 1.1475 to 1.26340.6138 to 1.14750.0790 to 0.6138 0.2995 to 0.5322 - 0.2245 to 0.2995 - 0.3356 to -0.1998 Eye (white) thin slag (yellow) fluid slag and barrel wall (brown)

An example of predicted bare metal area is shown in Figure 5, in which inert gas flow rate is incorporated as a function of gas injection time. As clearly shown in the figure, the bare metal area is a function of the inert gas flow rate. From the foregoing discussion it is clear that the method according to the invention can be used to characterize surface properties such as fragmentation of slag or bare metal and local solidification of slag and to quantify surface properties with respect to their area.

Since the second principal component _t2 represents the contrast or difference between pixel brightness at different wavelengths [Bharati and MacGregor, 1998], the average value of the second principal component was used to quantify the temperature of the melt pool. The relationship between temperature and brightness is also a function of the reflective properties of the material, which is to some extent a function of the ladle chemistry.

Fig. 6 shows the relationship between the molten pool temperature and the average second principal component _t2 for indicating the slag grade. As shown in Fig. 6, it can be obtained that the temperature of the molten pool can be represented by the average value of the second principal component _t2 . Therefore, it can be concluded that the temperature of the molten state (including slag, flux, metal and matte) can be determined using the average value of _t2 .

In order to use the image processing process results as real-time measurement data, it is important to be able to process the image in a reasonable amount of time. In the present work, the processing time for measuring the bare metal area is a few seconds. Therefore, it can be concluded that the calculation speed is sufficient for an in-line measurement system. The above calculations were performed using MATLAB ^(TM) Version 6 and MATLAB ^(TM) Image Processing Toolbox Version 3 on an IBM ^(TM) Pentium III/800MHz, 250MHz RAM personal computer running under Window ^(TM) 2000 environment.

Claims (8)

1. A method of identifying and quantifying information from a molten state product having an exposed surface area, said method comprising the steps of: a) develop standards for online evaluation of digital images; and b) performing the online assessment, wherein the criteria are developed using the following steps: i) acquiring a digital image of the exposed surface area of a product in a molten state to generate standard image data; ii) performing principal component analysis on said standard image data to identify score vectors _t1 and _t2 as features of said standard image data; iii) correlating the values of said score vectors _t1 and _t2 with characteristic attributes of said molten state product to define standard values for _t1 and _t2 ; And the evaluation is carried out using the following steps: iv) acquiring a digital image of an exposed surface area of a product in a molten state to generate online image data; v) performing principal component analysis on said online image data to define score vectors _t1 and _t2 that characterize said online image data; vi) assigning a characteristic attribute to the region of the online image data according to the standard values of _t1 and _t2 ; vii) Build an output of said feature attributes to identify and quantify states. 2. The method of claim 1, wherein the molten state includes any one of slag, flux, metal, matte, and glass. 3. The method of claim 1, wherein the digital image is taken in the visible spectrum. 4. The method of claim 1, wherein the digital image comprises a matrix of pixel elements of luminance values measured in at least three wavelength ranges. 5. The method of claim 4, wherein the pixel elements of the digital image have varying intensities of red, green, and blue colors. 6. The method of claim 1, wherein said characteristic attributes corrected with said score vectors _t and _t are selected from the group consisting of: state identification of molten state products; surface area occupied by each identified state; The temperature of each identified state. 7. A method of monitoring a steelmaking ladle having a high temperature molten state to distinguish regions having bare metal, bare metal covered with slag, and fluid slag, said method comprising the steps of: a) develop standards for online evaluation of digital images; and b) performing the online assessment, wherein the criteria are developed using the following steps: i) acquiring a digital image of the exposed surface area of the steelmaking ladle to generate standard image data; ii) performing principal component analysis on said standard image data to define score vectors _t1 and _t2 as attributes of said standard image data; iii) correlating the values of said score vectors _t1 and _t2 with characteristic attributes of said molten state product to define standard values for _t1 and _t2 ; And the evaluation is carried out using the following steps: iv) acquiring a digital image of the exposed surface area of the product in the molten state to generate online image data; v) performing principal component analysis on said online image data to define score vectors _t1 and _t2 as attributes of said online image data; vi) assigning feature attributes to the region of the online image data according to the standard values of _t1 and _t2 ; vii) Build an output of said feature attributes to identify and quantify states. 8. The method of claim 7, wherein said feature attributes corrected with said score vectors _t and _t are selected from the group consisting of: state identification; surface area occupied by each identified state; each identified state temperature.

Images