Method and device for evaluating Si and S content in molten iron based on neural network and area method

Abstract

The invention relates to a method and device for evaluating Si and S contents in molten iron based on a neural network and an area method. The method comprises: extracting a mushroom-like characteristic area in a mushroom-like photo of molten iron, calculating a ratio A / N of Si pixel points A to all pixel points N and a ratio D / N of S pixel points D to all pixel points N according to color values ​​in pixel points in the mushroom-like characteristic area, optimizing and training a-A / N and d-D / N corresponding relationship of a molten iron Si and S content evaluation model based on a neural network and Si content a and S content d of a standard sample. The device comprises a processor, a photographing device, a detection box and a high-temperature molten iron mold. After photographing, the processor calculates the Si and S content evaluation model with A / N and D / N, so that Si and S content evaluation results can be quickly obtained, the operation process is effectively simplified, the efficiency, accuracy and safety of molten iron component evaluation are improved, the timeliness is higher and manual observation experience errors and scalding risks are avoided, so as to timely and accurately respond to blast furnace operation and ensure smooth smelting and safety.

Description

Technical Field

[0001] The invention belongs to the technical field of high-temperature molten iron detection, and specifically relates to a method and a device for evaluating Si and S contents in molten iron based on a neural network and an area method. Background Art

[0002] At present, blast furnace ironmaking in China accounts for more than 90% of the entire ironmaking industry. The state of the blast furnace directly affects the output of ironmaking, the cost of ironmaking, and the work intensity and difficulty of ironmaking workers. Due to the delayed effect of blast furnace operation, the impact of many operations will not be revealed through furnace condition data until an hour or even several hours later. If potential problems are not discovered and handled in time, it is likely to cause problems with the blast furnace condition. At the least, there will be air blocking and hanging materials, and at the worst, it may cause the blast furnace to be difficult to operate and the furnace to cool. The above problems can be avoided to a large extent by predicting the Si and S content of the molten iron at the taphole. Therefore, sampling the molten iron in front of the furnace to evaluate the Si and S content of the molten iron during the blast furnace ironmaking process is a very important task for ironmaking workers.

[0003] The methods and devices for evaluating Si and S content in molten iron in the prior art have the following defects: (1) Sampling molten iron in front of the furnace for laboratory testing has the defects of complicated work process and operation, high equipment cost, low efficiency, and certain delay. Generally, a laboratory tests one bag of molten iron once, which is difficult to test multiple times in a short period of time and cannot provide timely feedback on blast furnace operation. (2) The current method for real-time measurement of molten iron composition in front of the blast furnace is mainly to observe the carbon and "oil skin" formed on the mushroom-like surface of the molten iron during the solidification process with the naked eye, and to make judgments based on the long-term experience accumulated by the staff. Manual observation may bring the risk of high temperature burns, rely heavily on experience, and easily cause inaccurate judgments, leading to misjudgment of furnace conditions and causing production accidents. Summary of the invention

[0004] The present invention aims to solve at least one of the above-mentioned technical problems to a certain extent. The present invention provides a method and device for evaluating the Si and S contents of molten iron based on a neural network and an area method, so as to respond to blast furnace operations in a timely and accurate manner and ensure smooth smelting and safety.

[0005] The technical solution adopted by the present invention to solve its technical problem is:

[0006] The method for evaluating Si and S content in molten iron based on neural network and area method includes:

[0007] Collect photos of molten iron mushrooms, extract mushroom-like feature areas in the photos, calculate the number of all pixel points N in the mushroom-like feature areas, and calculate the ratio A / N of Si pixel points A to all pixel points N and the ratio D / N of S pixel points D to all pixel points N according to the color values ​​of the pixels in the mushroom-like feature areas;

[0008] Detect the Si content a and S content d of the molten iron samples corresponding to multiple photos, optimize and train the corresponding relationship of aA / N and dD / N based on the neural network, and obtain the molten iron Si and S content evaluation model;

[0009] Input A / N and D / N to calculate the Si and S content evaluation model, output Si content a and S content d, and obtain the evaluation results.

[0010] Furthermore, the mushroom-like feature area in the photo is extracted including:

[0011] The format and size of each photo are unified based on the bilinear interpolation method;

[0012] Perform weighted averaging on the RGB components of the color image to obtain a reasonable grayscale image;

[0013] Calculate the optimal threshold of the photo according to the maximum inter-class variance method;

[0014] The pixel values ​​of the pixel group that are greater than or less than the optimal threshold are set to black and white to obtain a binary grayscale image.

[0015] Furthermore, the RGB three components of the color image are weighted averaged according to the formula f(i,j)=0.30R(i,j)+0.59G(i,j)+0.11B(i,j), where R represents the red channel of the image; G represents the green channel of the image; and B represents the blue channel of the image.

[0016] According to the formula T = Max[w0(t)*(u0(t)-u) 2 +w1(t)*(u1(t)-u) 2 )] Calculate the optimal threshold of the photo, where w0 is the background ratio, u0 is the background mean, w1 is the foreground ratio, u1 is the foreground mean, and u is the mean of the entire image.

[0017] Furthermore, the number N of all pixels in the mushroom-like feature area is read and calculated using the CV2 program.

[0018] Furthermore, the RGB value of all pixel points N in the mushroom-like feature area = (red) + (green*256) + (blue*256*256), the Si pixel point A is calculated according to the RGB value range of the Si content, and the S pixel point D is calculated according to the RGB value range of the S content.

[0019] Furthermore, the corresponding relationship of aA / N and dD / N is optimized and trained based on the BP neural network, specifically including:

[0020] A / N and D / N are used as input layers, and training data is established with Si content a and S content d corresponding to A / N and D / N;

[0021] Use the Relu function as the activation function for training the BP neural network and establish the BP neural network structure;

[0022] According to the direction of the error gradient descent, the connection weights between the hidden layer neurons and the output layer neurons, and the connection weights between the input layer neurons and the hidden layer neurons are corrected;

[0023] The training is optimized with the goal of minimizing the overall error between the expected value and the estimated value of the output layer neurons.

[0024] Furthermore, the input value of the i-th neuron in the input layer of the BP neural network is X i , the output value of the i-th neuron in the hidden layer is Y i , the output value of the i-th neuron in the output layer is Z i ; The network connection weight between the input layer neurons and the hidden layer neurons is M ij , the network connection weight between the hidden layer neurons and the output layer neurons is N j ; The activation threshold of the jth unit in the hidden layer is α j , the activation threshold of the kth unit in the output layer is β k ;

[0025] Output components of hidden layer neurons In the above formula: g1 is the transfer function of the hidden layer neurons, m is the number of neurons in the input layer;

[0026] Output components of output layer neurons In the above formula: g2 is the transfer function of the output layer neurons, and p is the number of hidden layer neurons.

[0027] Furthermore, the Relu function is: In the above formula: Y i is the neuron data of each layer, M i is the weight between neurons, b i is the bias that matches the weight.

[0028] Furthermore, the connection weight correction formula between the hidden layer neurons and the output layer neurons is: In the above formula: N j (n+1) is the connection weight between the hidden layer neurons and the output layer neurons after the (n+1)th modification, N j (n) is the connection weight between the hidden layer neurons and the output layer neurons after the nth correction, η is the learning rate of the BP neural network, η∈(0,1), and E(n) is the forward transmission error of the BP neural network after the nth training.

[0029] Furthermore, the connection weight correction formula between the input layer neurons and the hidden layer neurons is: In the above formula: M ij (n+1) is the connection weight between the input layer neurons and the hidden layer neurons after the (n+1)th modification, M ij (n) is the connection weight between the input layer neurons and the hidden layer neurons after the nth correction, η is the BP neural network learning rate, η∈(0,1), and E(n) is the forward transmission error of the BP neural network after the nth training.

[0030] Furthermore, A / N and D / N are input into the BP neural network structure to calculate the output layer estimated value Z l After the BP neural network passes the training data forward, the overall error is In the above formula: q is the number of neurons in the output layer, t l is the expected value of the output layer neuron.

[0031] A device for evaluating Si and S content in molten iron comprises a processor and a frame, wherein the frame is provided with a photographing device and an openable and closable detection box, the detection box is provided with a cooling structure and an exhaust structure, a high-temperature molten iron mold is provided in the detection box, the photographing device can take pictures of the inside of the high-temperature molten iron mold through the detection box and transmit the pictures to the processor, and the processor obtains the evaluation result based on the method for evaluating Si and S content in molten iron based on neural network and area method as described in any one of the above items.

[0032] Furthermore, the detection box includes a high temperature resistant sheet and a box body, the high temperature resistant sheet is connected to a first lifting mechanism, a transparent isolation sheet corresponding to the photographing equipment and the high temperature molten iron mold is provided on the high temperature resistant sheet, the cooling structure includes a cooling water reserved hole, and the exhaust structure includes a compressed air reserved hole and an exhaust reserved hole.

[0033] Furthermore, the high-temperature molten iron mold includes at least two matched mold bodies, and the mold bodies are connected to a second lifting mechanism, and the second lifting mechanism is used to drive a plurality of mold bodies to align and match to form a mold groove or to open in a staggered manner.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] (1) Based on the neural network optimization training of the corresponding relationship between aA / N and dD / N, the Si and S content evaluation model of molten iron is obtained. By inputting A / N and D / N to calculate the Si and S content evaluation model, the results of the Si and S content of molten iron can be quickly obtained. Compared with laboratory testing, it is more timely and simplifies the workflow of sampling and observing the composition of molten iron in front of the furnace, so as to provide timely feedback on blast furnace operation.

[0036] (2) Based on the area method, photos of molten iron mushroom samples were collected, and the mushroom-like feature area in the photos was extracted and converted into a binary grayscale image. The CV2 program was used to calculate the number of all pixels N in the mushroom-like feature area. The A / N and D / N were calculated according to the color values ​​of the pixels in the mushroom-like feature area. This improved the reliability and accuracy of intelligent identification of Si pixels and S pixels, and avoided the empirical errors of carbon and "oil skin" formed on the surface of molten iron mushroom samples during the solidification process by observing them with the naked eye, as well as the risk of high-temperature burns caused by manual observation.

[0037] (3) The Si and S content evaluation model of molten iron is based on BP neural network optimization training. The Relu function is used as the activation function. The overall error between the expected value and the estimated value of the output layer neurons is minimized. The activation threshold and connection weight are iteratively optimized to become a storable data structure, which improves the evaluation accuracy and avoids the problem that manual judgment is highly dependent on experience and is prone to inaccurate judgment, leading to misjudgment of furnace conditions and causing production accidents.

[0038] (4) The device uses a camera to take photos of the molten iron in the high-temperature molten iron mold at a suitable position through the detection box. The photos are automatically evaluated by the processor. The cooling structure promotes the rapid demoulding of the molten iron sample by opening the high-temperature molten iron mold after solidification. The exhaust structure eliminates water vapor interference. The high-temperature sample can be tested multiple times in a short time. It is easy to operate, reduces equipment costs compared to the laboratory, and avoids the risk of manual observation.

[0039] In summary, the method and device for evaluating the Si and S content of molten iron based on neural network and area method effectively simplify the operating process and improve the efficiency, accuracy and safety of molten iron composition evaluation, so as to respond to blast furnace operation in a timely and accurate manner and ensure smooth smelting and safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0041] Figure 1 is a flow chart of Embodiment 1 of the present invention;

[0042] Figure 2 is a flow chart of establishing a molten iron Si and S content evaluation model in Example 1 of the present invention;

[0043] Figure 3 is a stereogram of embodiment 2 of the present invention;

[0044] Figure 4 is an initial state diagram of Embodiment 2 of the present invention;

[0045] Figure 5 2 is a diagram showing the change in the demoulding state of Example 2 of the present invention.

[0046] Markings in the figure: frame 100, camera device 200, transparent protective cover 201, high temperature resistant sheet 301, box body 302, transparent isolation sheet 303, high temperature molten iron mold 400, mold body 401, mold groove 402, first lifting mechanism 500, cooling water reserved hole 600, compressed air reserved hole 701, exhaust reserved hole 702, second lifting mechanism 800. DETAILED DESCRIPTION

[0047] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0048] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "axial", "radial", "vertical", "horizontal", "inner", "outer" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first" and "second" are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, "multiple" and "several" mean two or more, unless otherwise clearly and specifically defined.

[0049] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0050] Embodiment 1:

[0051] like Figure 1-2 As shown, a preferred embodiment of the method for evaluating Si and S content in molten iron based on neural network and area method according to the present invention is shown, and the method comprises the following steps:

[0052] S1: Collect multiple photos of molten iron mushroom samples, detect the Si content a and S content d of the molten iron sample corresponding to each photo, and save them in the image information library as standard samples;

[0053] S2: Extracting the mushroom-like feature area in the photo in step S1, specifically including:

[0054] S201: unifying the format and size of each photo based on a bilinear interpolation method;

[0055] S202: Based on importance and other indicators, since the human eye is most sensitive to green and least sensitive to blue, a more reasonable grayscale image can be obtained by weighted averaging the RGB three components of the color image of the photo in step S201 according to the following formula (1);

[0056] f(i,j)=0.30R(i,j)+0.59G(i,j)+0.11B(i,j) (1)

[0057] In the above formula (1), R represents the red channel of the image; G represents the green channel of the image; B represents the blue channel of the image;

[0058] S203: According to the maximum inter-class variance method, namely the OTUS method, also known as the Otsu method, the optimal threshold of the entire photo in step 202 is obtained, and the threshold processing is performed after the grayscale processing is extracted as follows:

[0059] Assuming the grayscale level of the grayscale image is L, the grayscale range is [0, L-1]. The optimal threshold for calculating the photo using the OTSU algorithm is:

[0060] T=Max[w0(t)*(u0(t)-u) 2 +w1(t)*(u1(t)-u) 2 )] (2)

[0061] In the above formula (2), when the segmentation threshold is T, w0 is the background ratio, u0 is the background mean, w1 is the foreground ratio, u1 is the foreground mean, u is the mean of the entire image, 0<t<L-1, and T that maximizes the above formula (2) is the optimal threshold for segmenting the photo;

[0062] S204: using T in step S203, the data of the image in step 202 is divided into two parts: a pixel group greater than T and a pixel group less than T; setting the pixel values ​​of the pixel group greater than T in the image in step 202 to white or black, and setting the pixel values ​​of the pixel group less than T in the image in step 202 to black or white, thereby obtaining a binary grayscale image of the molten iron mushroom-like photo, and presenting an obvious black and white effect on the entire image;

[0063] S3: Calculate the number of all pixels N in the mushroom-like feature area in step S2 using the CV2 program:

[0064] img = cv2.imread(r`Feature Photo`, 1)

[0065] w,h,channels=img.shape

[0066] N = img.size

[0067] print(f`Total pixels: {N})

[0068] S4: Calculate the color value, i.e., RGB value, of the pixel point in the mushroom-like feature area, specifically including:

[0069] S401: Calculate the color values ​​of all pixels N in the feature area in step S3 according to the following formula (3);

[0070] RGB value = (red) + (green*256) + (blue*256*256) (3)

[0071] S402: The RGB value range of the Si content needs to be assigned according to the actual situation. The Si pixel point A is calculated according to the RGB value range of the Si content to obtain the A / N corresponding to the Si content a in step S1. A / N represents the ratio of the Si pixel point A to all the pixel points N in the entire feature area and is saved in the image information library.

[0072] S403: The RGB value range of the S content needs to be assigned according to the actual situation, and the S pixel point D is calculated according to the RGB value range of the S content to obtain the D / N corresponding to the S content d in step S1. D / N represents the ratio of the S pixel point D to all the pixel points N in the entire feature area, and is saved in the image information library;

[0073] S5: Optimize and train the corresponding relationship between aA / N and dD / N based on BP neural network, including:

[0074] S501: Determine the BP neural network structure as input layer, hidden layer, and output layer:

[0075] Input layer: A / N of step S402 and D / N of step 403; Output layer: Si content a and S content d corresponding to A / N and D / N of step S1, to establish training data;

[0076] S502: Define the input value of the i-th neuron in the input layer as X i , the output value of the i-th neuron in the hidden layer is Y i , the output value of the i-th neuron in the output layer is Z i ; The network connection weight between the input layer neurons and the hidden layer neurons is M ij , the network connection weight between the hidden layer neurons and the output layer neurons is Nj ; The activation threshold of the jth unit in the hidden layer is α j , the activation threshold of the kth unit in the output layer is β k ;

[0077] The output components of the hidden layer neurons are:

[0078] In the above formula (4): g1 is the transfer function of the hidden layer neurons, m is the number of neurons in the input layer;

[0079] The output components of the output layer neurons are:

[0080] In the above formula (5): g2 is the transfer function of the output layer neurons, p is the number of hidden layer neurons;

[0081] Use the Relu function as the activation function for training the BP neural network:

[0082]

[0083]

[0084]

[0085] In the above formulas (6), (7) and (8): Y i is the neuron data of each layer, M i is the weight between neurons, b i is the bias that matches the weight;

[0086] The Relu function running on the neurons of the BP neural network is responsible for mapping the input of the neuron to the output, obtaining the hidden layer neurons, and establishing the BP neural network structure;

[0087] S503: According to the direction of the error gradient descent, the connection weight correction formula between the hidden layer neurons and the output layer neurons is as follows (9):

[0088] N j (n+1)=ΔN j (n)+N j (n) (9)

[0089] In the above formula (9): N j (n+1) is the connection weight between the hidden layer neurons and the output layer neurons after the (n+1)th modification, N j (n) is the connection weight between the hidden layer neurons and the output layer neurons after the nth modification, ΔN j (n) is the nth weight modification of the hidden layer neurons and the output layer neurons, ΔN jThe calculation formula of (n) is as follows (10):

[0090]

[0091] In the above formula (10), η is the learning rate of the BP neural network, η∈(0,1), E(n) is the error of the forward transmission of the BP neural network after the nth training;

[0092] The connection weight correction formula between the input layer neurons and the hidden layer neurons is as follows (11):

[0093] M ij (n+1)=ΔM ij (n)+M ij (n) (11)

[0094] In the above formula (11): M ij (n+1) is the connection weight between the input layer neurons and the hidden layer neurons after the (n+1)th modification, M ij (n) is the connection weight between the input layer neurons and the hidden layer neurons after the nth revision, ΔM ij (n) is the nth weight correction of the input layer neurons and hidden layer neurons, ΔM ij The calculation formula of (n) is as follows (12):

[0095]

[0096] In the above formula (12): η is the learning rate of the BP neural network, η∈(0,1), E(n) is the error of the forward transmission of the BP neural network after the nth training;

[0097] S504: Input the A / N of step S402 and the D / N of step 403 into the BP neural network structure of step S502 to calculate the output layer estimated value Z l , after the BP neural network passes the training data forward, it obtains an overall error as the loss function:

[0098]

[0099] In the above formula (13), q is the number of neurons in the output layer; t l is the expected value of the output layer neuron, i.e., the measured Si content a and S content d corresponding to A / N and D / N;

[0100] Taking the overall error minimization as the goal, the activation threshold of S502 is iteratively optimized, and the connection weight of step S503 is corrected to obtain the corresponding relationship of aA / N and dD / N of the optimized training BP neural network, and the parameters are saved to establish the molten iron Si and S content evaluation model;

[0101] In the forward propagation of BP neural network, the signal is calculated by linear superposition before being transmitted from one neuron to the next layer of neurons. Before entering the reverse propagation, it needs to be transformed by nonlinear activation function and loss function before it can continue to be transmitted downward. This cycle is repeated. Based on the BP neural network optimization training of the corresponding relationship of aA / N and dD / N, the evaluation accuracy of the molten iron Si and S content evaluation model is improved.

[0102] S6: Collect photos of the molten iron mushroom samples to be evaluated;

[0103] S7: extracting the mushroom-like characteristic region in the photo of step S6 according to the method of step S2;

[0104] S8: Calculate the number of all pixels N in the mushroom-like feature area in step S7 according to the method in step S3;

[0105] S9: Calculate the color values ​​of all the pixels N in the feature area in step S8 according to the method of step S4, calculate the number of pixels A in the mushroom-like feature area whose RGB values ​​are in the range of Si, and obtain the ratio A / N of Si pixels A to all the pixels N in the entire feature area; calculate the number of pixels D in the mushroom-like feature area whose RGB values ​​are in the range of S, and obtain the ratio D / N of S pixels D to all the pixels N in the entire feature area;

[0106] S10: Input the A / N and D / N of step S9 as input values ​​into the molten iron Si and S content evaluation model of step S5, calculate and output the Si content a and S content d, obtain the evaluation results and save them in the image information library.

[0107] The above method optimizes the training of the corresponding relationship of aA / N and dD / N based on a neural network to obtain an evaluation model for the Si and S contents in molten iron. The Si and S content evaluation model is calculated by inputting A / N and D / N. The actual silicon and sulfur contents of the molten iron mushroom sample are evaluated based on the a and d values ​​of the standard sample. The results of the Si and S contents in the molten iron can be quickly obtained. Compared with laboratory testing, it is more timely, solves the defects of low efficiency and delay in the laboratory, and simplifies the workflow of sampling and observing the composition of molten iron in front of the furnace, so as to provide timely feedback on blast furnace operation.

[0108] The above method is based on the area method, collects photos of molten iron mushroom samples, extracts mushroom-like feature areas in the photos, unifies the photos based on bilinear interpolation, uses three-component weighted average to obtain a reasonable grayscale image, and obtains a binary grayscale image according to the maximum inter-class variance method. The CV2 program is used to calculate the number of all pixels N in the mushroom-like feature area, and A / N and D / N are calculated according to the color values ​​of the pixels in the mushroom-like feature area. The reliability and accuracy of intelligent identification of Si pixels and S pixels are improved, and the empirical error of carbon and "oil skin" formed on the surface of molten iron mushroom samples during the solidification process is avoided by naked eye observation, which solves the problem of high-temperature burns caused by manual observation.

[0109] The above method is based on the molten iron Si and S content evaluation model optimized and trained by the BP neural network, uses the Relu function as the activation function, and optimizes the training with the goal of minimizing the overall error between the expected value and the estimated value of the output layer neurons. According to the direction of the error gradient descent, the activation threshold and the connection weight are iteratively optimized to become a storable data structure, thereby improving the evaluation accuracy and solving the problem that when using the long-term accumulated experience of the staff to make judgments, the staff is highly dependent on the experience and is prone to inaccurate judgments, which leads to misjudgment of the furnace condition and causes production accidents.

[0110] Embodiment 2:

[0111] like Figure 3-5 As shown, a preferred embodiment of the device for evaluating Si and S content in molten iron according to the present invention is shown, the device includes a processor and a frame 100, the frame 100 is provided with a photographing device 200 and an openable and closable detection box, the detection box is provided with a cooling structure and an exhaust structure, a high-temperature molten iron mold 400 is provided in the detection box, the photographing device 200 can take pictures of the inside of the high-temperature molten iron mold 400 through the detection box and transmit the pictures to the processor, the processor obtains the evaluation result based on the method for evaluating Si and S content in molten iron based on neural network and area method described in Example 1, and the processing.

[0112] Furthermore, the photographing device 200 uses a high-definition camera, and a transparent protective cover 201 is provided outside the high-definition camera to play a protective role. The processor is set on an external computer.

[0113] Furthermore, the inspection box includes a high temperature resistant sheet 301 and a box body 302, the high temperature resistant sheet 301 is connected to a first lifting mechanism 500, and a transparent isolation sheet 303 corresponding to the photographing device 200 and the high temperature molten iron mold 400 is provided on the high temperature resistant sheet 301, the cooling structure includes a cooling water reserved hole 600, and the exhaust structure includes a compressed air reserved hole 701 and an exhaust reserved hole 702.

[0114] Furthermore, the cooling water reserved hole 600 is arranged on the high temperature resistant sheet 301. The cooling water reserved hole 600 is externally connected to a water tank pipeline and a nozzle, and is used to pass cooling water into the high temperature molten iron mold 400 to cool the molten iron sample. The cooling position can be adjusted by adjusting the nozzle angle.

[0115] Furthermore, the compressed air reserved hole 701 is arranged on the high temperature resistant sheet 301, and the compressed air reserved hole 701 is externally connected to a compressed air pipeline for purging the inside of the detection box. The exhaust reserved hole 702 is arranged on the box body 302, and the exhaust reserved hole 702 is externally connected to an exhaust device for extracting water vapor in the box body 302.

[0116] Furthermore, the high-temperature molten iron mold 400 includes at least two matched mold bodies 401, and the mold bodies 401 are connected to a second lifting mechanism 800, and the second lifting mechanism 800 is used to drive a plurality of mold bodies 401 to align and match to form a mold groove 402 or to open in a staggered manner.

[0117] Furthermore, there are two mold bodies 401 and they are symmetrically arranged. A semicircular mold groove 402 is provided on the mold body 401. When the two mold bodies 401 are closed, a full-circular mold groove 402 with an open top is formed.

[0118] Furthermore, the first lifting mechanism 500 and the second lifting mechanism 800 use hydraulic cylinders, which are installed on the frame 100. The hydraulic rods of the hydraulic cylinders are connected to the high temperature resistant sheet 301 or the mold body 401. The hydraulic cylinders are connected to a controller to control the extension and retraction of the hydraulic rods to achieve lifting operations.

[0119] The working steps of the above-mentioned device for evaluating Si and S content in molten iron are as follows:

[0120] F1: In the initial state, the first lifting mechanism 500 drives the high temperature resistant sheet 301 to rise so that the detection box is opened, and the second lifting mechanism 800 drives the corresponding mold body 401 to rise and fall so that the two mold bodies 401 are aligned and matched to form a mold groove 402;

[0121] F2: During the ironmaking process of the blast furnace, the high-temperature molten iron in front of the furnace is sampled, and the second lifting mechanism 800 drives the high-temperature molten iron mold 400 to rise as a whole, so as to facilitate pouring the high-temperature molten iron sample into the mold groove 402;

[0122] F3: After the first driving mechanism drives the high temperature resistant sheet 301 to descend, and the second driving mechanism drives the high temperature molten iron mold 400 to descend to a suitable position for photographing, the high-definition camera takes a photo of the molten iron sample in the high temperature molten iron mold 400. At the same time, about 7-10 seconds after the high temperature molten iron is poured into the high temperature molten iron mold 400 is the best time for photographing, that is, the above step S6 collects photos of the molten iron mushroom sample to be evaluated, and transmits the photos of the molten iron mushroom sample to the processor.

[0123] F4: The processor applies the steps of Embodiment 1 and processes to obtain an evaluation result;

[0124] F5: The first driving mechanism drives the high temperature resistant sheet 301 to descend until the high temperature resistant sheet 301 and the box body 302 are closed, and the second driving mechanism drives the high temperature molten iron mold 400 to descend to a suitable position in the detection box;

[0125] F6: External compressed air starts to purge the inside of the box 302 from the compressed air reserved hole 701;

[0126] F7: The external water tank sprays water from the cooling water reserved hole 600 to cool the molten iron sample, so that the molten iron sample solidifies in the high-temperature molten iron mold 400;

[0127] F8: The external exhaust device starts to suck the water vapor in the test box from the exhaust reserved hole 702;

[0128] F9: After about 10 seconds, the left mold body 401 of the high-temperature molten iron mold 400 is driven to rise by the corresponding second driving mechanism, and the right mold body 401 of the high-temperature molten iron mold 400 is driven to lower by the corresponding second driving mechanism, and the high-temperature molten iron mold 400 is opened by dislocation, so that the solidified molten iron sample can be demoulded quickly;

[0129] F10: After the molten iron sample is demoulded, the high-temperature molten iron mold 400 may have a small amount of water stains. Since the high-temperature molten iron mold 400 still has a certain temperature after demoulding, the water stains carried by the high-temperature molten iron mold 400 will evaporate completely in a short time under the blowing of compressed air;

[0130] F11: Repeat steps F1-F10 to perform the next task.

[0131] The above-mentioned device drives the inspection box to open and close by the first lifting mechanism 500, and adopts the photographing device 200 to take the photo of the high-temperature molten iron in the high-temperature molten iron mold 400 at the appropriate position through the transparent isolation plate 303 of the inspection box. The photo is automatically evaluated by the processor. After the cooling structure promotes the solidification of the molten iron sample, the second lifting mechanism 800 drives the high-temperature molten iron mold 400 to open and demold quickly. The exhaust structure eliminates the interference of water vapor, and the high-temperature sample can be tested multiple times in a short time, which solves the defects of laboratory testing workflow and cumbersome operation, high equipment cost, and the risk and error of manual observation.

[0132] In summary, the method and device for evaluating the Si and S content of molten iron based on neural network and area method effectively simplify the operating process and improve the efficiency, accuracy and safety of molten iron composition evaluation, so as to respond to blast furnace operation in a timely and accurate manner and ensure smooth smelting and safety.

[0133] The series of detailed descriptions listed above are only specific descriptions of feasible embodiments of the present invention. They are not intended to limit the scope of protection of the present invention. All equivalent embodiments or changes that do not deviate from the technical spirit of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for evaluating Si and S content in molten iron based on neural network and area method, characterized in that: The methods include: Collect photos of molten iron mushrooms, extract mushroom-like feature areas in the photos, calculate the number of all pixel points N in the mushroom-like feature areas, and calculate the ratio A / N of Si pixel points A to all pixel points N and the ratio D / N of S pixel points D to all pixel points N according to the color values ​​of the pixels in the mushroom-like feature areas; Detect the Si content a and S content d of the molten iron samples corresponding to multiple photos, optimize and train the corresponding relationship of aA / N and dD / N based on the neural network, and obtain the molten iron Si and S content evaluation model; Input A / N and D / N to calculate the molten iron Si and S content evaluation model, output Si content a and S content d, and obtain the evaluation result; The corresponding relationship of aA / N and dD / N is optimized and trained based on the BP neural network, specifically including: taking A / N and D / N as input layers, and taking Si content a and S content d corresponding to A / N and D / N as output layers, to establish training data; taking Relu function as an activation function for training the BP neural network, to establish a BP neural network structure; According to the direction of the error gradient descent, the connection weights between the hidden layer neurons and the output layer neurons, and the connection weights between the input layer neurons and the hidden layer neurons are corrected; the training is optimized with the goal of minimizing the overall error between the expected value and the estimated value of the output layer neurons.

2. The method for evaluating Si and S content in molten iron based on neural network and area method according to claim 1, characterized in that: Extracting mushroom-like feature areas in photos includes: The format and size of each photo are unified based on the bilinear interpolation method; Perform weighted averaging on the RGB components of the color image to obtain a reasonable grayscale image; Calculate the optimal threshold of the photo according to the maximum inter-class variance method; The pixel values ​​of the pixel group that are greater than or less than the optimal threshold are set to black and white to obtain a binary grayscale image.

3. The method for evaluating Si and S content in molten iron based on neural network and area method according to claim 2, characterized in that: The RGB three components of the color image are weighted averaged according to the formula f(i,j)=0.30R(i,j)+0.59G(i,j)+0.11B(i,j), where R represents the red channel of the image; G represents the green channel of the image; B represents the blue channel of the image; According to the formula T = Max[w0(t)*(u0(t)-u) 2 +w1(t)*(u1(t)-u) 2 ] Calculate the optimal threshold of the photo, where w0 is the background ratio, u0 is the background mean, w1 is the foreground ratio, u1 is the foreground mean, and u is the mean of the entire image.

4. The method for evaluating Si and S content in molten iron based on neural network and area method according to claim 1, characterized in that: The RGB value of all pixels N in the mushroom-like feature area = (red) + (green*256) + (blue*256*256). The Si pixel A is calculated according to the RGB value range of the Si content, and the S pixel D is calculated according to the RGB value range of the S content.

5. A device for evaluating Si and S content in molten iron, characterized in that: The invention comprises a processor and a frame (100), wherein the frame (100) is provided with a photographing device (200) and an openable and closable detection box, wherein the detection box is provided with a cooling structure and an exhaust structure, wherein a high-temperature molten iron mold (400) is arranged in the detection box, wherein the photographing device (200) can take pictures of the inside of the high-temperature molten iron mold (400) through the detection box and transmit the pictures to the processor, wherein the processor processes the method for evaluating Si and S content in molten iron based on a neural network and an area method as described in any one of claims 1 to 4 to obtain an evaluation result.

6. The device for evaluating Si and S content in molten iron according to claim 5, characterized in that: The detection box comprises a high temperature resistant sheet (301) and a box body (302); the high temperature resistant sheet (301) is connected to a first lifting mechanism (500); a transparent isolation sheet (303) corresponding to a photographing device (200) and a high temperature molten iron mold (400) is provided on the high temperature resistant sheet (301); the cooling structure comprises a cooling water reserved hole (600); and the exhaust structure comprises a compressed air reserved hole (701) and an exhaust reserved hole (702).

7. The device for evaluating Si and S content in molten iron according to claim 5, characterized in that: The high-temperature molten iron mold (400) comprises at least two matched mold bodies (401), wherein the mold bodies (401) are connected to a second lifting mechanism (800), and the second lifting mechanism (800) is used to drive a plurality of mold bodies (401) to align and match to form a mold groove (402) or to open in a staggered manner.

Images