Method for judging end point of copper smelting reduction period of anode furnace

Abstract

The invention discloses an end point judgment method for a copper smelting reduction period of an anode furnace, and belongs to the field of metallurgy. The method comprises the following steps: continuously acquiring flame image information by using a camera, obtaining refined data by using a flame image, converting a preprocessed image into an HSV color space, respectively extracting chromaticity, saturation and brightness values of each pixel point, extracting image feature information, and training by using a generalized regression neural network, namely a GRNN neural network structure data set. And inputting the data into an intelligent control system. The PID control algorithm is used for monitoring and controlling parameter optimization, loop control is achieved, and forecasting and alarming are completed at the end point of the reduction period. The method can effectively improve the end point hit rate of the anode furnace in the copper smelting reduction period, so that the production efficiency is improved.

Description

Technical Field

[0001] The present invention belongs to the field of metallurgy, and particularly relates to an end-point judgment system for the reduction period of copper smelting in an anode furnace using digital image processing technology and a neural network intelligent control system. Background Art

[0002] An anode furnace, also known as a rotary refining furnace, is suitable for refining molten blister copper. It has the characteristics of simple structure, high mechanization, and high automation, and can process blister copper with low impurity content through program control. Due to the above advantages, the anode furnace has become a standard pyrometallurgical refining equipment in copper smelters. The two key operation stages of pyrometallurgical anode refining of blister copper are the oxidation stage and the reduction stage. In actual production, the end-point judgment of the oxidation stage of the anode furnace is mainly based on the SO2 content in the flue gas. Since the SO2 content in the flue gas during the oxidation period of copper smelting in the anode furnace is relatively high, operators can rely on a sulfur dioxide analyzer to collect and record SO2 content data relatively accurately, and judge the end-point of the oxidation period through the data. However, when the copper smelting in the anode furnace reaches the reduction period, the SO2 content in the flue gas is low, and operators can no longer effectively judge the end-point of the reduction period based on the SO2 content. Workers will adopt an over-reduction operation method to avoid producing unqualified copper products, which will lead to a decline in product quality. To improve the production efficiency of the anode furnace, it is of great significance to accurately judge the end-point of the reduction stage for actual production.

[0003] With the continuous development of automation control technology, the end-point judgment based on image analysis during the reduction period becomes particularly important. By real-time monitoring and analyzing key parameters during the smelting process, such as temperature, furnace gas composition, etc., the end-point of the reduction period of copper smelting can be accurately judged to ensure that the smelting process reaches the best effect. This system can avoid human errors in traditional methods, improve the accuracy of end-point judgment, and provide a reliable basis for subsequent automation control.

[0004] Accurate end-point judgment during the reduction period not only helps to improve production efficiency, but also can effectively prevent situations such as overheating and over-smelting that may cause equipment damage or safety accidents, thus enhancing the safety of the production process. Summary of the Invention

[0005] To solve the above technical problems, the present invention provides a method for judging the end-point of the reduction period of copper smelting in an anode furnace, which is implemented by the following steps:

[0006] Step 1: Use a camera and a module collector to obtain flame image information;

[0007] Step 2: Separate the obtained flame image into three channels of HSV;

[0008] Step 3: Quantify the image numerically;

[0009] Step 4: Establish a multi-task model for the end point of the reduction period and train it;

[0010] Step 5: Use the prediction result of the neural network to judge the stage where the current real-time flame image is located and complete the alarm at the end point.

[0011] Further, the obtaining of the flame image information in step 1 is to obtain the furnace mouth image by using a high-speed camera at the anode furnace mouth, and use a module collector to collect the sulfur dioxide concentration, oxygen concentration, flue gas outlet temperature, air supply pressure, air supply flow rate, and oxygen flow rate in the anode furnace.

[0012] Further, the HSV three-channel separation in step 2 is to perform HSV three-channel separation on the real-time flame image, and set it as the chrominance channel, saturation channel, and lightness channel signals respectively.

[0013] Further, the establishment of the multi-task model for the end point of the reduction period in step 4 is to use the color moment ζi as the input eigenvalue of the generalized regression neural network (GRNN), and select n reduction-period flame images as the training set, and the response values are set as the initial reduction stage I, the middle reduction stage II, and the end reduction stage III respectively; then, select m reduction-period flame images as the test set to verify the stage where the flame image is located.

[0014] Through the prediction result of the neural network, the present invention can judge the reduction stage of the current flame image, and can also assist in verifying the judged stage.

[0015] The present invention has the following beneficial effects:

[0016] The present invention can avoid the errors existing in the manual judgment of the end point of the slag-making period in the copper converter blowing process, effectively improve the hit rate of the reduction refining end point of the copper anode furnace, thereby improving production efficiency and reducing production costs. Moreover, the process is simple, easy to operate, and easy to promote. The method of using the flame characteristic color moment, sulfur dioxide concentration, oxygen concentration, flue gas outlet temperature, air supply pressure, air supply flow rate, and oxygen flow rate as the input layer parameters of the neural network to study the end point of the reduction period in copper converter smelting can be an important alternative means for practical engineering applications. It not only has good accuracy but also can reduce the cost expenditure in the research and experimental direction, and has a very broad application prospect. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings that form a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation to this application. In the drawings:

[0018] Figure 1 It is a flow chart of the method of the present invention;

[0019] Figure 2To restore the HSV three-channel component information of the flame image throughout the reduction period;

[0020] Figure 3 To calculate the color moments of the flame image throughout the reduction period;

[0021] Figure 4 For the generalized regression neural network structure diagram (GRNN);

[0022] Figure 5 For the radial basis neural network model. Specific implementation mode

[0023] It should be noted that, without conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The following will refer to the drawings and combine the embodiments to detail this application.

[0024] It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0025] Embodiment 1

[0026] Step 1: Use a high-speed camera and a module collector to obtain flame image information; that is, obtain the furnace mouth image by using a high-speed camera at the anode furnace mouth, and use the module collector to collect the sulfur dioxide concentration, oxygen concentration, flue gas outlet temperature, air supply pressure, air supply flow rate, and oxygen flow rate inside the anode furnace.

[0027] Step 2: Separate the obtained flame image into three HSV channels; that is, separate the real-time flame image into three HSV channels, and set them as the chrominance channel, saturation channel, and lightness channel signals respectively;

[0028] Step 3: Numerically quantize the image;

[0029] Step 4: Establish a multi-task model for the end point of the reduction period and train it; that is, to establish a multi-task model for the end point of the reduction period, take the color moment ζi as the input feature value of the generalized regression neural network (GRNN), and select n reduction period flame images as the training set, and set the response values as the initial reduction stage I, the middle reduction stage II, and the end reduction stage III respectively; then, select m reduction period flame images as the test set to verify the stage where the flame image is located;

[0030] Step 5: Use the prediction results of the neural network to judge the stage where the current real-time flame image is located and complete the alarm at the end point.

[0031] Through the prediction results of the neural network, the present invention can determine the reduction stage of the current flame image and can also assist in verifying the determined stage.

[0032] Starting from the flame characteristics in the reduction period, the present invention separates the performances of each channel of the flame. During the process of making copper in the anode furnace, due to the influence of chemical elements in the furnace, molten bath temperature and other factors, the chromaticity, saturation and brightness of the flame will change. Therefore, by extracting relevant flame characteristic information, analyzing the color moment characteristics of the flame HSV image, and combining the generalized regression neural network (GRNN) to optimize and classify these characteristic values, the prediction and judgment of the end point of the slag-making period in converter copper smelting can be realized.

[0033] The present invention utilizes the color information of the chromaticity, saturation and brightness channels in the HSV color space of the flame image, calculates the correlation coefficient ξ(μ,v), and analyzes the color characteristics of the flame image according to this coefficient, so as to judge the stage of the reduction process of copper smelting in the anode furnace and accurately predict the end point of the reduction period.

[0034] Initial stage: When the value of ξ(μ,v) is close to 0.5 and remains stable, it indicates that the current flame characteristics conform to the initial state and are in the initial stage of reduction.

[0035] Middle stage: When the value of ξ(μ,v) is close to 1 and shows a stable peak, it indicates that the current flame characteristics reach the middle state of reduction.

[0036] End stage: When 0.5 < ξ(μ,v) ≤ 0.8 and this value remains stable, it indicates that the reduction process is about to end, the flame color characteristics are close to the ideal state, and it is at the end point of the reduction period.

[0037] Color moment ξ i Comprehensively represents the characteristic information of the three channels of image chromaticity, saturation and brightness.

[0038] Color moment formula: First moment (mean value)

[0039] Second moment (standard deviation)

[0040] Third moment (skewness)

[0041] Judging the end point of the reduction period based on the flame color characteristics has significant advantages. First of all, this method can reflect parameters such as in-furnace chemical reactions and temperature changes in real time, providing a fast and efficient monitoring means. Since it is a non-contact monitoring, it avoids equipment loss and danger in traditional methods. In addition, the changes in flame chromaticity, saturation and brightness are closely related to various factors in the furnace. By accurately extracting image features, high-precision and high-sensitivity judgment can be achieved.

[0042] The working principle of GRNN is relatively simple and usually consists of four layers: the input layer, the pattern layer, the summation layer, and the output layer. Its training process does not require the traditional backpropagation algorithm but directly estimates based on the training data. GRNN can be designed through radial basis neurons and linear neurons.

[0043] Radial basis function neural network (RBFNN) shows significant advantages in multiple application scenarios, especially when dealing with complex nonlinear problems. Its training speed is relatively fast, and the optimization process is simple, which benefits from the simplicity of the network structure and fewer parameters, thus effectively reducing the computational burden. RBFNN has a powerful function approximation ability and can accurately handle complex nonlinear relationships to ensure high-precision data fitting. In addition, the number of neurons in the hidden layer of RBFNN is relatively small, the structure is simple, and it is easy to understand and implement.

[0044] RBFNN has strong generalization ability during the training process and can effectively adapt to different types of data distributions through nonlinear mapping, reducing the risk of overfitting. It has strong adaptability and can adapt to the changes and distributions of different data by adjusting the centers and widths of the basis functions. This network can effectively handle noise and abnormal data to ensure the stability and accuracy of the model.

[0045] In addition, due to the independent calculations of neurons in RBFNN, it has good parallel computing ability and can efficiently process large-scale data sets to meet the requirements of high-concurrency and high-performance data processing. Therefore, RBFNN demonstrates strong technical advantages and broad application prospects in big data processing, complex pattern recognition, and other applications that require efficient handling of nonlinear relationships.

[0046] Collect and organize the training data related to the problem. The data set usually includes input features (vectors) and corresponding output labels.

[0047] The number of neurons in the input layer is the same as the number of features of the input data.

[0048] The radial basis function usually uses the Gaussian function, and other types of radial basis functions can also be selected according to the problem. The form of the Gaussian function is usually:

[0049]

[0050] where c is the center vector and σ is the width parameter.

[0051] Once the centers and widths of the radial basis functions are determined, the weights of the output layer of the network can be solved by the least squares method. This is usually a linear regression problem, that is, to solve:

[0052] W=Φ -1 T

[0053] Among them, Φ is a matrix composed of the outputs of the hidden layer (basis function values), and T is the expected output value.

[0054] Assuming the input is x and the output is y, the output of the network can be expressed as:

[0055]

[0056] Among them, W i is the weight of the output layer, φ i is the output of the i-th radial basis function, and N is the number of neurons in the hidden layer.

[0057] The basic principle, main features and advantages of the present invention have been detailedly demonstrated and described in the above content. Those skilled in the art should understand that the present invention is not limited to the specific forms of the above embodiments. The descriptions in the embodiments and the specification are only the preferred implementation schemes of the present invention and do not constitute a limitation on the scope of the present invention. Without departing from the basic idea and technical scope of the present invention, the present invention can be variously changed and improved, and these changes and improvements are all within the protection scope of the present invention. The protection scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for judging the end point of the reduction period in copper smelting in an anode furnace, characterized in that: The method is implemented by the following steps: Step 1: Obtain the flame image information by using a camera and a module collector; Step 2: Separate the obtained flame image into three HSV channels; Step 3: Numerically quantize the image; Step 4: Establish a multi-task model for the end point of the reduction period and train it; Step 5: Use the prediction result of the neural network to judge the stage of the current real-time flame image and complete the alarm at the end point.

2. The method for judging the end point of the reduction period in copper smelting in the anode furnace according to claim 1, wherein: In Step 1, the flame image information is obtained by using a high-speed camera to take the image at the anode furnace mouth and using a module collector to collect the sulfur dioxide concentration, oxygen concentration, flue gas outlet temperature, air supply pressure, air supply flow rate, and oxygen flow rate inside the anode furnace.

3. The method for judging the end point of the reduction period in copper smelting in the anode furnace according to claim 1, wherein: In Step 2, the HSV three-channel separation is to separate the real-time flame image into three HSV channels, which are respectively set as the chrominance channel, saturation channel, and lightness channel signals.

4. The method for judging the end point of the reduction period in copper smelting in the anode furnace according to claim 1, wherein: In Step 4, to establish the multi-task model for the end point of the reduction period, the color moment ζi is used as the input eigenvalue of the generalized regression neural network, and n reduction-period flame images are selected as the training set, and the response values are respectively set as the initial reduction stage I, the middle reduction stage II, and the end reduction stage III; then, m reduction-period flame images are selected as the test set to verify the stage where the flame image is located.

Images