Soft measurement method for sintering end point based on image recognition

Abstract

The invention discloses a sintering end point soft measurement method based on image recognition, belongs to the field of calculation of an end point position in the sintering process in the metallurgical industry, and aims at clustering and dividing different sintering working condition types based on a distribution diagram of the current air bellow exhaust gas temperature and determining the position of the sintering end point by adopting a specific rule for each type. The method comprises the following implementation processes: collecting the temperature of the sintering flue waste gas; establishing an air bellow exhaust gas temperature distribution curve image by adopting an interpolation mode; establishing an image clustering network to group the bellows exhaust gas temperature distribution diagram; formulating different sintering end point position calculation rules for different groups of images; and an image classification network is established to classify the input bellow exhaust gas temperature map, and the sintering end point position is calculated by adopting the rule of the corresponding category according to the classification result.

Description

Technical Field

[0001] The present invention relates to the technical field of calculating an endpoint position in a sintering process in the metallurgical industry, and in particular to a sintering endpoint soft measurement method based on image recognition. Background Art

[0002] The location of the burn-through point (BTP) is a key parameter reflecting the sintering state. Achieving a stable BTP during sintering is crucial for increasing production, improving quality, and reducing energy consumption. Furthermore, soft sensing technology can replace some hardware sensors, addressing issues such as high cost and difficult maintenance.

[0003] The main technical means for soft measurement of the sintering endpoint position is to establish a mathematical model, such as fitting a quadratic curve using the exhaust gas temperature, and confirming the vertex, i.e., the sintering endpoint position, based on the fitted curve; or to establish a computational model based on artificial intelligence algorithms such as support vector machines, neural networks, and other machine learning methods using various production parameters and the sintering endpoint position to establish a mapping relationship. However, the sintering process is a complex physical and chemical reaction process. In actual production, depending on different working conditions, the data it relies on, such as exhaust gas temperature, will show completely different distributions, making it difficult to accurately simulate using mathematical models; computational models such as support vector machines, BP neural networks, and RBF neural networks that rely on state control parameters such as temperature field, permeability, and trolley speed have poor versatility due to differences in selected parameters, measurement accuracy, and hysteresis, and it is difficult to explain the relationship between dependent parameters and the endpoint position. Summary of the Invention

[0004] The embodiment of the present invention provides a new soft measurement method for sintering endpoint based on image recognition, aiming to further improve the accuracy and interpretability of existing soft measurement of sintering endpoint, and solve the problems of existing methods' reliance on the measurement of various production and state parameters during the sintering process, and the low versatility of calculation methods under different working conditions.

[0005] The present invention provides a sintering endpoint soft measurement method based on image recognition, comprising: Preprocess the data of one or more waste gas temperature measurement points of the same wind box, and interpolate the preprocessed wind box waste gas temperature by interpolation method; For the interpolated temperature data, the coordinate axis of the windbox position-temperature is established, the temperature distribution curve is drawn, and the grayscale image of the windbox exhaust gas temperature distribution is obtained; The feature vector of the grayscale image is extracted through a deep convolutional neural network, and the feature vector of the grayscale image is clustered using a clustering algorithm; For each clustering result, corresponding rules are formulated according to the characteristics of the temperature distribution image to calculate the sintering endpoint position.

[0006] In some examples, the interpolated temperature data, establishing a wind box position-temperature coordinate axis, drawing a temperature distribution curve, and obtaining a grayscale image of the wind box exhaust gas temperature distribution include: For the interpolated temperature data, the coordinate axis of the windbox position-temperature is established, the temperature distribution curve is drawn, and the space between the x-axis and the curve is filled. The curve graph is converted into an area graph to obtain a grayscale image of the windbox exhaust gas temperature distribution.

[0007] In some examples, extracting a feature vector of a grayscale image using a deep convolutional neural network includes: A deep convolutional neural network consists of multiple blocks, each of which contains several convolutional layers and a pooling layer; The grayscale image is output as a feature vector after multiple convolutions and pooling.

[0008] In some examples, for each clustering result, a corresponding rule is formulated according to the characteristics of the temperature distribution image to calculate the sintering endpoint position, including: If there is an obvious temperature drop point near the highest point, several points near the highest point are selected for quadratic curve fitting, and the sintering end point position is solved according to the fitting curve; If there is no obvious temperature drop point near the highest point, find the temperature drop point from the highest point toward the tail of the machine. If there is a point with a significantly lower temperature than the highest point at a certain position, select several temperature points from the temperature drop point toward the head of the machine for quadratic curve fitting, and solve the curve vertex position as the sintering end point. If the highest point temperature does not reach the set temperature, there is no sintering end point.

[0009] In some examples, the method further comprises: An image classifier is constructed to classify grayscale images. The current bellows temperature is processed into a grayscale image and used as the input of the image classifier to obtain the output probability of each category. The category with the highest probability is selected as the classification result. According to the classification result of the current wind box temperature distribution, the sintering end point position is calculated using the category rule corresponding to the classification result.

[0010] In some examples, constructing an image classifier to classify a grayscale image includes: After clustering, add corresponding category labels to each grayscale image, and use all grayscale images with labeled categories as sample sets; Construct a deep convolutional neural network classifier, which takes as input a grayscale image of temperature distribution, passes through several convolution blocks, and finally outputs the probability value of each classification through an output layer containing several neurons; The grayscale image sample set with category annotations is trained as a deep convolutional neural network classifier.

[0011] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art: This method collects and preprocesses flue gas temperature data, interpolates the windbox exhaust gas temperature using interpolation, creates a grayscale image of the windbox exhaust gas temperature distribution, builds a deep convolutional neural network to extract features from the grayscale image, uses a clustering algorithm to group the images, and establishes different rules for each group to calculate the sintering endpoint position. This method aims to further improve the accuracy and interpretability of existing soft-measurement of sintering endpoint. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0013] Figure 1 It is a schematic diagram of the method flow provided by an embodiment of the present invention; Figure 2 Schematic diagram of the image feature extraction network result based on deep convolution provided by an embodiment of the present invention; Figure 3 This is a schematic diagram of the automatic clustering algorithm provided by an embodiment of the present invention; Figure 4 Schematic diagram of different clustering results of wind box exhaust gas temperature distribution provided by an embodiment of the present invention; Figure 5 This is a schematic diagram of the image classification structure based on deep convolution provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0014] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0015] In the following description, specific embodiments of the present invention will be described with reference to steps and symbols performed by one or more computers, unless otherwise specified. Therefore, these steps and operations will be mentioned several times as being performed by a computer, and computer execution as referred to herein includes operations by a computer processing unit that represents electronic signals of data in a structured form. This operation converts the data or maintains it at a location in the computer's memory system, which can be reconfigured or otherwise change the operation of the computer in a manner familiar to testers in the field. The data structure in which the data is maintained is a physical location in the memory, which has specific characteristics defined by the data format. However, the principles of the present invention are described in the above text, which does not represent a limitation, and testers in the field will understand that the various steps and operations below can also be implemented in hardware.

[0016] As used herein, the terms "module" or "unit" may be considered software objects executed on the computing system. The various components, modules, engines, and services herein may be considered implementation objects on the computing system. While the devices and methods herein are preferably implemented in software, they may also be implemented in hardware and remain within the scope of protection of the present invention.

[0017] It will be understood by those skilled in the art that, unless expressly stated otherwise, the singular forms "a", "an", and "the" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, "connected" or "coupled" as used herein may include wireless connections or wireless couplings. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.

[0018] In an embodiment of the present invention, a sintering endpoint soft measurement method based on image recognition is provided, such as Figure 1 As shown, the following steps are included: Step 1: Collect and preprocess flue gas temperature data; Step 2: Use interpolation to supplement the windbox exhaust gas temperature; Step 3: Create a grayscale image of the windbox exhaust temperature distribution; Step 4: Establish a deep convolutional neural network to extract features from grayscale images; Step 5: Use clustering algorithm to group the images; Step 6: Formulate different rules for each group to calculate the sintering endpoint position; Step 7: Establish a deep convolutional neural network to classify images; Step 8: Take the current bellows temperature as input, create a grayscale image, and determine the category through the classification network; Step 9: Calculate the sintering endpoint using the corresponding rules based on the classification results.

[0019] Furthermore, in step 3, the exhaust gas bellows temperature is collected and interpolated to create a temperature distribution graph, which is then saved as an image. Specifically, a two-dimensional array of bellows position and temperature is created based on the exhaust gas bellows temperature data, and smooth interpolation is performed according to the required accuracy. A temperature distribution curve is then plotted for the interpolated result, which is then converted into an area graph and saved as a grayscale image.

[0020] Furthermore, in step 4, feature extraction is performed on the temperature distribution image. The deep convolutional network for feature extraction is composed of multiple similar blocks, each of which contains several convolutional layers and a pooling layer. After multiple convolutions and pooling, the grayscale image is output as a feature vector.

[0021] Furthermore, in step 5, a clustering algorithm is used to classify the image feature vectors into several groups. The clustering algorithm should not predefine the final number of groups, but rather rely on the connections between the images themselves and the set boundary conditions to determine the groupings. After grouping each image in the training set, the corresponding category labels are added.

[0022] Furthermore, in step 6, according to the division result of the temperature distribution image, different rules corresponding to different distribution categories are established to calculate the sintering end point position.

[0023] Furthermore, in step 7, the input distribution image is classified based on the classifier. Specifically, the different categories of images obtained by clustering are used as sample sets to train a convolutional classification neural network. The current wind box temperature distribution map is used as input, the corresponding classification results are output, and the sintering endpoint position is calculated based on the corresponding rules according to the category.

[0024] Furthermore, in step 1, the flue gas temperature data is collected and preprocessed, and the average value is taken after removing the abnormal values ​​for one or more waste gas temperature measurement points at the same wind box position.

[0025] Furthermore, in step 2, the windbox exhaust gas temperature is supplemented by the cubic spline interpolation method so that the discrete points of temperature collection can generate a smooth curve, as follows: For the entire interpolation interval , if exists Known value, Divided into subintervals, in the interval The upper interpolation function is a polynomial of degree not more than 3, intervals need to be solved Undetermined coefficients:

[0026] exist On an internal node, Continuous and smooth, that is, the first-order derivative and the second-order derivative Existing and continuous, we get Conditions:

[0027]

[0028]

[0029] Based on the known values, we get n+1 conditions:

[0030] Consider endpoints The characteristics at the endpoint are considered to be 0, that is:

[0031]

[0032] common The equation can be solved Complete interpolation for each interval.

[0033] It should be noted that the cubic spline interpolation method used in the embodiment of the present invention to generate the temperature distribution curve is not mandatory, and the use of other algorithms to obtain the temperature distribution curve is also within the scope of protection of the present invention.

[0034] Furthermore, in step 3, after obtaining the interpolated temperature point, the coordinate axis of the bellows position-temperature is established, the temperature distribution curve is drawn, and the area between the x-axis and the curve is filled with black to obtain the bellows temperature area diagram and save it as a grayscale image to simplify the subsequent image processing calculation complexity.

[0035] Again, any design that converts the temperature distribution curve into a picture format for subsequent clustering and classification does not depart from the spirit and scope of the present invention.

[0036] Furthermore, in step 4, the temperature distribution image feature vector is extracted. In the embodiment of the present invention, a deep convolutional network is used as Figure 2, taking a grayscale image as input and extracting the corresponding one-dimensional feature vector. This deep convolutional network consists of several convolutional layers and flattening layers. The convolution layer uses the same, smaller convolution kernel and stride to improve feature representation. Max pooling layers are used between each convolutional layer to reduce the size of the feature map while retaining salient features. All hidden layers use the ReLU function as the activation unit function. The flattening layer preserves the feature order and converts it into a one-dimensional feature vector for subsequent processing.

[0037] It should be noted that in this example, a deep convolutional network is used to extract one-dimensional feature vectors of images for subsequent clustering, but using other methods to extract feature vectors of images still falls within the scope of protection of the present invention.

[0038] Furthermore, in step 5, a clustering algorithm is used to cluster the characteristic vectors of the temperature distribution graph. This automatic clustering algorithm does not require a pre-specified number of clusters. By adjusting other boundary parameters of the clustering algorithm, a limited number of clusters (num_class should be greater than or equal to 2) can be obtained. This is consistent with the various distributions of wind box waste temperatures in actual production conditions.

[0039] In the embodiment of the present invention, a density-based clustering algorithm is used. Figure 3 , we get 3 kinds of (num_class=3) distribution results as follows Figure 4 The clustering process may select other clustering algorithms to obtain image groups, such as spatial or hierarchical clustering algorithms, without departing from the scope of protection of the present invention.

[0040] Furthermore, in step 6, for different grouping results of the temperature distribution image, corresponding rules are formulated according to the characteristics of the temperature distribution image to calculate the sintering end point position.

[0041] In the embodiment of the present invention, Figure 4 The following rules are designed for calculating the sintering endpoint based on the three temperature distribution images: 1) There is a point with obvious temperature drop near the highest point, such as after the highest point appears, compared with the highest point, the temperature drops by a first preset value. In the embodiment of the present invention, a point with a temperature drop of 15 degrees appears after the highest point, such as Figure 4 (a) circumstances; Then take several points near the highest point, generally 4 to 5 points including the highest point, to perform quadratic curve fitting. The exhaust gas temperature Ti and the wind box position xi are approximately in a quadratic relationship (m is the total number of wind boxes):

[0042] Solve the BTP position according to the fitting curve:

[0043] It should be noted that the threshold for determining a significant temperature drop, i.e., the first preset temperature value, should be reasonably set based on empirical experience and the on-site temperature distribution, and the embodiment of the present invention does not impose any unique limitation.

[0044] 2) There is no obvious temperature drop point near the highest point. For example, after the highest point appears, the temperature drop does not exceed the first preset temperature value compared with the highest point. In the embodiment of the present invention, the temperature drop after the highest point does not exceed 15 degrees. Figure 4 (b) circumstances; Find the temperature drop point from the highest point toward the tail. If there is a point with a significantly lower temperature than the highest point at a certain position, , then select several temperature points from the temperature drop point toward the nose direction as Perform quadratic curve fitting and determine the vertex of the curve as the sintering end point.

[0045] 3) If the highest point temperature does not reach the set temperature (e.g. 400°C), Figure 4 (c) circumstances; If the highest point temperature does not reach the set temperature, the sintering machine may be in the start-up or stop state, which is an abnormal operating condition. At this time, the sintering end point is meaningless, that is, there is no sintering end point.

[0046] Furthermore, in step 7, an image classifier is constructed to classify the temperature distribution map, which specifically includes the following steps: 1) Create a dataset. After clustering in step 5, add corresponding category labels to each image. Use all labeled images as the sample set.

[0047] 2) Construct a deep convolutional neural network classifier. In the embodiment of the present invention, a deep convolutional network is used for image classification. Figure 5 The input is the temperature distribution map, which passes through several convolution blocks and finally outputs num_class probability values ​​from the output layer containing num_class neurons. Where num_class is the total number of classes after the image is clustered in step 5.

[0048] 3) Train the neural network classifier. Use the temperature distribution image sample set with category labels as the classifier data source for training. The classifier inputs the temperature distribution image and outputs the corresponding category.

[0049] In particular, the embodiment of the present invention uses a convolutional neural network as a classifier for the temperature distribution map. Classifiers implemented using other machine learning methods or neural networks, using clustered and labeled samples as a sample set for parameter training, should also be included in the scope of protection of the present invention.

[0050] Furthermore, in step 8, the current bellows temperature is processed into a grayscale image through steps 1 to 3 and used as the input of the classification network in step 7 to obtain the probability output of each category. The category with the highest probability is selected as the classification result.

[0051] Furthermore, in step 9, according to the classification result of the current wind box temperature distribution in step 8, the corresponding classification rule in step 6 is adopted to calculate the sintering end point position.

[0052] The above is a detailed introduction to a sintering endpoint soft measurement method based on image recognition provided by an embodiment of the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the ideas of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A sintering endpoint soft measurement method based on image recognition, characterized in that: include: Preprocess the data of one or more waste gas temperature measurement points of the same wind box, and interpolate the preprocessed wind box waste gas temperature by interpolation method; For the interpolated temperature data, the coordinate axis of the windbox position-temperature is established, the temperature distribution curve is drawn, and the grayscale image of the windbox exhaust gas temperature distribution is obtained; The feature vector of the grayscale image is extracted through a deep convolutional neural network, and the feature vector of the grayscale image is clustered using a clustering algorithm; For each clustering result, corresponding rules are formulated according to the characteristics of the temperature distribution image to calculate the sintering endpoint position.

2. The method according to claim 1, characterized in that The interpolated temperature data is used to establish a wind box position-temperature coordinate axis, draw a temperature distribution curve, and obtain a grayscale image of the wind box exhaust gas temperature distribution, including: For the interpolated temperature data, the coordinate axis of the windbox position-temperature is established, the temperature distribution curve is drawn, and the space between the x-axis and the curve is filled. The curve graph is converted into an area graph to obtain a grayscale image of the windbox exhaust gas temperature distribution.

3. The method according to claim 2, characterized in that The method of extracting a feature vector of a grayscale image by using a deep convolutional neural network includes: A deep convolutional neural network consists of multiple blocks, each of which contains several convolutional layers and a pooling layer; The grayscale image is output as a feature vector after multiple convolutions and pooling.

4. The method according to claim 3, characterized in that For each clustering result, a corresponding rule is formulated according to the characteristics of the temperature distribution image to calculate the sintering end point position, including: If there is an obvious temperature drop point near the highest point, several points near the highest point are selected for quadratic curve fitting, and the sintering end point position is solved according to the fitting curve; If there is no obvious temperature drop point near the highest point, find the temperature drop point from the highest point toward the tail of the machine. If there is a point with a significantly lower temperature than the highest point at a certain position, select several temperature points from the temperature drop point toward the head of the machine for quadratic curve fitting, and solve the curve vertex position as the sintering end point. If the highest point temperature does not reach the set temperature, there is no sintering end point.

5. The method according to any one of claims 1 to 4, characterized in that The method further comprises: An image classifier is constructed to classify grayscale images. The current bellows temperature is processed into a grayscale image and used as the input of the image classifier to obtain the output probability of each category. The category with the highest probability is selected as the classification result. According to the classification result of the current wind box temperature distribution, the sintering end point position is calculated using the category rule corresponding to the classification result.

6. The method according to claim 5, characterized in that The constructing of an image classifier to classify grayscale images includes: After clustering, add corresponding category labels to each grayscale image, and use all grayscale images with labeled categories as sample sets; Construct a deep convolutional neural network classifier, which takes as input a grayscale image of temperature distribution, passes through several convolution blocks, and finally outputs the probability value of each classification through an output layer containing several neurons; The grayscale image sample set with category annotations is trained as a deep convolutional neural network classifier.

Images