Method for identifying information of waste steel entering converter

Abstract

The invention relates to a method for identifying information of waste steel entering a converter, and belongs to the technical field of ferrous metallurgy. According to the technical scheme, the method comprises the steps that scrap steel is loaded into a scrap steel bucket batch by batch through a crown block by means of a suction magnetic disk crane, the type of the loaded scrap steel is recognized through a deep learning artificial intelligence model, and meanwhile the scrap steel adding time and the weighing weight are recorded; after the scrap steel loading operation is completed, recording the time when the ground vehicle is driven out as the loading ending time; an industrial camera is installed on a converter platform, before a crown block is hoisted to the converter platform, a converter starts a scrap steel adding operation signal, the number of a scrap steel hopper is recognized through the industrial camera, and meanwhile the scrap steel entering time is recorded. According to the method, the accuracy of the information of the steel scrap entering the converter is improved, accurate basic data are provided for automatic, modeling and intelligent production of the converter, the labor intensity of operators is effectively relieved, and a foundation is laid for intelligent transformation and high-quality development of converter smelting.

Description

Technical Field

[0001] The invention relates to a method for identifying scrap steel information entering a converter, and belongs to the technical field of iron and steel metallurgy. Background Art

[0002] Scrap steel is one of the main raw materials for converter smelting, and is an important metal material and coolant for converter smelting. Reasonable scrap steel addition can significantly reduce the consumption of converter auxiliary materials such as lime, limestone, light-burned dolomite, and sintered ore, while increasing production output. Scrap steel input information, including the type and weight of scrap steel input, is the basis for converter smelting model calculation, and is also the main factor restricting converter production stability, modeling, and intelligence.

[0003] There are two main problems in identifying scrap steel entering the furnace. One is the classification and identification of scrap steel in the scrap steel bucket, and the other is the correspondence between the scrap steel bucket and the converter.

[0004] There are many types of scrap steel used in converter smelting, including self-produced scrap steel such as continuous casting billet heads and tails, steel rolling trimmings and recycled steel slag, and purchased scrap steel such as steel bar cuttings, ship plate scrappings, automobile plate briquettes, and recycled scrap steel. The shapes, components, densities and other factors that affect smelting of various scrap steels are different, which increases the difficulty of scrap steel identification. Especially for steel mills that were planned and built in the early stage, the overall space of the scrap steel room is small, and there is no sufficient space for scrap steel classification. It is impossible to classify and stack different types of scrap steel at fixed points. The placement location is often changed at will. At the same time, there is the problem of placing multiple types of scrap steel in the same silo, which further increases the difficulty of scrap steel type identification.

[0005] At present, there are two main methods for identifying the type of scrap steel entering the furnace. One is to rely on manual methods by operators to confirm the scrap steel lifted by the overhead crane to the scrap steel bucket, and then record the weight and type, and transmit it to the converter database through the secondary system. This method increases the labor intensity of manual labor and there are errors in manual recording of scrap steel; the second is to use the positioning method of the overhead crane to automatically determine the type and weight of scrap steel by adsorbing scrap steel at different positions of the electromagnetic crane in the scrap steel room. The scrap steel information is manually matched with the scrap steel bucket number, and it is transmitted to the converter database through the secondary system. This method needs to clarify and fix the storage locations of different types of scrap steel in the scrap steel identification process in order to accurately identify the type of scrap steel. It cannot cope with the problem that the space in the scrap steel room is small, the scrap steel storage silo cannot be clearly stored, or a variety of scrap steel is stored in one silo. At the same time, the manual correspondence between the scrap steel bucket number and the scrap steel type will also cause errors.

[0006] The problem of the correspondence between the scrap bucket and the converter is mainly to correspond the scrap bucket number with the converter seat to be added. For the converter process that is not smelted in a single seat, due to the uncertainty of converter smelting, including changes in production plans, uncertainty in converter smelting cycles, and the impact of emergencies, it is impossible to correspond the scrap bucket with the seat in advance, which increases the difficulty of automatically identifying converter and scrap bucket information. The main corresponding method is manual correspondence. For example, Danieli's converter smelting model requires manual confirmation of the scrap bucket and converter seat before converter smelting, and then find the manual connection of the corresponding scrap bucket number from the database, and then perform converter model calculation. This method has errors caused by manual participation in confirmation.

[0007] The patent with application number 201310484859.X discloses "a weighing method and system for converter scrap steel raw materials". The method consists of a monitoring device, a wireless positioning device for an overhead crane and a weighing device. The material yard is divided into grid areas horizontally and vertically, the types of scrap steel in the grid are defined, and the monitoring camera is used for data transmission. The weighing of the scrap steel raw materials is completed by lifting the position and weighing the weight.

[0008] The utility model patent with application number 201720733116.5 discloses "a device for automatically collecting and proportioning scrap steel from a converter". The method consists of a scrap steel pool, a scrap steel pool positioning device, an electromagnetic crane, an electromagnetic crane positioning device, an electromagnetic crane weighing device and a scrap steel bucket weighing device. The type of scrap steel is determined by positioning the scrap steel pool and the electromagnetic crane, and finally the type and weight of scrap steel are collected.

[0009] The above methods cannot adapt to the changing scrap steel placement conditions, and cannot effectively solve the problem of small scrap steel space and unstable scrap steel placement. At the same time, they cannot solve the problem of accurately and automatically transmitting scrap steel information to the converter smelting model. Summary of the invention

[0010] The purpose of the present invention is to provide a method for identifying scrap steel information entering a converter. By identifying the type and weight of scrap steel added by a scrap steel electromagnetic crane, the method is widely applicable to various scrap steel room site environments, and can effectively solve the problem that the scrap steel room is small and automatic identification of scrap steel cannot be performed through overhead crane positioning. Through scrap steel bucket number identification, the correspondence between the converter and the scrap steel is achieved, the drawback of manual entry of scrap steel information is solved, the problem of manual participation in the process of obtaining converter scrap steel information is reduced, the accuracy of the scrap steel information entering the converter is improved, and accurate basic data is provided for converter automation, modeling and intelligent production, which effectively reduces the labor intensity of operators, lays the foundation for the intelligent transformation and high-quality development of converter smelting, and effectively solves the above-mentioned problems existing in the background technology.

[0011] The technical solution of the present invention is: a method for identifying scrap steel information entering a converter, comprising the following steps:

[0012] (1) An industrial camera is installed at the scrap steel loading position. When the local vehicle runs to the scrap steel loading limit area, a limit signal is generated, and the scrap steel bucket visual recognition unit is activated to identify and record the scrap steel bucket number and the loading start time;

[0013] (2) The overhead crane loads the scrap steel into the scrap steel bucket batch by batch through the suction magnetic disk crane, and uses the deep learning artificial intelligence model to identify the type of scrap steel loaded this time, while recording the time of scrap steel addition and weighing weight;

[0014] (3) When the scrap steel loading operation is completed, the ground vehicle drives the scrap steel bucket out. During this process, the scrap steel bucket leaves the loading limit position, generating a loading end signal, and the time when the ground vehicle drives out is recorded as the loading end time;

[0015] (4) Install an industrial camera on the converter platform, wait for the overhead crane to be lifted to the converter platform, and then receive the converter start-up signal to add scrap steel. The industrial camera identifies the scrap steel bucket number and records the time when the scrap steel enters the furnace.

[0016] In the step (1), the scrap bucket visual recognition unit identifies the operating status of the scrap bucket through an industrial camera, identifies the operating status of the scrap bucket through an operating status recognition logic, and identifies the number of the scrap bucket through a picture processing program and a deep learning artificial intelligence model.

[0017] In the step (2), the deep learning artificial intelligence model includes a suction magnetic disk crane detection model and a scrap steel type recognition model;

[0018] The suction disk crane detection model is established through machine vision detection and deep learning methods. Its key steps are as follows:

[0019] ① Use 576*576 pixels to represent the features of the disk crane with material and the disk crane without material, and determine it as the size of the dataset image;

[0020] ② Use industrial cameras to take photos, select images of disk cranes at different positions and angles, and cover different light changes and different dust diffusion levels as the data set;

[0021] ③ The data set is divided into a training set and a test set, with the training set accounting for 75% of the total number of photos and the test set accounting for 25% of the total number of photos;

[0022] ④ Manually calibrate these photos to form training set labels;

[0023] ⑤ The number of images of the two categories of disk crane without material and disk crane with material in the training set is relatively balanced to prevent category bias during training and improve the generalization ability of the model;

[0024] ⑥ The laboratory trained the model for 100 rounds to make the accuracy of the test set reach the expected value, and then used the trained weights to detect the suction disk crane;

[0025] ⑦ The trained magnetic suction crane detection model is used to detect whether the magnetic suction crane appears in the charging workshop and its location information when it appears, so as to provide recognition images for subsequent scrap steel types;

[0026] The scrap steel type recognition model is established through machine vision detection and deep learning methods. The key steps are as follows:

[0027] ①Use industrial cameras to take photos and collect scrap steel images;

[0028] ② The data set is divided into a training set and a test set, with the training set accounting for 75% of the total number of photos and the test set accounting for 25% of the total number of photos;

[0029] ③ Manually calibrate these photos to form training set labels;

[0030] ④ The number of images of heavy scrap steel, small scrap steel, light scrap steel, baled scrap steel, plate edge materials and magnetic separation slag scrap steel in the training set is relatively balanced to prevent category bias during training and improve the generalization ability of the model;

[0031] ⑤ The laboratory trains the model for 100 rounds to make the accuracy of the test set reach the expected value, and then uses the trained weights to detect the types of scrap steel.

[0032] In step (3), when T 进 (n) <T 吊 (x i ) <T 结 (n) is the total amount of scrap steel in bucket n, where n is the scrap steel bucket number, T 进 (n) is the loading start time, T 吊 (x i ) is the time of adding scrap steel, T 结 (n) is the loading end time, MT 吊 (x i ) is the weighing weight, x i is the type of scrap steel, and i is the scrap steel type number.

[0033] In step (4), when T 结 (n) <T 入 (n), the time corresponding to the nth bucket The type and weight of scrap steel loaded into the converter; when the converter is in the scrap steel loading end operation signal, the corresponding bucket of No. n is Record deletion to avoid data duplication, where n is the scrap bucket number, T 结(n) is the loading end time, T 入 (n) is the time when scrap steel enters the furnace, MT 吊 (x i ) is the weighing weight.

[0034] The beneficial effects of the present invention are as follows: by identifying the type and weight of scrap steel added by the scrap steel electromagnetic crane, the present invention is widely applicable to various scrap steel room environment, and can effectively solve the problem that the scrap steel room is small and the scrap steel cannot be automatically identified by overhead crane positioning. Through the scrap steel bucket number identification, the correspondence between the converter and the scrap steel is achieved, the drawbacks of manual entry of scrap steel information are solved, the problem of manual participation in the process of obtaining converter scrap steel information is reduced, the accuracy of the scrap steel information entering the converter is improved, and accurate basic data is provided for converter automation, modeling and intelligent production, which effectively reduces the labor intensity of operators and lays the foundation for the intelligent transformation and high-quality development of converter smelting. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 is a flow chart of the method of the present invention;

[0036] Figure 2 It is the industrial camera layout diagram of the present invention;

[0037] Figure 3 It is a schematic diagram of the working area of ​​the present invention;

[0038] Figure 4 It is a reasoning diagram of the material suction magnetic disk crane detection model of the present invention;

[0039] Figure 5 It is the furnace scrap steel identification system diagram and the furnace number described in the embodiment of the present invention;

[0040] In the figure: overhead crane operating span 1, scrap steel pile 2, loading station 3, converter station 4, industrial camera 5, converter furnace body 6, converter overhead crane 7, converter main control 8, scrap steel ground car 9, scrap steel loading operation area A, waiting area for overhead crane lifting B, converter scrap steel loading area C. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solutions and advantages of the invention implementation cases clearer, the technical solutions in the implementation cases of the present invention will be clearly and completely described below in conjunction with the drawings in the implementation cases. Obviously, the implementation cases described are only a small part of the implementation cases of the present invention, rather than all the implementation cases. Based on the implementation cases in the present invention, all other implementation cases obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0042] A method for identifying scrap steel information entering a converter comprises the following steps:

[0043] (1) An industrial camera is installed at the scrap steel loading position. When the local vehicle runs to the scrap steel loading limit area, a limit signal is generated, and the scrap steel bucket visual recognition unit is activated to identify and record the scrap steel bucket number and the loading start time;

[0044] (2) The overhead crane loads the scrap steel into the scrap steel bucket batch by batch through the suction magnetic disk crane, and uses the deep learning artificial intelligence model to identify the type of scrap steel loaded this time, while recording the time of scrap steel addition and weighing weight;

[0045] (3) When the scrap steel loading operation is completed, the ground vehicle drives the scrap steel bucket out. During this process, the scrap steel bucket leaves the loading limit position, generating a loading end signal, and the time when the ground vehicle drives out is recorded as the loading end time;

[0046] (4) Install an industrial camera on the converter platform, wait for the overhead crane to be lifted to the converter platform, and then receive the converter start-up signal to add scrap steel. The industrial camera identifies the scrap steel bucket number and records the time when the scrap steel enters the furnace.

[0047] In the step (1), the scrap bucket visual recognition unit identifies the operating status of the scrap bucket through an industrial camera, identifies the operating status of the scrap bucket through an operating status recognition logic, and identifies the number of the scrap bucket through a picture processing program and a deep learning artificial intelligence model.

[0048] In the step (2), the deep learning artificial intelligence model includes a suction magnetic disk crane detection model and a scrap steel type recognition model;

[0049] The suction disk crane detection model is established through machine vision detection and deep learning methods. Its key steps are as follows:

[0050] ① Use 576*576 pixels to represent the features of the disk crane with material and the disk crane without material, and determine it as the size of the dataset image;

[0051] ② Use industrial cameras to take photos, select images of disk cranes at different positions and angles, and cover different light changes and different dust diffusion levels as the data set;

[0052] ③ The data set is divided into a training set and a test set, with the training set accounting for 75% of the total number of photos and the test set accounting for 25% of the total number of photos;

[0053] ④ Manually calibrate these photos to form training set labels;

[0054] ⑤ The number of images of the two categories of disk crane without material and disk crane with material in the training set is relatively balanced to prevent category bias during training and improve the generalization ability of the model;

[0055] ⑥ The laboratory trained the model for 100 rounds to make the accuracy of the test set reach the expected value, and then used the trained weights to detect the suction disk crane;

[0056] ⑦ The trained magnetic suction crane detection model is used to detect whether the magnetic suction crane appears in the charging workshop and its location information when it appears, so as to provide recognition images for subsequent scrap steel types;

[0057] The scrap steel type recognition model is established through machine vision detection and deep learning methods. The key steps are as follows:

[0058] ①Use industrial cameras to take photos and collect scrap steel images;

[0059] ② The data set is divided into a training set and a test set, with the training set accounting for 75% of the total number of photos and the test set accounting for 25% of the total number of photos;

[0060] ③ Manually calibrate these photos to form training set labels;

[0061] ④ The number of images of heavy scrap steel, small scrap steel, light scrap steel, baled scrap steel, plate edge materials and magnetic separation slag scrap steel in the training set is relatively balanced to prevent category bias during training and improve the generalization ability of the model;

[0062] ⑤ The laboratory trains the model for 100 rounds to make the accuracy of the test set reach the expected value, and then uses the trained weights to detect the types of scrap steel.

[0063] In step (3), when T 进 (n) <T 吊 (x i ) <T 结 (n) is the total amount of scrap steel in bucket n, where n is the scrap steel bucket number, T 进 (n) is the loading start time, T 吊 (x i ) is the time of adding scrap steel, T 结 (n) is the loading end time, MT 吊 (x i ) is the weighing weight, x i is the type of scrap steel, and i is the scrap steel type number.

[0064] In step (4), when T 结 (n) <T 入 (n), the time corresponding to the nth bucket The type and weight of scrap steel loaded into the converter; when the converter is in the scrap steel loading end operation signal, the corresponding bucket of No. n is Record deletion to avoid data duplication, where n is the scrap bucket number, T 结(n) is the loading end time, T 入 (n) is the time when scrap steel enters the furnace, MT 吊 (x i ) is the weighing weight.

[0065] In practical applications, the present invention uses a computer for visual recognition, an industrial camera for scrap steel bucket identification, and an industrial camera for scrap steel type identification.

[0066] The specific steps include:

[0067] 1. Install an industrial camera at the scrap steel loading position. When the vehicle runs to the scrap steel loading limit area, a limit signal is generated, and the scrap steel bucket visual recognition unit starts to identify and record the scrap steel bucket number n, and also records the loading start time T. 进 (n).

[0068] The scrap steel bucket visual recognition unit mainly identifies the operating status of the scrap steel bucket through industrial cameras, identifies the operating status of the scrap steel bucket through operating status recognition logic, and identifies the numbers of the scrap steel bucket through image processing programs and deep learning artificial intelligence models.

[0069] 2. The scrap steel is loaded into the scrap steel bucket batch by batch through the overhead crane using the suction magnetic disk crane. The scrap steel is sucked into the scrap steel bucket by the suction magnetic disk crane, and the type of scrap steel loaded this time is automatically identified by the industrial camera using the deep learning artificial intelligence model. i , i is the scrap steel type number. At the same time, the scrap steel adding time T is recorded 吊 (xi) and weighing weight MT 吊 (x i ).

[0070] The deep learning artificial intelligence model includes a suction magnetic disk crane detection model and a scrap steel type recognition model. Since the suction magnetic disk crane to be detected and the magnetic disk crane without material are similar in appearance, especially when the mass of the sucked scrap steel is small, the two can be easily misidentified, so two samples of magnetic disk crane with material and magnetic disk crane without material are selected. The suction magnetic disk crane detection model is used to detect whether the suction magnetic disk crane appears in the charging workshop and the location information when it appears, thereby providing an identification image for the subsequent scrap steel type.

[0071] The suction disk crane detection model is established through machine vision detection and deep learning methods. Its key steps are as follows:

[0072] ① The pixel size of 576*576 is sufficient to characterize the features of the disk crane with material and the disk crane without material, and is determined as the size of the dataset image.

[0073] ② Use industrial cameras to take a large number of photos, select images of disk cranes at different positions and angles, and cover different light changes and different degrees of dust diffusion as the data set.

[0074] ③ The data set is divided into a training set and a test set, with the training set accounting for 75% of the total number of photos and the test set accounting for 25% of the total number of photos;

[0075] ④Manually calibrate these photos to form training set labels.

[0076] ⑤ The number of images of the two categories of disk crane without material and disk crane with material in the training set is relatively balanced to prevent category bias in the training process and improve the generalization ability of the model.

[0077] ⑥ The laboratory trained the model for 100 rounds to make the accuracy of the test set reach the expected value, and then used the trained weights to detect the suction disk crane.

[0078] ⑦ The trained suction magnetic disk crane detection model is used to detect whether the suction magnetic disk crane appears in the charging workshop and its location information when it appears, thereby providing an identification image for the subsequent scrap steel types.

[0079] The scrap steel type recognition model is established through machine vision detection and deep learning methods. The key steps are as follows:

[0080] ① Use industrial cameras to take a large number of photos and collect scrap steel images.

[0081] ② The data set is divided into a training set and a test set, with the training set accounting for 75% of the total number of photos and the test set accounting for 25% of the total number of photos;

[0082] ③Manually calibrate these photos to form training set labels.

[0083] ④ The number of images of heavy scrap steel, small scrap steel, light scrap steel, packaged scrap, plate edge materials, magnetic separation slag and other scrap steel types in the training set is relatively balanced to prevent category bias in the training process and improve the generalization ability of the model.

[0084] ⑤ The laboratory trains the model for 100 rounds to make the accuracy of the test set reach the expected value, and then uses the trained weights to detect the types of scrap steel.

[0085] When the scrap steel loading operation is completed, the ground vehicle will drive the scrap steel bucket out. During this process, the scrap steel bucket leaves the loading limit, generating a loading end signal, and the time when the ground vehicle drives out is recorded as the loading end time T 结 (n). When T 进 (n) <T 吊 (x i ) <T 结 (n) is the total amount of scrap steel in bucket No. n.

[0086] 4. Install an industrial camera on the converter platform. Wait for the overhead crane to be lifted to the converter platform, and the converter starts to add scrap steel operation signal. The industrial camera identifies the scrap steel bucket number n and records the scrap steel entering the furnace time T. 入 (n). When T 结 (n) <T 入 (n), the time corresponding to the nth bucket The type and weight of scrap steel loaded into the converter. When the converter is in the scrap steel loading end operation signal, the corresponding bucket of No. n is Record deletion to avoid data duplication.

[0087] Example:

[0088] The specific implementation steps are as follows:

[0089] 1. The scrap bucket is transported to the scrap loading station by a ground vehicle and loaded on site through the No. 7 scrap bucket. During the transportation of the scrap bucket, the scrap bucket number is visually identified as No. 7, and the start time of the operation is recorded as T 进 (7) = 00:15:00, the operation end time is T 结 (7)=00:22:50.

[0090] 2. In T 进 (7) With T 结 (7) A total of 10 loads of scrap steel were identified, of which 7 types were plate edges, M T吊(板边) A total of 16.25 tons; magnetic separation slag 1 time, M T吊(磁选渣) Total 5.6t; heavy scrap 2 times M T吊(重废) Total 3.6t; scrap steel A total of 25.45 tons.

[0091] 3. The scrap steel bucket is hoisted to the front of the converter for scrap steel loading. The scrap steel bucket number is identified as No. 7 by the visual recognition camera in front of the 2# furnace, and the start time of loading scrap steel is recorded as T 入 (n) = 01: 15: 01. The comprehensive information of the type and weight of scrap steel used in the converter is transmitted to the converter secondary system for converter secondary model calculation.

Claims

1. A method for identifying scrap steel information entering a converter, characterized in that The following steps are involved: (1) An industrial camera is installed at the scrap steel loading position. When the local vehicle runs to the scrap steel loading limit area, a limit signal is generated, and the scrap steel bucket visual recognition unit is activated to identify and record the scrap steel bucket number and the loading start time; (2) The overhead crane loads the scrap steel into the scrap steel bucket batch by batch through the suction magnetic disk crane, and uses the deep learning artificial intelligence model to identify the type of scrap steel loaded this time, while recording the time of scrap steel addition and weighing weight; (3) When the scrap steel loading operation is completed, the ground vehicle drives the scrap steel bucket out. During this process, the scrap steel bucket leaves the loading limit position, generating a loading end signal, and the time when the ground vehicle drives out is recorded as the loading end time; (4) Install an industrial camera on the converter platform, wait for the overhead crane to be lifted to the converter platform, and then receive the converter start-up signal to add scrap steel. The industrial camera identifies the scrap steel bucket number and records the time when the scrap steel enters the furnace.

2. The method for identifying scrap steel information entering a converter according to claim 1, characterized in that: In the step (1), the scrap bucket visual recognition unit identifies the operating status of the scrap bucket through an industrial camera, identifies the operating status of the scrap bucket through an operating status recognition logic, and identifies the number of the scrap bucket through a picture processing program and a deep learning artificial intelligence model.

3. The method for identifying scrap steel information entering a converter according to claim 1, characterized in that: In the step (2), the deep learning artificial intelligence model includes a suction magnetic disk crane detection model and a scrap steel type recognition model; The suction disk crane detection model is established through machine vision detection and deep learning methods. Its key steps are as follows: ① Use 576*576 pixels to characterize the features of the disk crane with material and the disk crane without material, and determine it as the size of the dataset image; ② Use industrial cameras to take photos, select images of disk cranes at different positions and angles, and cover different light changes and different dust diffusion levels as the data set; ③ The data set is divided into a training set and a test set, with the training set accounting for 75% of the total number of photos and the test set accounting for 25% of the total number of photos; ④ Manually calibrate these photos to form training set labels; ⑤ The number of images of the two categories of disk crane without material and disk crane with material in the training set is relatively balanced to prevent category bias during training and improve the generalization ability of the model; ⑥ The laboratory trained the model for 100 rounds to make the accuracy of the test set reach the expected value, and then used the trained weights to detect the suction disk crane; ⑦ The trained magnetic suction crane detection model is used to detect whether the magnetic suction crane appears in the charging workshop and its location information when it appears, so as to provide recognition images for subsequent scrap steel types; The scrap steel type recognition model is established through machine vision detection and deep learning methods. The key steps are as follows: ①Use industrial cameras to take photos and collect scrap steel images; ② The data set is divided into a training set and a test set, with the training set accounting for 75% of the total number of photos and the test set accounting for 25% of the total number of photos; ③ Manually calibrate these photos to form training set labels; ④ The number of images of heavy scrap steel, small scrap steel, light scrap steel, baled scrap steel, plate edge materials and magnetic separation slag scrap steel in the training set is relatively balanced to prevent category bias during training and improve the generalization ability of the model; ⑤ The laboratory trains the model for 100 rounds to make the accuracy of the test set reach the expected value, and then uses the trained weights to detect the types of scrap steel.

4. The method for identifying scrap steel information entering a converter according to claim 1, characterized in that: In step (3), when T 进 (n) <T 吊 (x i ) <T 结 (n) is the total amount of scrap steel in bucket n, where n is the scrap steel bucket number, T 进 (n) is the loading start time, T 吊 (x i ) is the time of adding scrap steel, T 结 (n) is the loading end time, MT 吊 (x i ) is the weighing weight, x i is the type of scrap steel, and i is the scrap steel type number.

5. The method for identifying scrap steel information entering a converter according to claim 1, characterized in that: In step (4), when T 结 (n) <T 入 (n), the time corresponding to the nth bucket The type and weight of scrap steel loaded into the converter; when the converter is in the scrap steel loading end operation signal, the corresponding bucket of No. n is Record deletion to avoid data duplication, where n is the scrap bucket number, T 结 (n) is the end time of scrap steel loading, T 入 (n) is the time when scrap steel enters the furnace, MT 吊 (x i ) is the weighing weight.

Images