CN-121975990-A - Intelligent slag splashing control method for converter

CN121975990ACN 121975990 ACN121975990 ACN 121975990ACN-121975990-A

Abstract

The intelligent slag splashing control method for the converter acquires flame images of the converter mouth, carries out real-time identification on the flame images by utilizing an image classification model based on deep learning, outputs state types of slag splashing process, wherein the state types comprise early stage, middle normal state, middle less 1, middle less 2, rear normal state, rear less 1 and rear less 2 of slag splashing, determines the adjustment quantity of the gun position of an oxygen gun according to the state types, calculates the height of a target gun position, controls an oxygen gun executing mechanism according to the height of the target gun position to carry out gun position adjustment, realizes closed-loop control on the slag splashing process, and can automatically and real-time identify flame form, color and dynamic change by an image acquisition device and an image classification model based on a convolutional neural network, and the fuzzy visual experience is converted into seven quantized states of early stage, middle normal state, middle less 1, middle less 2, rear normal state, rear less 1 and rear less 2, objective continuous accurate perception on the slag splashing process is realized, and the dependence on manual experience is thoroughly eliminated.

Inventors

FENG CHUNSONG
CAO YONGGEN
QIN WEIFENG
LI JIAYUN
HU TONGFANG
Xue Zhenhao

Assignees

江苏省镔鑫钢铁集团有限公司

Dates

Publication Date: 20260505
Application Date: 20260126

Claims (9)

1. The intelligent slag splashing control method of the converter is characterized by comprising the following steps of: (1) Acquiring flame images of the converter mouth in real time through an image acquisition device; (2) Carrying out real-time identification on the flame image by using an image classification model based on deep learning, and outputting state categories of a slag splashing process, wherein the state categories comprise early stage, middle normal, middle less 1, middle less 2, rear normal, rear less 1 and rear less 2 of slag splashing; (3) Determining the adjustment quantity of the lance position of the oxygen lance according to the state type, and calculating the height of the target lance position; (4) Controlling an oxygen lance executing mechanism to adjust the lance position according to the target lance position height, so as to realize closed-loop control on the slag splashing process; Wherein, the step (2) adopts a staged time triggering strategy to execute image recognition: the default output state is the earlier stage of slag splashing within the first 30 seconds after the slag splashing starts; Starting a first classification model from 30 seconds, identifying an input flame image, and outputting one of three states of normal, middle and small 1, middle and small 2; starting a second classification model from the 90 th second, identifying the input flame image, and outputting one of the three states of normal state, 1 less later and 2 less later; And when the second classification model continuously identifies the last 2 states of the preset times, judging that the slag splashing process is finished.
2. The intelligent slag splashing control method of the converter of claim 1, wherein the first classification model and the second classification model are classification models based on a convolutional neural network; The structure of the convolutional neural network comprises an input normalization layer, at least two convolutional layers, a flattening layer, a full-connection layer and a Softmax classification output layer; wherein the convolutional layer uses a3×3 convolutional kernel, a ReLU activation function, and is followed by a2×2 pooling layer.
3. The intelligent slag splashing control method of a converter according to claim 2, wherein the convolutional neural network comprises two convolutional layers, the first convolutional layer has 32 3×3 convolutional kernels, and the second convolutional layer has 64 3×3 convolutional kernels.
4. The intelligent slag splashing control method of a converter according to claim 1, wherein in the step (3), a gun position correction amount Δh is preset for each state type, and a mapping rule of the gun position correction amount Δh is as follows: when the status category is middle normal or rear normal, Δh=0; when the state class is less than 1 in or less than 1 after, Δh= -20mm; When the state class is middle less than 2 or later less than 2, Δh= -40mm.
5. The intelligent slag splashing control method of a converter according to claim 4, wherein in the step (3), the target gun position height Hcmd (t) is dynamically calculated according to the current gun position height Hcur (t) and the gun position correction amount Δh, and a calculation formula is as follows: Hcmd(t)=Hcur(t)+ΔH; and, the calculated target gun position height Hcmd (t) needs to satisfy the following physical constraint conditions: (a) High amplitude limiting, namely Hmin is less than or equal to Hcmd (t) is less than or equal to Hmax, wherein Hmin and Hmax are the lowest and highest safe positions allowed by the oxygen lance respectively; (b) The speed limiting is that the value of [ Hcmd (t) -Hcmd (t-delta t) ]/delta t is less than or equal to Vmax, hcmd (t-delta t) is the target gun position height of the previous control period, wherein Vmax is the maximum lifting speed allowed by the oxygen gun.
6. The intelligent slag splashing control method of the converter of claim 1, wherein the state type is represented by a state variable S (t), S (t) =1, 2,3,4,5,6,7 corresponds to early stage, middle normal, middle less 1, middle less 2, rear normal, rear less 1 and rear less 2 of slag splashing respectively, a control mode of the system is determined according to the current identification state S (t), and a final control instruction is synthesized: Defining a take-over mark variable M (t), wherein the assignment rule is that if S (t) is the earlier stage of slag splashing, M (t) =0, the system adopts a static table query mode to control, otherwise M (t) =1, and the system is switched to a dynamic database control mode based on a real-time identification result; The control command Uplc (t) finally issued to the actuator is determined by the following equation: When M (t) =1, uplc (t) =uddynamicdb (t), which is a control instruction dynamically generated based on a real-time image recognition result; when M (t) =0, uplc (t) = UStaticTable (t), and UStaticTable (t) is a control instruction obtained by querying according to a preset static parameter table.
7. The intelligent slag splashing control method of the converter of claim 1, wherein the step (4) is specifically realized by modifying the original programmable logic controller program of the converter, the target gun position height is issued to the programmable logic controller, the controller drives the oxygen gun lifting executing mechanism to act, and meanwhile, the state type, the target gun position height and the slag splashing process parameters are displayed in real time through a man-machine interaction interface.
8. The intelligent slag splashing control method for the converter according to claim 1 or 5, wherein the slag splashing end is assisted and judged through a continuous state counter C (t), and the updating rule of the counter is as follows: if the current recognition state S (t) is less than 2, let C (t) =c (t- Δt) +1; If S (t) is not the latter 2 less, let C (t) =0; Wherein Δt is the recognition period.
9. The intelligent slag splashing control method of the converter according to claim 8, wherein a slag splashing end instruction is triggered when the value of the continuous state counter C (t) reaches or exceeds a preset threshold N; the ending instruction comprises setting the oxygen lance target height Hcmd (t) as the lance safety height Hlift and setting a Splash End flag splash_end.

Description

Intelligent slag splashing control method for converter Technical Field The invention relates to the technical field of steel industry, in particular to an intelligent slag splashing control method of a converter. Background In order to overcome the defects of the prior art, the invention aims to provide an intelligent slag splashing model of a converter, which realizes real-time monitoring and dynamic adjustment of a slag splashing process by integrating a flame (slag splashing) image recognition technology based on time phase division and a dynamic intelligent control model. In the steel production process, converter steelmaking is used as a core link, and intelligent transformation is important to improving production efficiency, reducing cost and guaranteeing safety. The slag splashing furnace protection of the converter is a key process for prolonging the service life of a furnace lining and ensuring the smooth operation of steelmaking, and the operation effect directly influences the production cost and the equipment safety. At present, the traditional slag splashing furnace protection operation of the converter mainly depends on manual experience. The operator can judge the slag splashing process, slag lifting amount and effect by visual characteristics of the flame at the furnace mouth, such as color, brightness, morphology and the like through naked eyes, and can manually adjust the technological parameters of the oxygen lance position, nitrogen flow and the like according to the slag splashing process, the slag lifting amount and the effect. This manually dominated mode of operation suffers from the following significant drawbacks: (1) The operation is unstable, the quality fluctuation is large, the manual observation subjectivity is strong, the interference of factors such as ambient light, individual experience difference, visual fatigue and the like is easy, the judgment of the slag splashing state is inconsistent and inaccurate, the slag splashing operation stability is poor, and the fluctuation of the furnace protection effect is large. (2) The production efficiency and the cost control are difficult to optimize, the optimal real-time cooperative control of the technological parameters is difficult to realize depending on experience, the slag splashing time is possibly caused to affect the production rhythm excessively, or the medium consumption such as nitrogen is uneconomical, so that the production cost is increased. (3) The labor intensity is high, the safety risk is high, operators need to continuously concentrate on observation and operation at high-temperature and high-noise industrial sites, the mental pressure and physical burden is heavy, and the safety risk such as splashing caused by judgment lag or error exists. (4) Knowledge inheritance and standardization are difficult, experience of excellent operators is difficult to accurately quantify, solidify and inherit, standardization and continuous optimization of the process are not facilitated, and improvement of production management level is restricted. In recent years, with the advancement of intelligent transformation in the steel industry, basic automation or simple models are tried to be introduced for auxiliary control, but most of the intelligent transformation is still limited to fixed time sequence control or simple feedback based on single and static parameters, and the intelligent transformation method lacks the capability of real-time accurate sensing and intelligent decision making of multidimensional and dynamic changes in the slag splashing process, so that the problems cannot be fundamentally solved. Therefore, the converter slag splashing control method and system capable of accurately identifying the slag splashing state in real time and performing intelligent dynamic regulation and control according to the slag splashing state are developed, and the method and system have urgent practical significance and important application value for realizing the intellectualization and standardization of slag splashing furnace protection operation, improving the production efficiency, stabilizing the process quality, and reducing the consumption and the labor intensity. Disclosure of Invention The invention aims to solve the technical problem of providing a converter intelligent slag splashing control method based on a real-time visual perception and intelligent decision closed loop. The technical problem to be solved by the invention is realized by the following technical scheme, and the intelligent slag splashing control method of the converter comprises the following steps: (1) Acquiring flame images of the converter mouth in real time through an image acquisition device; (2) Carrying out real-time identification on the flame image by using an image classification model based on deep learning, and outputting state categories of a slag splashing process, wherein the state categories comprise early stage, middle normal, mid