CN210765379U

CN210765379U - Device for intelligent tapping of converter

Info

Publication number: CN210765379U
Application number: CN201921393774.XU
Authority: CN
Inventors: 李培玉; 沈国振; 王国春; 王伟
Original assignee: Hangzhou Putedy Industrial Co ltd
Current assignee: Hangzhou Putedy Industrial Co ltd
Priority date: 2019-08-26
Filing date: 2019-08-26
Publication date: 2020-06-16
Anticipated expiration: 2029-08-26

Abstract

The utility model relates to a metallurgical steelmaking technology, aiming at providing a device for intelligent tapping of a converter. The device comprises a converter steelmaking secondary control system, a converter slag tapping detection system, a steel tapping hole slag stopping system, a converter hole monitoring module, a steel tapping monitoring module, a communication control module and a deep learning host; the tapping monitoring module consists of a tapping monitoring probe and a tapping monitoring processing unit; the communication control module is respectively connected to the deep learning host, the furnace mouth monitoring module, the steel tapping monitoring module, the converter steelmaking secondary control system, the converter slag tapping detection system and the steel tapping hole slag stopping system through signal lines, and bidirectional intercommunication of data information and control signals is achieved. The utility model discloses the device can be used for monitoring tapping of fire door and ladle, real-time developments such as sediment down, for utilizing real-time supervision data, carries out automatic control according to set control rule to the converter tapping operation and provides the hardware equipment condition.

Description

Device for intelligent tapping of converter

Technical Field

The utility model belongs to the technical field of metallurgical steelmaking, in particular to a device for intelligent tapping of converter.

Background

In the converter steelmaking process, when the components and the temperature of the molten steel reach the requirements of the current steel grade, the steel is ready to be tapped. When tapping, the converter body is tilted to a preset angle, and molten steel is poured into a ladle through a tapping hole. In the tapping process, a tapping worker observes the comprehensive conditions of tapping steel flow, liquid level in the converter, large ladle emptying and the like at an observation window behind the converter, the tilting angle of the converter is continuously increased, and meanwhile, the positions of a ladle trolley and a slag ladle car are adjusted, so that the phenomenon that molten steel coiled slag flows into the ladle due to the fact that the liquid level of the steel in the converter is too close to a tapping hole is avoided, and in addition, the phenomenon that the molten steel of steel slag in the converter overflows from the converter mouth due to the fact that the converter cannot be tilted too fast is. In the final stage of tapping, if slag is discharged from a tapping hole, tapping is stopped in time, the converter is swung to a zero position, the ladle trolley is driven out of the tapping position, and the tapping can be completed only by at least 2-3 experienced operators in the whole process through mutual cooperation.

At present, most steel mills adopt a mode of manually controlling converter tapping, and the following problems exist in manually controlled tapping: the converter tilts too fast, the steel slag and the molten steel easily flow out from a large furnace mouth; the converter tilts too slowly, and the slag is easy to flow into a large ladle in the tapping process; the converter tilting angle and the position of the ladle trolley are not well matched, so that molten steel does not flow into a ladle or is not fully stirred; the alloy rotary launder swings and is matched with the tilting angle of the converter and the moving position of the ladle car by mistake, so that the alloy adding time and the adding amount are deviated or the mixing and stirring are not uniform; the moment when tapping is finished is not judged in time, so that steel slag is rolled into a steel ladle to influence the quality of molten steel; the tapping amount is larger than the bearing capacity of the steel ladle, so that molten steel in the steel ladle overflows; multiple persons cooperate with cooperative operation, operation errors are easy to occur in the communication process, and potential safety hazards exist; because the molten steel is in a glowing state, the working environment is severe, and the long-term observation of the glowing molten steel state by naked eyes is not beneficial to the labor protection of workers.

Therefore, it is necessary to provide a more comprehensive apparatus for providing hardware device conditions for intelligent control. Therefore, the intelligent control can be carried out based on the device, the real-time dynamics of tapping, slag discharging and the like of the furnace mouth and the ladle can be monitored, and the automatic operation of the converter tapping operation is realized according to the set control rule by utilizing the real-time monitoring data.

SUMMERY OF THE UTILITY MODEL

The to-be-solved technical problem of the utility model is to overcome the deficiencies in the prior art and provide a device for intelligent tapping of converter.

For solving the technical problem, the utility model discloses a solution is:

the device for the intelligent tapping of the converter comprises a converter steelmaking secondary control system, a converter slag tapping detection system and a tapping hole slag stopping system; the device also comprises a furnace mouth monitoring module, a tapping monitoring module, a communication control module and a deep learning host; the communication control module is respectively connected to the deep learning host, the furnace mouth monitoring module, the steel tapping monitoring module, the converter steel-making secondary control system, the converter slag tapping detection system and the steel tapping hole slag stopping system through signal lines, and bidirectional intercommunication of data information and control signals is achieved.

As an improvement, the furnace mouth monitoring module consists of a furnace mouth monitoring probe and a furnace mouth monitoring processing unit.

As an improvement, the tapping monitoring module consists of a tapping monitoring probe and a tapping monitoring processing unit.

As an improvement, the deep learning host is a learning type computer host integrated with multiple GPUs.

Compared with the prior art, the beneficial effects of the utility model are that:

1. the device can be used for monitoring tapping, slag discharging and other real-time dynamics of a furnace mouth and a steel ladle, and provides hardware equipment conditions for utilizing real-time monitoring data and automatically controlling tapping operation of the converter according to set control rules.

2. The deep learning host of the device is a learning type computer host integrated with a plurality of GPUs, a user can install corresponding software function modules according to needs, analyze and process image data acquired by a furnace mouth monitoring module and a steel tapping monitoring module, generate specific control parameters by utilizing the processed data and a built-in preset rule, and transmit the specific control parameters to a converter steel-making secondary control system through a communication control module to be operated by the converter steel-making secondary control system, so that manual operation is replaced.

Drawings

FIG. 1 is a schematic view of an intelligent tapping device for a converter according to a preferred embodiment of the present invention.

The reference numbers in the figures are: 1, a furnace mouth monitoring probe; 2, a furnace mouth monitoring and processing unit; 3, a communication control module; 4, a converter steelmaking secondary control system; 5, a converter slag discharging detection system; 6, deep learning host; 7, buggy ladle; 8, converter; 9 tapping monitoring probe; and 10 tapping monitoring and processing units.

Detailed Description

It should be noted that the present invention relates to a device for intelligent tapping of a converter, and the technical innovation lies in the connection relationship and position layout of each hardware device. The to-be-solved technical problem of the utility model does not relate to intelligent tapping control process itself, and its technical effect lies in providing hardware equipment condition for follow-up realization intelligent tapping control method. Although the following embodiments will describe the embedded software functional modules of some devices (such as the deep learning host, the fire hole monitoring processing unit, the tapping monitoring processing unit, etc.), these software modules are only illustrative of the implementation of the control method. The utility model discloses describe it and only be based on the utility model discloses an use the show, nevertheless never mean that these software modules also belong to the technical content of the utility model.

The converter intelligent tapping device in the utility model is shown in figure 1 and comprises a converter steelmaking secondary control system 4, a converter slag-off detection system 5 and a tapping hole slag-stopping system; the device also comprises a furnace mouth monitoring module, a tapping monitoring module, a communication control module and a deep learning host 6; the furnace mouth monitoring module consists of a furnace mouth monitoring probe 1 and a furnace mouth monitoring processing unit 2, and the steel tapping monitoring module consists of a steel tapping monitoring probe 9 and a steel tapping monitoring processing unit 10. The communication control module is respectively connected to the deep learning host 6, the furnace mouth monitoring module, the tapping monitoring module, the converter steelmaking secondary control system 4, the converter slag tapping detection system 5 and the tapping hole slag stopping system through signal lines, and bidirectional intercommunication of data information and control signals is realized.

The utility model discloses in, fire door monitoring probe 1 is used for gathering the fire door image of tapping in-process converter in real time, and fire door monitoring processing unit 2 can be through opening, closing, temperature control, dust removal etc. of its PLC control fire door monitoring probe, and furthest ensures monitoring probe long-term steady operation to send fire door monitoring image for degree of depth study host computer 6 through the ethernet. The tapping monitoring probe 9 is used for acquiring a steel flow image from a tapping hole to a steel ladle in the tapping process in real time, the tapping monitoring processing unit 10 can control the monitoring probe to be opened, closed, temperature controlled, dedusted and the like through a PLC (programmable logic controller), so that the monitoring probe can be ensured to operate stably for a long time to the maximum extent, and the tapping monitoring image is sent to the deep learning host 6 through the Ethernet. The deep learning host 6 is a learning type computer host integrated with a plurality of GPUs, can be used for analyzing and processing image data acquired by a furnace mouth monitoring module and a tapping monitoring module after corresponding software function modules are installed, and performs iterative training through a neural network model to improve the accuracy rate of target state recognition in an image; the communication control module is respectively connected to the deep learning host 6, the furnace mouth monitoring module, the tapping monitoring module, the converter steelmaking secondary control system 4, the converter slag tapping detection system 5 and the tapping hole slag stopping system through the Ethernet, and bidirectional intercommunication of data information and control signals is realized.

The utility model discloses in, some equipment are existing equipment or prior art. For example, a converter steelmaking secondary control system, a converter slag tapping detection system and a tap hole slag stopping system are all universal devices which are widely applied by large-scale steelmaking enterprises at home and abroad. The furnace mouth monitoring probe 1 can adopt a mode of monitoring probe combination, for example, a machine vision thermal imager of FLIR A615 model of Philier corporation in America is adopted, and a high-temperature camera of DS-NXCN3A204 model of Haikonweiv corporation is matched to realize double monitoring of the furnace mouth of the converter in a visible light region and a far infrared light wavelength region. The furnace mouth monitoring and processing unit 2 can select a PLC model S7-200SMART of Germany Siemens company, and is matched with a corresponding peripheral control circuit and control software. Tapping monitoring probe 9 may be a machine vision thermal imager model FLIR a615, phililes, usa. The tapping monitoring processing unit 10 can select a PLC model S7-200SMART of Siemens Germany, and is matched with a corresponding peripheral control circuit and control software. The deep learning host 6 is a learning type computer host integrated with multiple GPUs, and can select a T7920 model double-path GPU server of DELL company.

Finally, it should be noted that the above-mentioned embodiments are only specific embodiments of the present invention. Obviously, the present invention is not limited to the above embodiments, and many variations are possible. All modifications which can be derived or suggested by a person skilled in the art from the disclosure of the invention should be considered as within the scope of the invention.

How further the utility model discloses a realize intelligent control on the basis and introduce:

the intelligent tapping control method of the converter based on the device comprises the following steps:

(1) after the converter finishes converting and the sampling is qualified, starting a tapping monitoring module and a furnace mouth monitoring module, and starting to acquire real-time images of a furnace mouth and a steel flow in the tapping process;

(2) the deep learning host 6 sends a tapping signal to the converter steelmaking secondary control system 4 through the communication control module, and the tapping signal controls the converter 8 to tilt to a preset initial tapping position to start tapping;

(3) the furnace mouth monitoring module sends the furnace mouth image monitored in real time to the deep learning host 6, and the deep learning host 6 analyzes the furnace mouth boundary and the slag liquid surface boundary position in the image by using a neural network model to form vectorization parameters; judging whether the furnace mouth has slagging or not or whether slagging risks exist by comparing the furnace mouth with existing data stored in a database of a neural network model;

(4) the steel tapping monitoring module sends the steel flow image monitored in real time to the deep learning host 6, and the deep learning host 6 analyzes the steel flow shape, position, width and steel tapping time in the image by using a neural network model to form vectorization parameters; judging whether slag rolling or slag discharging phenomenon exists in the steel flow or not by comparing the steel flow with existing data stored in a database of a neural network model;

(5) the deep learning host 6 generates specific control parameters of the tilting action time, the tilting target angle and the target angle retention time according to the furnace mouth slagging state and the tapping steel flow state fed back by the neural network model and according to preset rules, and transmits the specific control parameters to the converter steelmaking secondary control system 4 through the communication control module to execute the operation.

In the later stage of the tilting operation, when a converter slag tapping detection system 5 detects a slag tapping signal, a communication control module synchronously transmits the signal to a deep learning host 6; the deep learning host 6 sends an instruction to the converter steelmaking secondary control system 4 through the communication control module, and the deep learning host controls the tapping hole slag stopping system to execute slag stopping operation; after the operation is completed, the tap hole slag trap system transmits a tap end signal to the steelmaking secondary control system 4, which returns to the tilting furnace 8 according to a predetermined scheme.

The deep learning host 6 extracts abnormal target data of slag to be discharged from the furnace mouth, slag discharged from the furnace mouth or slag coil of the steel stream from the image data of the steel stream state and the state change of the furnace mouth, and inputs the corresponding image data as a new sample into the neural network model for iterative training so as to improve the identification accuracy.

The deep learning type neural network technology is an application of an artificial intelligence technology in the field of industrial control, more mature technologies can be directly utilized, and parameters can be adjusted according to actual requirements in a specific application process.

Taking the YOLO neural network model as an example, the training process is as follows: designing a neural network model structure, extracting a characteristic vector through a convolution layer, and obtaining a predicted value through a connection layer; different state targets, the layer number design is different; after the design is finished, pre-training is carried out, a mean square error loss function is adjusted, and weights of different parts are distinguished; then, obtaining a most probable result as a predicted value through a non-maximum inhibition algorithm, and optimizing a neural network to finally reach a result meeting the condition through the iterative model training by increasing the number of samples;

analyzing the image data of the furnace mouth slag-off state and the tapping steel flow state by utilizing a neural network model, and comprising the following steps: adjusting the width and height of images, each image represented by a matrix of pixel values, stacked by rows or columns into a plurality of long vectors; calculating the difference of the image along the horizontal X axis and the vertical Y axis respectively to calculate the image gradient, and then synthesizing the image gradient into a two-dimensional vector; after removing unnecessary parameters by using a vector mask or a filter, loading a YOLO neural network model and setting input for preprocessing; classifying by a YOLO neural network model, giving a probability vector, and calibrating the steel stream boundary position and the slag entrapment boundary position by using the probability vector; and analyzing the steel flow form through the probability vector, counting steel tapping time values, processing the converted information, comparing the processed information with existing data stored in a database of the neural network model, and judging whether the information is abnormal or not.

The deep learning host 6 is mainly used for a machine vision part, and enables a computer to autonomously identify and mark a specified target or state by analyzing and processing data of a specific state characteristic picture. The aim of reducing unnecessary manual operation is achieved by directly transmitting parameters among modules in a computer program. The system is mainly applied to analyzing and judging the states of the steel flow shape, the width, the boundary position of the furnace mouth, the liquid level position of slag in the furnace, whether the furnace mouth is slagging, the surface shape of the slag liquid level and the like. The deep learning host 6 adopts a learning type computer host integrating a plurality of GPUs on hardware. A built-in software algorithm part uses a neural network model, and the target identification accuracy is up to more than 95% after the neural network is constructed under the condition that the number of target samples is continuously accumulated to be enough. Meanwhile, in the running process of the system, new target samples can be continuously collected to carry out continuous iterative model training, so that the neural network is optimized, and the accuracy of target state identification is improved.

An example neural network model training process is as follows:

firstly, a YOLO neural network model (reference: https:// pjredbie. com/darknet/YOLO /) is designed and selected, a characteristic vector is extracted through a convolutional layer, and a connection layer obtains a predicted value. The layer number design is different for different state targets. After the design is finished, pre-training is carried out on the network, the mean square error loss function is adjusted, and weights of different parts are distinguished. And then, obtaining a most possible result as a predicted value through a non-maximum suppression algorithm (non maximum suppression), and optimizing the neural network to finally reach a result meeting the condition through increasing the number of samples and training an iterative model.

Taking the example of judging whether the furnace mouth is submerged slag or not, in the tapping process, the tapping furnace mouth video SPV transmitted back in real time from the furnace mouth monitoring module is used as input information, and the machine identification subsystem MIS in the model calculates the extreme value area of the first derivative through the area of the rapid change of the pixel value function, so that the boundary position of the slag liquid level in the furnace, the boundary position of the furnace mouth and the like can be automatically marked, and whether the slag liquid level boundary exceeds the furnace mouth boundary or not is further calculated, so that whether the furnace mouth is submerged slag or not can be judged.

The following description of the control method for intelligent tapping of the converter by a specific example is as follows:

(1) the converter finishes converting, after sampling is qualified, intelligent tapping is started by one key, and a tapping monitoring module and a converter mouth monitoring module are started to work;

(2) the deep learning host 6 informs the communication control module to send a steel tapping signal to the converter steel-making secondary control system 4 according to the intelligent steel tapping initial model, and the converter steel-making secondary control system 4 tilts the converter to a preset initial steel tapping position to start steel tapping;

(3) the tapping monitoring module monitors tapping steel flow, preprocessing tapping steel flow image information and then sending the preprocessed tapping steel flow image information to the deep learning host 6, and the deep learning host 6 analyzes the shape, position, width, tapping time and the like of the steel flow in the image through a YOLO neural network model; an exemplary procedure is as follows: first, the width and height of the image are adjusted. Each image is represented by a matrix of pixel values, which may be stacked in rows or columns into individual long vectors, and the simplest way to calculate the image gradient is to calculate the difference of the image along the horizontal (X) and vertical (Y) axes separately and then synthesize them into a two-dimensional vector. After removing unnecessary parameters with a vector mask or filter, the YOLO neural network is loaded and input pre-processing is set as needed. After classification, the network gives a probability vector, and through the probability vector, the program can calibrate the steel flow boundary position, the slag entrapment boundary position and the like. And analyzing the steel flow form through the probability vector, counting time equivalence of steel tapping, and carrying out subsequent processing on the converted information, comparing the converted information with the stored values in the model database and judging whether the converted information is abnormal or not.

(4) After monitoring a furnace mouth picture, a furnace mouth monitoring module preprocesses furnace mouth image information and sends the furnace mouth image information to the deep learning host 6, the deep learning host 6 processes the picture by using a trained neural network model, and analyzes the equivalence of the furnace mouth boundary and the slag liquid surface boundary position so as to judge whether slag is discharged from the furnace mouth or not or whether slag discharge risk exists;

(5) the deep learning host 6 guides the tilting action time, the tilting target angle and the target angle retention time of the steelmaking secondary control system according to the feedback output information of the intelligent tapping neural network model, such as the taphole slag-off state and whether the tapping steel flow is abnormal;

(6) when the converter slag discharging detection system sends a slag discharging alarm signal, the deep learning host 6 synchronously detects the signal, outputs a control instruction to give a steel notch slag stopping system for slag stopping operation, and sends a steel tapping end signal to the steel-making secondary control system 4, and the secondary system returns to the converter according to a preset scheme;

(7) after tapping is finished, the deep learning host 6 inputs the images into a neural network learning model as new samples for iterative training when abnormal targets (slag is about to be discharged from the furnace mouth or discharged from the furnace mouth and slag is rolled up from the steel stream) appear according to the steel stream state change images and the furnace mouth state change images during the current tapping process, so that the subsequent identification accuracy is improved, and the next tapping is waited.

Claims

1. A device for intelligent tapping of a converter comprises a converter steelmaking secondary control system, a converter slag tapping detection system and a tapping hole slag stopping system; the device is characterized by also comprising a furnace mouth monitoring module, a tapping monitoring module, a communication control module and a deep learning host; the communication control module is respectively connected to the deep learning host, the furnace mouth monitoring module, the steel tapping monitoring module, the converter steel-making secondary control system, the converter slag tapping detection system and the steel tapping hole slag stopping system through signal lines, and bidirectional intercommunication of data information and control signals is achieved.

2. The apparatus of claim 1, wherein the fire door monitoring module is composed of a fire door monitoring probe and a fire door monitoring processing unit.

3. The apparatus of claim 1, wherein the tapping monitoring module is comprised of a tapping monitoring probe and a tapping monitoring processing unit.

4. The apparatus of claim 1, wherein the deep learning host is a multi-way GPU integrated learning-type computer host.