CN115187653A - Image recognition-based slag-raking machine control method and system - Google Patents

Image recognition-based slag-raking machine control method and system Download PDF

Info

Publication number
CN115187653A
CN115187653A CN202110354364.XA CN202110354364A CN115187653A CN 115187653 A CN115187653 A CN 115187653A CN 202110354364 A CN202110354364 A CN 202110354364A CN 115187653 A CN115187653 A CN 115187653A
Authority
CN
China
Prior art keywords
slag
raking
image
ladle
liquid level
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202110354364.XA
Other languages
Chinese (zh)
Inventor
张宇军
于艳
孙兴洪
李火中
汪圣军
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Baoshan Iron and Steel Co Ltd
Original Assignee
Baoshan Iron and Steel Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Baoshan Iron and Steel Co Ltd filed Critical Baoshan Iron and Steel Co Ltd
Priority to CN202110354364.XA priority Critical patent/CN115187653A/en
Publication of CN115187653A publication Critical patent/CN115187653A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/70—Determining position or orientation of objects or cameras
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B22—CASTING; POWDER METALLURGY
    • B22D—CASTING OF METALS; CASTING OF OTHER SUBSTANCES BY THE SAME PROCESSES OR DEVICES
    • B22D43/00—Mechanical cleaning, e.g. skimming of molten metals
    • B22D43/005—Removing slag from a molten metal surface
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B22—CASTING; POWDER METALLURGY
    • B22D—CASTING OF METALS; CASTING OF OTHER SUBSTANCES BY THE SAME PROCESSES OR DEVICES
    • B22D46/00—Controlling, supervising, not restricted to casting covered by a single main group, e.g. for safety reasons
    • C—CHEMISTRY; METALLURGY
    • C21—METALLURGY OF IRON
    • C21C—PROCESSING OF PIG-IRON, e.g. REFINING, MANUFACTURE OF WROUGHT-IRON OR STEEL; TREATMENT IN MOLTEN STATE OF FERROUS ALLOYS
    • C21C1/00—Refining of pig-iron; Cast iron
    • C21C1/02—Dephosphorising or desulfurising
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/13—Edge detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/60—Analysis of geometric attributes
    • G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Chemical & Material Sciences (AREA)
  • Mechanical Engineering (AREA)
  • Materials Engineering (AREA)
  • Metallurgy (AREA)
  • Organic Chemistry (AREA)
  • Geometry (AREA)
  • Furnace Charging Or Discharging (AREA)

Abstract

The invention discloses a control method of a slag raking machine based on image recognition, which comprises the following steps: acquiring a liquid level image of the foundry ladle by adopting a first image acquisition device, identifying a boundary to obtain the distribution of molten iron and slag and the boundary of the foundry ladle, and carrying out region segmentation on the liquid level image of the foundry ladle to obtain a plurality of sub-regions; calculating the slag area of each sub-area in the boundary of the molten iron ladle; constructing and training an improved deep residual shrinkage network, inputting the slag area of each sub-region into the improved deep residual shrinkage network, outputting an optimal slag raking path, and determining the swing angle of a slag raking arm and the expansion amount of a slag raking head; acquiring a liquid level of molten iron and a boundary image of the bottom of the slag raking head by adopting a second image acquisition device, and acquiring the lifting amount of the slag raking arm and the inclination angle of the slag raking arm according to the image; and controlling the action of the slag raking machine based on the swing angle of the slag raking arm, the telescopic amount of the slag raking head, the lifting amount of the slag raking arm and the inclination angle of the slag raking arm. In addition, the invention also discloses a control system of the slag raking machine, which is used for implementing the method.

Description

Image recognition-based slag raking machine control method and system
Technical Field
The invention relates to a molten iron pretreatment method and a molten iron pretreatment system, in particular to a molten iron pretreatment slag skimming method and a molten iron pretreatment slag skimming system.
Background
In the field of steel making in the metallurgical industry, a ladle KR mechanical stirring method is a preferred process technology for the existing desulphurization process of blast furnace molten iron pretreatment.
In the process of desulfurization, sulfur-containing slag generated by KR mechanical stirring can be removed in a slag skimming mode, and currently, the advanced KR slag skimming system at home and abroad generally skims slag manually, namely, an operator carries out remote manual slag skimming in a remote control room.
In the prior art, the remote manual slag-raking method can realize slag-raking on the premise of protecting the safety of operators. However, the process means depends on manual work, the production cost is high, the production efficiency is low, the process means has high requirements on the professional operation skills of operators, the difference of the slag raking time of different operators is large, the slag raking effect and quality are different, and the remote manual slag raking is greatly influenced by human factors. Therefore, users and the market urgently need an automatic slag-raking method to realize automatic slag-raking.
At present, some researchers have developed an automatic slag skimming system based on image recognition, and the system firstly reads a real-time image of a camera through an image recognition processing module to obtain information such as a ladle wall constraint range, slag distribution, molten iron distribution and the like; secondly, a path planning module is applied, under the constraint of a certain slag-raking strategy, a slag-raking path which is large in slag-raking amount, short in slag-raking time and meets the constraint condition of collision avoidance of a slag-raking head and a ladle is obtained through an optimization algorithm, the path is further converted into coordinate data of the slag-raking head, and then the coordinate data are transmitted to a PLC (programmable logic controller) for real-time control; and finally, when the slag amount is less than a certain threshold value, the slag skimming end point is reached, and the slag skimming is stopped.
In the automatic slag-raking technology based on image recognition, the key point is the automation and intellectualization of determining 4 control parameters of the slag-raking machine (including the swing angle of a slag-raking arm, the elongation of a slag-raking head, the up-down displacement of the slag-raking arm and the inclination angle of the slag-raking arm). However, in the prior art, the automatic slag skimming technique cannot achieve a good automatic slag skimming effect, and cannot automatically, real-timely and accurately measure the four control parameters.
Based on the above, aiming at the defects and shortcomings of the slag-raking process in the prior art, in order to further reduce cost and improve efficiency, the invention is expected to obtain the control method and the control system of the slag-raking machine based on image recognition. By adopting the control method and the system of the slag raking machine, intelligent slag raking can be completed without manual intervention, so that the production efficiency can be effectively improved, the production cost can be reduced, and the physical and psychological health of field operators can be guaranteed.
Disclosure of Invention
One of the purposes of the invention is to provide a control method of the slag-raking machine based on image recognition, the control method of the slag-raking machine is convenient to operate and simple in flow, automatic slag raking can be realized based on the image recognition, the image recognition accuracy is high, and the operation control parameters of the slag-raking machine can be accurately obtained. By adopting the control method of the slag raking machine, intelligent slag raking can be completed without manual intervention, the production efficiency can be effectively improved, the production cost can be reduced, and the physical and psychological health of field operators can be guaranteed.
In order to achieve the purpose, the invention provides a control method of a slag raking machine based on image recognition, which comprises the following steps:
(1) Acquiring a liquid level image of the ladle by adopting a first image acquisition device positioned above the ladle;
(2) Carrying out boundary identification on the liquid level image of the foundry ladle to obtain the distribution of molten iron and slag and the boundary of the foundry ladle; the image of the liquid level of the ladle is subjected to region segmentation to obtain a plurality of sub-regions;
(3) Calculating the slag area of each subregion in the ladle boundary;
(4) Constructing and training an improved depth residual shrinkage network for inputting the slag area of each sub-region into the improved depth residual shrinkage network, wherein the improved depth residual shrinkage network outputs an optimal slag skimming path;
(5) Determining the swing angle of the slag-raking arm and the telescopic amount of the slag-raking head based on the optimal slag-raking path;
(6) Acquiring the liquid level of molten iron and a boundary image of the bottom of the slag-raking head by adopting a second image acquisition device positioned on the side surface of the ladle, and acquiring the lifting amount of the slag-raking arm and the inclination angle of the slag-raking arm based on the liquid level of the molten iron and the boundary image of the bottom of the slag-raking head;
(7) And controlling the slag raking action of the slag raking machine based on the swing angle of the slag raking arm, the telescopic amount of the slag raking head, the lifting amount of the slag raking arm and the inclination angle of the slag raking arm.
In the technical scheme of the invention, the key of the control method of the slag-raking machine is the automation and intellectualization of determining 4 control parameters of the slag-raking machine (including the swing angle of a slag-raking arm, the elongation of a slag-raking head, the up-down displacement of the slag-raking arm and the inclination angle of the slag-raking arm).
In the invention, the swing angle of the slag-raking arm and the elongation of the slag-raking head can be identified and measured by the first image acquisition device above the ladle, and the optimal slag-raking path of slag-raking in the horizontal plane can be calculated by carrying out boundary identification and slag amount calculation on the ladle liquid level image acquired by the first image acquisition device, so that the swing angle of the slag-raking arm and the expansion amount of the slag-raking head at each point on the optimal slag-raking path are determined. The lifting amount and the inclination angle of the slag raking arm can be identified and measured by a second image acquisition device on the side surface of the ladle, and are obtained by the acquired molten iron liquid level and the boundary image of the bottom of the slag raking head.
It should be noted that, in the present embodiment, an improved Deep Residual Shrinkage network (MDRSN) is constructed and trained, and the slag area of each sub-region may be output based on the improved Deep Residual Shrinkage network to an optimal drossing path. Among these, MDRSN is an integration of a deep residual network, an attention mechanism, and an improved soft threshold function.
The MDRSN can eliminate the redundant features by adopting a special soft thresholding threshold value according to the characteristics of each image identification sample. The MDRSN can identify the noise characteristics through an attention mechanism, and set the noise characteristics to be zero through a soft threshold function; meanwhile, important features are reserved through an attention mechanism, so that the capability of a deep neural network for extracting useful features from a noise-containing signal is enhanced; when slag and molten iron classification is carried out in a slag skimming area, some noises including dust environment noises, gaussian noises and the like or abnormal information such as faults of a slag skimming machine exists inevitably in an image sample. This information (noise) may adversely affect the classification effect.
In the invention, the improved depth residual shrinkage network (MDRSN) can adopt a special soft thresholding model for noise reduction treatment; from the aspect of deep learning, the features corresponding to the noise should be deleted inside the deep neural network to avoid affecting the image recognition effect; even in the same image sample set, the amount of noise often varies from sample to sample. The positions of the molten iron slag in the pictures can be different, and the attention mechanism can identify the position of the slag for each picture; in addition, when the molten iron and slag classifier is trained, images labeled as "slag" may be interfered by a slag raking arm and smoke dust, resulting in a decrease in classification accuracy. Therefore, in the improved deep residual shrinkage network (MDRSN), an attention mechanism is adopted to identify the interference of smoke dust and the shadow area of the slag raking arm, so that the accuracy of the slag and molten iron classifier is effectively improved.
Further, in the image recognition-based slag removal machine control method, in the step (2), the boundary recognition is performed on the ladle liquid level image by adopting image threshold processing and image enhancement.
In the technical scheme of the invention, the boundary identification is carried out on the ladle liquid level image by adopting the traditional image processing method, including image threshold processing, image enhancement and the like, so as to obtain the distribution of molten iron and slag and the boundary with the ladle.
Correspondingly, in the step (2) of the slag-raking machine control method, the automatic smoke and dust identification processing can be carried out on the ladle liquid level image, and the area is divided to obtain a plurality of sub-areas.
It should be noted that, in some embodiments, the area may be divided into a uniform fine grid with a length (50-1920) and a width (10-1280), which is intended to prepare data for a deep learning algorithm, each characteristic channel in the deep learning is a divided sub-area (minimum single pixel), then the slag area of each sub-area in the ladle boundary may be calculated in the subsequent step (3), and may be corrected by the slag color, for example, for a gray scale, the slag color may be similar to the thickness of slag, and then the total slag amount and the slag removal path are respectively labeled by experts. Wherein, the total slag amount is used for judging the end point slag, namely the automatic slag-raking process is terminated when the real-time surface residual slag amount is less than the set end point slag amount; the slag raking path is a plurality of preset optimized paths (from experience of industry experts), the industry experts mark according to actual slag and molten iron distribution, the deep learning network automatically trains an optimal path, each path is composed of 3-10 points, each point corresponds to a certain pixel point on an image, and the image calibration or reference comparison method is adopted to correspond to the horizontal and vertical coordinates of the actual slag raking head one by one. Finally, the image identification technology adopting the improved depth residual shrinkage network (MDRSN) in the step (4) can output the optimal slag-off path for the slag area of each sub-area. The application of the deep residual error network aims to solve the problem of degradation of the deep network and avoid gradient disappearance or gradient explosion phenomena, and is used for training and learning to determine the total slag amount and the optimal path number corresponding to the slag area distribution and the color distribution, while the attention mechanism automatically learns a group of threshold values by adding a small network in the deep residual error network, the small network obtains a threshold value for each characteristic channel, the threshold value is equivalent to the weight of each characteristic, and the weight determines whether the characteristic is emphasized or not, or the characteristic is reserved or deleted and the like; the invention adopts an attention mechanism to weaken the adverse effects of abnormal noises such as dust environment noise, the influence of the shadow area of a slag-raking arm and the like on slag quantity identification and path identification, and in addition, a soft threshold function processing method is improved for shielding the abnormal noises.
Further, in the image recognition-based crawler loader control method according to the present invention, in step (4), the improved soft threshold function S (x, T) of the improved depth residual shrinkage network is:
Figure BDA0003003231140000041
wherein x represents the slag area of the sub-region; t denotes a threshold value automatically learned by the attention mechanism; alpha denotes a correction factor.
It should be noted that the invention can modify the threshold size of the conventional soft threshold function through actual field experience fitting data to improve the classification accuracy of slag amount identification and path identification under the determined environment.
Further, in the image recognition-based control method for the crawler loader according to the present invention, the correction factor is obtained based on the following formula:
α＝β 0 +β 1 ×X+β 2 ×X 2 +β 3 ×X 3
wherein X represents the mean gray value of the sub-region; beta is a 0 、β 1 、β 2 、β 3 All represent fitting coefficients, where 0 < beta 0 ≤10，0＜β 1 ≤5，0＜β 2 ≤3，0＜β 3 ≤3。
Further, in the image recognition-based control method of the slag raking machine, the improved deep residual error shrinkage network outputs the optimal slag raking path in the form of a tag index number.
Further, in the image recognition-based control method for the slag raking machine, in the step (5), an image calibration method or a reference comparison method is adopted to enable the abscissa and the ordinate of the point on the optimal slag raking path to correspond to the swing angle of the slag raking arm and the telescopic amount of the slag raking head one by one, so that the swing angle of the slag raking arm and the telescopic amount of the slag raking head are determined.
Further, in the image recognition-based control method for the slag raking machine, in the step (6), obtaining the lifting amount of the slag raking arm based on the molten iron liquid level and the boundary image of the bottom of the slag raking head comprises the following steps: preprocessing the liquid level of the molten iron and the boundary image of the bottom of the slag raking head; performing region segmentation on the preprocessed image to obtain an identification subregion; adopting different identification thresholds to identify the lower boundary of the slagging-off head and the molten iron liquid level position in each identification subarea; and obtaining the lifting amount of the slag raking arm by adopting an image calibration method.
In the technical scheme of the invention, the control of the lifting amount of the slag-raking arm is the insertion depth of the slag-raking head in molten iron, the slag-raking iron loss is easily increased due to too deep insertion of the slag-raking head into the molten iron, and the removal efficiency is too low due to too shallow insertion of the slag-raking head into the molten iron, so that the up-and-down displacement of the slag-raking arm needs to be accurately determined. Meanwhile, the parameter and the slag raking and hardening height are closely related to parameters such as the liquid level height of the ladle and the like, particularly, the slag raking and hardening height is a random value, the melting loss and the slagging state of a slag raking plate change greatly along with different working conditions, and the height of the slag raking plate can be increased or decreased, so that the slag raking and hardening height is difficult to accurately calculate only by a direct calculation method; the method comprises the steps of dividing an image into different identification areas by a digital image processing technology, including methods of binarization, image enhancement, morphological processing and the like, adopting a sub-area identification technology, respectively identifying the lower boundary of a slag raking head and the position of the liquid level of molten iron in different sub-areas by adopting different identification thresholds, and obtaining lifting data of a slag raking arm through image calibration so as to achieve the purpose of intelligent control.
Further, in the image recognition-based control method for the slag raking machine, in the step (6), obtaining the inclination angle of the slag raking arm based on the molten iron liquid level and the boundary image of the bottom of the slag raking head comprises the following steps: acquiring the height difference of the slag raking head at the near end of the standby position and the far end of the standby position based on the molten iron liquid level and the boundary image of the bottom of the slag raking head; obtaining a center line of the slag raking arm based on the height difference; and obtaining the inclination angle of the slag raking arm based on the included angle between the central line of the slag raking arm and the horizontal line.
In the technical scheme of the invention, the preferable scheme of the inclination angle of the slag-raking arm is that the slag-raking arm is set to be kept horizontal in the slag-raking operation, the insertion depth of a slag-raking head is controlled only by adjusting the up-down displacement of the slag-raking arm, and the horizontal control of the slag-raking arm cannot realize the feedback control of the inclination angle of the slag-raking arm because the slag-raking machine cannot realize the inclination angle of the slag-raking arm, therefore, the invention adopts an image recognition method to carry out calibration control, namely, the height information of the slag-raking arm on an image is measured by a machine vision recognition algorithm, if the heights are consistent, the horizontal is considered to be kept, and if the heights are inconsistent, the inclination angle horizontal value of the slag-raking arm is adjusted and set according to the comparison between the height difference of the image and the actual height difference.
Accordingly, another object of the present invention is to provide a control system for a slag raking machine based on image recognition, which can perform intelligent slag raking without manual intervention when the control system is used for slag raking, and which can effectively replace manual operations in the existing slag raking process, greatly shorten the slag raking time, improve the production efficiency, and facilitate the reduction of the production cost.
In order to achieve the above object, the present invention provides an image recognition-based control system for a slag-raking machine, comprising:
the first image acquisition device is arranged above the ladle and acquires images of the liquid level of the ladle;
the second image acquisition device is arranged on the side surface of the ladle and acquires the liquid level of the molten iron and a boundary image of the bottom of the slag-raking head;
a control module that performs the steps of:
carrying out boundary identification on the ladle liquid level image to obtain the distribution of molten iron and slag and the ladle boundary, and carrying out region segmentation on the ladle liquid level image to obtain a plurality of sub-regions;
calculating the slag area of each subregion in the ladle boundary;
constructing and training an improved depth residual shrinkage network for inputting the slag area of each sub-region into the improved depth residual shrinkage network, wherein the improved depth residual shrinkage network outputs an optimal slag skimming path;
determining the swing angle of the slag-raking arm and the telescopic amount of the slag-raking head based on the optimal slag-raking path;
acquiring the lifting amount and the inclination angle of the slag-raking arm based on the liquid level of molten iron and a boundary image of the bottom of the slag-raking head;
and controlling the slag raking action of the slag raking machine based on the swing angle of the slag raking arm, the telescopic amount of the slag raking head, the lifting amount of the slag raking arm and the inclination angle of the slag raking arm.
Further, in the control system of the crawler loader according to the present invention, the improved soft threshold function S (x, T) of the improved deep residual shrinkage network is:
Figure BDA0003003231140000071
wherein x represents the slag area of the sub-region; t denotes a threshold value automatically learned by the attention mechanism; alpha denotes a correction factor.
Further, in the crawler loader control system according to the present invention, the correction factor is obtained based on the following formula:
α＝β 0 +β 1 ×X+β 2 ×X 2 +β 3 ×X 3
wherein X represents the mean gray value of the sub-region; beta is a 0 、β 1 、β 2 、β 3 Uniform meterShowing fitting coefficients, where 0 < beta 0 ≤10，0＜β 1 ≤5，0＜β 2 ≤3，0＜β 3 ≤3。
Compared with the prior art, the image recognition-based crawler loader control method and system have the following advantages and beneficial effects:
(1) The image recognition-based control method of the slag raking machine can acquire all control parameters of the slag raking machine in real time by adopting a multi-machine vision integration method, so that the intelligent control level can be effectively improved.
(2) The image recognition-based control method of the slag raking machine adopts an image recognition technology based on an improved deep residual error shrinkage network (MDRSN) aiming at slag amount recognition and optimal slag raking path recognition, and can effectively improve the recognition and classification accuracy under the actual metallurgical high-temperature dust environment.
(3) According to the image recognition and measurement method for the slagging-off height of the slagging-off plate based on the image recognition control method, the distance parameter between the slagging-off arm and the slagging-off plate can be determined by recognizing the position of the horizontal line of the slagging-off arm in the image and recognizing the position of the bottom edge line of the slagging-off plate in the image, so that the slagging-off height of the slagging-off plate can be calculated; the image identification and measurement method of the liquid level height of the foundry ladle is to identify the position of the liquid level of the foundry ladle in an image and compare the position with a standard height reference value to determine the liquid level height; the distance between the slag skimming plate and the liquid level of the ladle is determined by identifying the position of the bottom of the slag skimming plate in the image and the position of the liquid level of the molten iron in the image.
In conclusion, the image recognition-based control method of the slag raking machine is convenient to operate and simple in process, automatic slag raking can be achieved based on image recognition, the image recognition accuracy is high, and the operation control parameters of the slag raking machine can be accurately obtained. By adopting the control method of the slag raking machine, intelligent slag raking can be completed without manual intervention, the production efficiency can be effectively improved, the production cost can be reduced, and the physical and psychological health of field operators can be guaranteed.
Accordingly, the image recognition-based crawler loader control system can be used for implementing the crawler loader control method, and has the advantages and beneficial effects.
Drawings
Fig. 1 schematically shows a process flow diagram of a crawler loader control method based on image recognition according to an embodiment of the present invention.
Detailed Description
The image recognition-based crawler loader control method and system according to the present invention will be further described with reference to the following detailed embodiments of the invention and the accompanying drawings, but the description is not intended to limit the invention.
In the invention, the slag-raking machine control system based on image recognition can comprise: the device comprises a first image acquisition device, a second image acquisition device and a control device. The first image acquisition device is arranged above the ladle and can acquire images of the liquid level of the ladle; the second image acquisition device is arranged on the side face of the ladle and can acquire the liquid level of molten iron and the boundary image of the bottom of the slag-raking head.
It should be noted that, in the system of the present invention, the control module may perform boundary identification on the ladle liquid level image to obtain the distribution of molten iron and slag and the ladle boundary, and perform region segmentation on the ladle liquid level image to obtain a plurality of sub-regions; then calculating the slag area of each subregion in the molten iron ladle boundary; constructing and training an improved deep residual shrinkage network (MDRSN) for inputting the slag area of each sub-region into the improved deep residual shrinkage network (MDRSN), wherein the improved deep residual shrinkage network (MDRSN) outputs an optimal slag-off path; then, the swing angle of the slag-raking arm and the stretching amount of the slag-raking head can be determined based on the optimal slag-raking path, and the lifting amount of the slag-raking arm and the inclination angle of the slag-raking arm can be obtained based on the molten iron liquid level and the boundary image of the bottom of the slag-raking head; the slag raking action of the slag raking machine can be controlled based on the swing angle of the slag raking arm, the telescopic amount of the slag raking head, the lifting amount of the slag raking arm and the inclination angle of the slag raking arm.
In the invention, the image recognition-based slag-raking machine control system can be used for implementing the image recognition-based slag-raking machine control method.
In order to further explain the implementation of the control system of the crawler loader, the invention adopts a specific embodiment, and in this embodiment, the process flow of the control method of the crawler loader implemented by the system can be as shown in the following fig. 1.
Fig. 1 schematically shows a process flow diagram of a crawler loader control method based on image recognition according to an embodiment of the present invention.
In this embodiment, the first image capturing device in the system for executing the slag-raking machine control method according to the present invention may be disposed at a position 30 ° above the ladle at an angle and 5 m away from the ladle, and may capture a ladle liquid level image, and perform real-time measurement and recognition by using an image boundary recognition method and a slag amount recognition method on the ladle liquid level image captured here.
Correspondingly, the second image acquisition device is arranged at the lateral position between the ladle and the slag raking machine and can acquire the liquid level of the molten iron and the boundary image of the bottom of the slag raking head. In this embodiment, the first image capturing device and the second image capturing device may be selected as cameras to capture images.
As shown in fig. 1, in this embodiment, the crawler loader control method implemented by the system of the present invention may include the following steps:
(1) And collecting the liquid level image of the ladle by adopting a first image collecting device positioned above the ladle.
(2) Carrying out boundary identification on the liquid level image of the foundry ladle to obtain the distribution of molten iron and slag and the boundary of the foundry ladle; and the ladle level image is subjected to region segmentation to obtain a plurality of sub-regions.
In the step (2) of the present invention, the boundary recognition of the ladle level image may be performed by image thresholding and image enhancement.
(3) And calculating the slag area of each sub-area in the molten iron ladle boundary.
(4) And constructing and training an improved depth residual shrinkage network for inputting the slag area of each sub-region into the improved depth residual shrinkage network, wherein the improved depth residual shrinkage network outputs an optimal slag skimming path.
In the above step (4) of the present invention, in the modified depth residual puncturing network (MDRSN), the modified soft threshold function S (x, T) can be expressed as:
Figure BDA0003003231140000091
wherein x represents the slag area of the sub-region; t denotes a threshold value automatically learned by the attention mechanism; alpha denotes a correction factor. Accordingly, the correction factor α may be further obtained based on the following formula:
α＝β 0 +β 1 ×X+β 2 ×X 2 +β 3 ×X 3
in the above formula, X represents the average gray value of the sub-region; beta is a beta 0 、β 1 、β 2 、β 3 All represent fitting coefficients, where 0 < beta 0 ≤10，0＜β 1 ≤5，0＜β 2 ≤3，0＜β 3 ≤3。
It should be noted that, in this embodiment, the threshold size of the soft threshold function may be modified and improved through actual field experience fitting data, so that the classification accuracy of optimal drossing path identification and slag amount identification performed by the improved deep residual shrinkage network (MDRSN) in a certain environment is effectively improved, and the improved deep residual shrinkage network (MDRSN) may output the optimal drossing path in the form of a tag index number.
(5) And determining the swing angle of the slag-raking arm and the telescopic amount of the slag-raking head based on the optimal slag-raking path.
In the step (5) of the invention, an image calibration method or a reference comparison method can be further adopted to respectively correspond the abscissa and the ordinate of the point on the optimal slag-raking path to the swing angle of the slag-raking arm and the stretching amount of the slag-raking head one by one so as to determine the swing angle of the slag-raking arm and the stretching amount of the slag-raking head.
(6) And acquiring the liquid level of the molten iron and the bottom boundary image of the slag-raking head by adopting a second image acquisition device positioned on the side surface of the molten iron ladle, and acquiring the lifting amount of the slag-raking arm and the inclination angle of the slag-raking arm based on the liquid level of the molten iron and the bottom boundary image of the slag-raking head.
In the step (6), the process of obtaining the lifting amount of the slag-raking arm based on the molten iron liquid level and the boundary image of the bottom of the slag-raking head may include the steps of: preprocessing a boundary image of the liquid level of molten iron and the bottom of a slag raking head; performing region segmentation on the preprocessed image to obtain an identification subregion; adopting different identification thresholds to identify the lower boundary of the slagging-off head and the molten iron liquid level position in each identification subarea; and obtaining the lifting amount of the slag raking arm by adopting an image calibration method.
Correspondingly, the process of obtaining the inclination angle of the slag-raking arm based on the molten iron liquid level and the boundary image of the bottom of the slag-raking head comprises the following steps: the method for obtaining the inclination angle of the slag-raking arm based on the molten iron liquid level and the boundary image of the bottom of the slag-raking head comprises the following steps: acquiring the height difference of the slag raking head at the near end of the standby position and the far end of the standby position based on the molten iron liquid level and the boundary image of the bottom of the slag raking head; obtaining a center line of the slag raking arm based on the height difference; and obtaining the inclination angle of the slag raking arm based on the included angle between the central line of the slag raking arm and the horizontal line.
(7) And controlling the slag raking action of the slag raking machine based on the swing angle of the slag raking arm, the telescopic amount of the slag raking head, the lifting amount of the slag raking arm and the inclination angle of the slag raking arm.
When the real-time control method of the slag-raking machine is adopted, in the present embodiment, when the image of the liquid level of the ladle is divided into regions in the step (2), the image mesh may be divided into 100 parts by length and 50 parts by width to obtain 5000 sub-regions, 8000 training images and 2500 test images.
Accordingly, in the present embodiment, the above step (4) needs to construct and train an improved depth residual shrinkage network (MDRSN), wherein the fitting coefficient β of the improved soft threshold function 0 May be 0.15, fitting coefficient beta 1 May be 1.2, fitting coefficient beta 2 May be 0.3, fitting coefficient beta 3 May be 0.1.
In the embodiment, after the positions of the first image acquisition device and the second image acquisition device are determined, and the improved deep residual shrinkage network (MDRSN) -based construction and training are completed, the slag removing machine control system can be used for effectively controlling slag removing of the slag removing machine, and the real-time control power of the slag removing machine control system is more than 95%.
In conclusion, the image recognition-based control method for the slag removing machine is convenient to operate and simple in flow, automatic slag removing can be achieved based on image recognition, image recognition accuracy is high, and operation control parameters of the slag removing machine can be accurately obtained. By adopting the control method of the slag raking machine, intelligent slag raking can be completed without manual intervention, the production efficiency can be effectively improved, the production cost can be reduced, and the physical and psychological health of field operators can be guaranteed.
Accordingly, the image recognition-based crawler loader control system can be used for implementing the crawler loader control method, and has the advantages and beneficial effects.
It should be noted that the prior art in the protection scope of the present invention is not limited to the examples given in the specification, and all the prior art which is not inconsistent with the technical solution of the present invention, including but not limited to the prior patent documents, the prior publications and the like, can be included in the protection scope of the present invention.
In addition, the combination of the features in the present application is not limited to the combination described in the claims of the present application or the combination described in the embodiments, and all the features described in the present application may be freely combined or combined in any manner unless contradictory to each other.
It should also be noted that the above-mentioned embodiments are only specific embodiments of the present invention. It is apparent that the present invention is not limited to the above embodiments and similar changes or modifications can be easily made by those skilled in the art from the disclosure of the present invention and shall fall within the scope of the present invention.

Claims (11)

1. A control method of a slag-raking machine based on image recognition is characterized by comprising the following steps:
(1) Acquiring a liquid level image of the ladle by adopting a first image acquisition device positioned above the ladle;
(2) Carrying out boundary identification on the ladle liquid level image to obtain the distribution of molten iron and slag and the ladle boundary; the image of the liquid level of the ladle is subjected to region segmentation to obtain a plurality of sub-regions;
(3) Calculating the slag area of each subregion in the ladle boundary;
(4) Constructing and training an improved depth residual shrinkage network for inputting the slag area of each sub-region into the improved depth residual shrinkage network, wherein the improved depth residual shrinkage network outputs an optimal slag skimming path;
(5) Determining the swing angle of the slag-raking arm and the telescopic amount of the slag-raking head based on the optimal slag-raking path;
(6) Acquiring a molten iron liquid level and a slag raking head bottom boundary image by adopting a second image acquisition device positioned on the side surface of the ladle, and acquiring a lifting amount of a slag raking arm and an inclination angle of the slag raking arm based on the molten iron liquid level and the slag raking head bottom boundary image;
(7) And controlling the slag raking action of the slag raking machine based on the swing angle of the slag raking arm, the telescopic amount of the slag raking head, the lifting amount of the slag raking arm and the inclination angle of the slag raking arm.
2. The image recognition-based slag crawler control method according to claim 1, wherein in the step (2), the image threshold processing and the image enhancement are adopted to perform the boundary recognition on the ladle liquid level image.
3. The image recognition-based crawler loader control method according to claim 1, wherein in step (4) the improved soft threshold function S (x, T) of the improved depth residual shrinkage network is:
Figure FDA0003003231130000011
wherein x represents the slag area of the sub-region; t denotes a threshold value automatically learned by the attention mechanism; alpha denotes a correction factor.
4. The image recognition-based crawler loader control method according to claim 3, wherein said correction factor is obtained based on the following formula:
α＝β 0 +β 1 ×X+β 2 ×X 2 +β 3 ×X 3
wherein X represents the mean gray value of the sub-region; beta is a beta 0 、β 1 、β 2 、β 3 All represent fitting coefficients, where 0 < beta 0 ≤10，0＜β 1 ≤5，0＜β 2 ≤3，0＜β 3 ≤3。
5. The image recognition-based crawler loader control method according to claim 1, wherein said improved deep residual shrinkage network outputs an optimal crawler loader path in the form of a tag index.
6. The image recognition-based slag-raking machine control method according to claim 1, wherein in the step (5), an image calibration method or a reference comparison method is adopted to correspond the abscissa and the ordinate of the point on the optimal slag-raking path to the swing angle of the slag-raking arm and the telescopic amount of the slag-raking head in a one-to-one manner, so as to determine the swing angle of the slag-raking arm and the telescopic amount of the slag-raking head.
7. The image recognition-based crawler loader control method according to claim 1, wherein in the step (6), obtaining the lifting amount of the crawler arm based on the molten iron level and the crawler head bottom boundary image comprises the steps of: preprocessing a boundary image of the liquid level of molten iron and the bottom of a slag raking head; performing region segmentation on the preprocessed image to obtain an identification subregion; adopting different identification thresholds to identify the lower boundary of the slagging-off head and the molten iron liquid level position in each identification subarea; and obtaining the lifting amount of the slag raking arm by adopting an image calibration method.
8. The image recognition-based slag-raking machine control method according to claim 1, wherein in the step (6), the step of obtaining the inclination angle of the slag-raking arm based on the liquid level of the molten iron and the boundary image of the bottom of the slag-raking head comprises the steps of: acquiring the height difference of the slag raking head at the near end of the standby position and the far end of the standby position based on the molten iron liquid level and the boundary image of the bottom of the slag raking head; obtaining a center line of the slag raking arm based on the height difference; and obtaining the inclination angle of the slag raking arm based on the included angle between the central line of the slag raking arm and the horizontal line.
9. A crawler loader control system based on image recognition is characterized by comprising:
the first image acquisition device is arranged above the ladle and acquires images of the liquid level of the ladle;
the second image acquisition device is arranged on the side surface of the ladle and acquires the liquid level of the molten iron and a boundary image of the bottom of the slag-raking head;
a control module that performs the steps of:
carrying out boundary identification on the liquid level image of the foundry ladle to obtain the distribution of molten iron and slag and the boundary of the foundry ladle, and carrying out region segmentation on the liquid level image of the foundry ladle to obtain a plurality of sub-regions;
calculating the slag area of each sub-area in the boundary of the molten iron ladle;
constructing and training an improved depth residual shrinkage network for inputting the slag area of each sub-region into the improved depth residual shrinkage network, wherein the improved depth residual shrinkage network outputs an optimal slag-raking path;
determining the swing angle of the slag-raking arm and the telescopic amount of the slag-raking head based on the optimal slag-raking path;
acquiring the lifting amount of the slag-raking arm and the inclination angle of the slag-raking arm based on the liquid level of molten iron and the boundary image of the bottom of the slag-raking head;
and controlling the slag raking action of the slag raking machine based on the swing angle of the slag raking arm, the telescopic amount of the slag raking head, the lifting amount of the slag raking arm and the inclination angle of the slag raking arm.
10. The crawler control system of claim 9, wherein the improved soft threshold function S (x, T) of the improved deep residual shrinkage network is:
Figure FDA0003003231130000031
wherein x represents the slag area of the sub-region; t denotes a threshold value automatically learned by the attention mechanism; alpha denotes a correction factor.
11. The crawler loader control system of claim 10, wherein said correction factor is obtained based on the following equation:
α＝β 0 +β 1 ×X+β 2 ×X 2 +β 3 ×X 3
wherein X represents the mean gray value of the sub-region; beta is a 0 、β 1 、β 2 、β 3 All represent fitting coefficients, where 0 < beta 0 ≤10，0＜β 1 ≤5，0＜β 2 ≤3，0＜β 3 ≤3。
CN202110354364.XA 2021-04-01 2021-04-01 Image recognition-based slag-raking machine control method and system Pending CN115187653A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110354364.XA CN115187653A (en) 2021-04-01 2021-04-01 Image recognition-based slag-raking machine control method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110354364.XA CN115187653A (en) 2021-04-01 2021-04-01 Image recognition-based slag-raking machine control method and system

Publications (1)

Publication Number Publication Date
CN115187653A true CN115187653A (en) 2022-10-14

Family

ID=83512034

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110354364.XA Pending CN115187653A (en) 2021-04-01 2021-04-01 Image recognition-based slag-raking machine control method and system

Country Status (1)

Country Link
CN (1) CN115187653A (en)

Similar Documents

Publication Publication Date Title
CN112017145B (en) Efficient automatic slag skimming method and system for molten iron pretreatment
CN110413013B (en) Intelligent argon blowing system and control method thereof
CN112233133B (en) Power plant high-temperature pipeline defect detection and segmentation method based on OTSU and area growth method
CN106881462A (en) A kind of on-line checking for selective laser fusing forming defects and optimization system
CN115100191B (en) Metal casting defect identification method based on industrial detection
CN110747306A (en) Method, device and equipment for controlling slag overflow in converter tapping process and storage medium
CN114897908B (en) Machine vision-based method and system for analyzing defects of selective laser powder spreading sintering surface
CN111126206B (en) Smelting state detection system and method based on deep learning
CN110751669A (en) Novel CBOCP online infrared converter tapping steel flow automatic detection and tracking method and system
CN104392213A (en) Image information state recognizing system applicable to melting process
CN113838114B (en) Blast furnace burden surface depth estimation method and system based on edge defocus tracking
CN115187653A (en) Image recognition-based slag-raking machine control method and system
CN112091206B (en) Safe and reliable molten iron pretreatment automatic slag skimming method and system
CN105047239B (en) Nuclear fuel assembly repairs tracking detection method and device
CN113063474A (en) Slag liquid level real-time detection method, device, equipment and storage medium
CN103302253A (en) Liquid level detection method and system
CN101294945A (en) White edge detecting method for hot galvanizing alloying plate
CN114932292A (en) Narrow-gap passive visual weld joint tracking method and system
CN113467437B (en) Optimization method of KR automatic slag skimming intelligent path
CN113538319B (en) Slag amount calculation method based on gray scale ratio of slag skimming image
CN115612765B (en) Real-time detection control method and system for blast furnace tapping state
CN115464661B (en) Robot control system based on digital twins
CN110438284B (en) Intelligent tapping device of converter and control method
CN117187942B (en) Crucible position control method and device in crystal pulling process
CN116162762A (en) Method and system for adjusting parameters of slag removing arm based on machine vision

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination