Intelligent crust breaking and discharging system based on machine vision

Abstract

The invention provides an intelligent crust breaking and blanking system based on machine vision. The intelligent crust breaking and blanking system comprises an active detection module, an image acquisition module, an image processing and calculating module, a data communication module, a cell control machine, a high-temperature-resistant insulation structure and a cooling device, the active detection module forms a reference line in an aluminum electrolysis cell monitoring area; the image acquisition module is used for respectively acquiring image data of a motion state of a crust breaking hammer and a material falling process of a blanking port; the image processing and calculating module is used for carrying out preprocessing, feature extraction and calculation on the image to obtain a crust breaking depth parameter and blanking flow data; the data communication module transmits a processing result to the cell control machine, and the cell control machine adjusts the crust breaking set depth and optimizes the blanking interval according to the processing result; the high-temperature-resistant insulation structure carries out physical protection and electrical isolation on key modules, and the cooling device keeps the internal temperature stable and prevents the lens from being polluted. Therefore, the crust breaking depth and the blanking quantity control precision can be improved, the maintenance cost is reduced, and the adaptability to a high-temperature strong electric field environment is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of aluminum electrolysis production equipment, and in particular to an intelligent crust breaking and discharging system based on machine vision. BACKGROUND

[0002] In the aluminum electrolysis production process, the precise control of the crust breaking and discharging process directly affects the operating efficiency and energy consumption level of the electrolytic cell. The traditional technical solutions mainly have the following technical bottlenecks: Firstly, in the aspect of crust breaking state monitoring, the existing system generally uses displacement sensors to collect hammer head displacement data, but cannot obtain key parameters such as hammer head wear degree and electrolyte adhesion thickness. The displacement sensor is prone to signal drift in a high-temperature strong magnetic field environment, and its sealing structure is difficult to resist electrolyte splashing corrosion, so it needs to be replaced every 3 months on average, resulting in high maintenance costs.

[0003] Secondly, in the control link of discharging amount, due to the poor flowability of alumina powder, arching phenomenon is easy to occur in the material box, causing the fixed volume discharger to be empty or half full. The current control system can only act according to the preset program, and cannot sense the actual discharging amount, causing systematic deviation between the process calculation value and the actual feeding amount. Some aluminum plants try to use weighing sensors for measurement, but due to the influence of material wall hanging, the measurement error often exceeds 15%, which cannot meet the needs of fine production.

[0004] More prominent problem is that the existing technical solutions lack effective closed-loop control mechanism. The crust breaking depth adjustment and discharging interval setting mainly rely on manual experience, and it is difficult to respond to changes in working conditions such as electrolyte level fluctuation and material consumption rate change in time, which is easy to cause problems such as crust hardening or uneven discharging.

[0005] Therefore, how to improve the control precision of crust breaking depth and discharging amount, reduce maintenance cost, and adapt to high-temperature strong electric field environment is a technical problem to be solved by those skilled in the art. SUMMARY

[0006] The present application provides an intelligent crust breaking and discharging system based on machine vision to improve the control precision of crust breaking depth and discharging amount, reduce maintenance cost, and adapt to high-temperature strong electric field environment.

[0007] In one aspect, the present application provides an intelligent crust breaking and discharging system based on machine vision, which comprises: The intelligent crust breaking and discharging system based on machine vision comprises: An active detection module for forming a reference baseline in the monitoring area of the aluminum electrolytic cell; An image acquisition module for acquiring movement images of the crust breaking hammer head and material images of the discharging port in the aluminum electrolytic cell; An image processing calculation module is configured to pre-process, extract features and calculate data of the crust breaking hammer head movement image and the material image at the discharge port, so as to obtain a depth parameter of the crust breaking hammer head entering the electrolyte and actual flow data of the material at the discharge port. A data communication module is configured to transmit the depth parameter and the actual flow data to a tank control machine. The tank control machine is configured to adjust a crust breaking setting depth according to the depth parameter and optimize a discharge interval according to the actual flow data, so as to form a closed-loop control of the crust breaking and the discharging. A high-temperature-resistant insulation structure is configured to physically protect, resist high temperature and electrically isolate the image acquisition module, the image processing calculation module and the data communication module. A cooling device is configured to reduce the temperature inside the high-temperature-resistant insulation structure and block dust in the aluminum electrolysis tank from adhering to a lens of the image acquisition module.

[0008] According to the intelligent crust breaking and discharging system based on machine vision, the image processing calculation module is specifically configured to: match the crust breaking hammer head movement image with a preset LAB depth threshold template to obtain a first matching result; if the first matching result indicates that there is a coherent electrolyte feature area matched with the LAB depth threshold template, frame the coherent electrolyte feature area by using a minimum bounding rectangle, and calculate the width, height and pixel area of the minimum bounding rectangle; convert the pixel area into an actual depth value of the crust breaking hammer head entering the electrolyte according to a preset conversion relationship between the pixel area and the actual size, wherein the conversion relationship between the pixel area and the actual size is determined by a calibration parameter when the system is installed.

[0009] According to the intelligent crust breaking and discharging system based on machine vision, the image processing calculation module is further configured to: when the system is installed, place a calibration object with a known size at a working position of the crust breaking hammer head, acquire a calibration object image by using the image acquisition module, and establish a conversion relationship between the pixel area and the actual size; acquire an internal temperature of the high-temperature-resistant insulation structure and an external tank temperature of the aluminum electrolysis tank, and calculate a pixel area correction coefficient at the current temperature according to a pre-calibrated temperature and distortion coefficient correlation model; extract a hammer head profile in the crust breaking hammer head movement image, calculate a current diameter of the hammer head, and calculate a hammer head wear amount by subtracting an initial diameter of the hammer head from the current diameter of the hammer head, and determine a depth correction amount according to a preset correlation between the wear amount and the depth correction amount; after the pixel area is corrected by using the pixel area correction coefficient, obtain a preliminary depth according to the conversion relationship between the pixel area and the actual size, and superimpose the depth correction amount to obtain a corrected depth; The correction depth of the continuous multi-frame shell hammer head moving image is corrected, the abnormal value is removed by using the Grubbs criterion, the arithmetic average value of the remaining effective depth is obtained, and the actual depth value is obtained.

[0010] According to the intelligent shell cracking and discharging system based on machine vision provided by the application, the image processing and calculation module is further used for: According to the first preset period, the first identification accuracy of the LAB depth threshold template is counted; If the first identification accuracy in the current period is less than the first preset accuracy for a continuous time length reaching the first preset time length, adaptive learning is triggered; The first preset number of first effective images in the current period is collected, the LAB color space threshold range is recalibrated by using the maximum inter-class variance method, the adjusted LAB depth threshold template is generated, and the trial mode is entered; In the use mode, if the second identification accuracy of the adjusted LAB depth threshold template is greater than or equal to the first preset accuracy, the LAB depth threshold template is replaced; if the second identification accuracy is still less than the first preset accuracy, optimization is repeated until the first stop optimization condition is met.

[0011] According to the intelligent shell cracking and discharging system based on machine vision provided by the application, the image processing and calculation module is further used for: The material image of the discharging port is matched with the preset LAB flow threshold template to obtain a second matching result; If the second matching result indicates that there is a material feature area matched with the LAB flow threshold template, timing is started, and the rectangular pixel area of the first detected material feature area is recorded; The material feature areas of the discharging port are continuously monitored, the pixel areas of all the material feature areas are accumulated, and the timing is stopped when the last material feature area disappears to obtain the total accumulated pixel area and the timing duration; The accumulated pixel area total sum and the timing duration are multiplied by the size conversion coefficient to obtain the discharging flow, and the accumulated pixel area total sum and the size conversion coefficient are multiplied to obtain the total amount of single discharging.

[0012] According to the intelligent shell cracking and discharging system based on machine vision provided by the application, the image processing and calculation module is further used for: According to the second preset period, the third identification accuracy of the LAB flow threshold template is counted; If the third identification accuracy in the current period is less than the second preset accuracy for a continuous time length reaching the second preset time length, adaptive learning is triggered; Collecting a second preset number of second effective images in a current period, and recalibrating the LAB color space threshold range by using the maximum inter-class variance method to generate an adjusted LAB flow threshold template, and entering a trial mode; In the use mode, if a fourth accuracy of the adjusted LAB flow threshold template is greater than or equal to a preset accuracy, the LAB flow threshold template is replaced; if the fourth accuracy is still less than the preset accuracy, optimization is repeated until a second stop optimization condition is met.

[0013] According to the intelligent shell cracking and discharging system based on machine vision provided by the application, the slot control machine is specifically used for: Establishing a correlation model of pixel area and set shell cracking depth; When it is monitored that the pixel area exceeds a preset upper threshold, it is determined that the shell cracking is too deep, an adjustment instruction is sent to the shell cracking execution mechanism, and the set shell cracking depth is decreased according to the gradient; wherein the preset upper threshold is determined by aluminum electrolysis production process parameters; When it is monitored that the pixel area is lower than a preset lower threshold, an adjustment instruction is sent to the shell cracking execution mechanism, and the set shell cracking depth is increased; wherein the preset lower threshold is determined by aluminum electrolysis production process parameters.

[0014] According to the intelligent shell cracking and discharging system based on machine vision provided by the application, the slot control machine is further used for: Real-time collection of aluminum electrolysis production process parameters; the aluminum electrolysis production process parameters include cell temperature, electrolyte molecular ratio, electrolyte level and aluminum level; The cell temperature, electrolyte molecular ratio, electrolyte level and aluminum level are substituted into a preset nonlinear mapping formula to calculate the preset upper threshold and the preset upper threshold.

[0015] According to the intelligent shell cracking and discharging system based on machine vision provided by the application, the slot control machine is further used for: Comparing the discharging flow with a preset first process demand value, and comparing the total discharging amount with a preset second process demand value; If the discharging flow is lower than the first process demand value, or the total discharging amount does not reach the second process demand value, the discharging interval time is shortened; wherein the shortening range is dynamically adjusted according to a first deviation proportion of the discharging flow and the first process demand value, and a second deviation proportion of the total discharging amount and the second process demand value, and the maximum shortening rate is less than a preset shortening rate; If the discharging flow is higher than the first process demand value, or the actual total discharging amount exceeds the second process demand value, the discharging interval time is extended.

[0016] According to the intelligent shell cracking and discharging system based on machine vision provided by the application, further comprising: A maintenance management device for outputting a maintenance strategy; The maintenance strategy includes: Daily maintenance items, including using a strong light flashlight to irradiate the lens surface of the image acquisition module, observing whether there are dust and electrolyte splashes and other stain accumulations on the lens surface; if the lens transparency decreases to affect the picture clarity, using special lens paper dipped in anhydrous ethanol to gently wipe the lens surface, and prohibiting the use of hard objects or ordinary paper towels; after cleaning, previewing the picture through the image processing calculation module to confirm that the lens has no stain residues and the imaging is clear; The air cooling device and temperature monitoring item includes checking the air pressure gauge reading of the cooling device to ensure that the air pressure is stable within the preset pressure range, and if the air pressure is lower than the lower limit pressure of the preset pressure range, contacting the air compressor station to adjust the air supply pressure; using an infrared temperature detector to detect the temperature of the outer shell of the active detection module and the body of the image acquisition module to ensure that the surface temperature is less than the first preset temperature; if the surface temperature exceeds the second preset temperature, checking whether the vortex cooling pipe is blocked and whether the air curtain outlet is unobstructed, and disassembling and cleaning the impurities inside the cooling pipe; Weekly special maintenance insulation performance retest items, including using an insulation resistance tester to detect the insulation resistance of the camera shell and the fixed frame again to ensure that the insulation resistance is greater than the preset resistance, and if the insulation resistance is less than or equal to the preset resistance, checking whether the insulation support and gasket are aged and damaged, and replacing the failed components in time; Communication link checking items, including logging in the background of the image processing calculation module to check whether the camera data transmission packet loss rate is less than the preset packet loss rate; at the same time, checking the sealing condition of the network cable connector, and if the sealing glue is found to be aged and fallen off, reapplying high-temperature resistant sealing glue for protection; Maintenance records and abnormality handling items, including establishing a system maintenance account book, and recording in detail the time of each maintenance, maintenance items, detection data and abnormal conditions; if serious scratches on the lens, continuous over-standard equipment temperature and other problems that cannot be solved on site are found, the equipment is stopped and reported to the equipment management department, and a standby monitoring device is enabled at the same time.

[0017] The intelligent shell breaking and discharging system based on machine vision provided by the application constructs a reference reference line through an active detection module, acquires dynamic images of the shell breaking hammer head movement and the material falling at the discharging port through an image acquisition module, accurately extracts shell breaking depth parameters and actual discharging flow data in combination with an image processing calculation module, and forms a closed loop control of shell breaking and discharging in cooperation with a tank control machine; at the same time, the synergistic effect of the high-temperature resistant insulation structure and the cooling device effectively overcomes the adverse effects of the high temperature and dust environment of the aluminum electrolysis tank on the detection equipment, so that the shell breaking depth and discharging amount control precision can be improved, the maintenance cost can be reduced, and the adaptability to the high temperature and strong electric field environment can be improved. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present application or the prior art, the drawings required to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort on the basis of these drawings.

[0019] Figure 1 is a structural schematic diagram of an intelligent shell breaking and discharging system based on machine vision provided by an embodiment of the present application. DETAILED DESCRIPTION

[0020] In order to make the objects, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be clearly and completely described below in combination with the drawings in the present application. Obviously, the described embodiments are some embodiments of the present application, rather than all the embodiments. On the basis of the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative effort belong to the protection scope of the present application.

[0021] In the prior art, the monitoring and control of the shell breaking and discharging link in the aluminum electrolysis production process have long relied on traditional sensor technology. The displacement sensor and the pull rope sensor can only collect the displacement parameters of the shell breaking hammer head, and cannot reflect the key working condition information such as the wear degree of the hammer head and the adhesion state of the electrolyte. At the same time, the high temperature and strong electromagnetic field environment leads to a high sensor failure rate, and the average failure-free operation time is less than three months. The discharging system is affected by poor material flowability, and there is a bridging and wall-hanging phenomenon, and the error of the traditional weighing measurement method is as high as 15%, which cannot meet the process precision requirement. In the actual case of a certain aluminum electrolysis plant, the shell breaking depth is out of control due to the failure of the sensor, which causes abnormal fluctuation of the electrolyte level, and causes the daily output of a single tank to decrease by 2.3%.

[0022] In order to solve the above problems, it is found through research that simply improving the performance of the sensor cannot break through the environmental adaptability bottleneck, and instead a non-contact monitoring scheme is explored. Through analyzing the working condition characteristics of the electrolytic tank, it is found that high temperature radiation and electrolyte splashing are the main factors affecting the stability of the equipment, and strong electric field interference easily leads to misoperation of electronic equipment. In order to solve the problem of image acquisition clarity, an active light source is introduced to enhance the feature contrast. After many field tests, the penetration ability of the laser marking line in the dust environment is verified, and a conversion model of the pixel area and the physical parameters is developed. Finally, a technical route combining machine vision and structural protection is formed.

[0023] Therefore, the present application provides the following technical solutions: Figure 1 is a structural schematic diagram of an intelligent shell breaking and discharging system based on machine vision provided by an embodiment of the present application. As Figure 1As shown, the machine vision-based intelligent crust breaking and discharging system of the embodiment can include a proactive detection module 11, an image acquisition module 12, an image processing and calculation module 13, a data communication module 14, a tank control machine 15, a high-temperature-resistant insulation structure 16, and a cooling device. The proactive detection module 11 forms a reference baseline in the monitoring area, the image acquisition module 12 acquires the movement image of the crust breaking hammer head and the material image of the discharging port, the image processing and calculation module 13 extracts the features of each image and calculates the depth parameter and the flow data, the data communication module 14 transmits data to the tank control machine 15 to realize closed-loop control, the high-temperature-resistant insulation structure 16 provides physical protection and electrical isolation, and the cooling device maintains the working temperature of the equipment and prevents lens contamination.

[0024] The proactive detection module 11 refers to a device that enhances target features through controllable light sources. A high-power linear laser marker can be used, and the laser line width can be flexibly adjusted according to the requirements of the on-site environment. Due to the complex environment in the aluminum electrolysis tank, there are interference factors such as high temperature and dust. This module can form a clear reference baseline in the monitoring area by actively emitting high-brightness laser markings, significantly improving the image contrast and feature recognition of key targets such as crust breaking hammer heads and discharging ports, and providing favorable conditions for the accurate analysis of the image processing and calculation module 13.

[0025] The image acquisition module 12 refers to an optical acquisition device with environmental adaptability, such as a high-temperature-resistant special camera, which has excellent high-temperature environmental adaptability and imaging performance. Its core parameters include: the upper temperature limit reaches 250°C, the resolution is 1080P high-definition level, and electric zoom adjustment is supported to accurately capture target details; the protection level reaches IP68, which can effectively resist the erosion of dust, water vapor, and other harsh environments in the electrolytic tank. In terms of communication protocols, it supports multiple standard protocols such as TCP / IP, ICMP, HTTP, HTTPS, FTP, DHCP, DNS, DDNS, RTP, and RTSP, ensuring the stability and compatibility of data transmission; storage uses local hard disk storage, and power supply supports POE (Power over Ethernet), simplifying wiring deployment. The core function of this module is to perform real-time and clear image acquisition of the movement state of the crust breaking hammer head in the aluminum electrolysis tank and the material falling process of the discharging port, providing high-quality visual data for subsequent processing.

[0026] The image processing calculation module 13 refers to a processing unit for performing feature recognition and parameter conversion, for example, a high-performance RISC-V dual-core processor + KPU neural network processor architecture, with a computing power of 6TOPS equivalent computing power, supporting INT8 / INT16 two data precision operations, which can efficiently run complex image recognition and parameter calculation algorithms. In terms of hardware configuration, the running memory is 2GBytes (LPDDR4), and the storage capacity can support up to 512GB, meeting the real-time data caching and algorithm model storage requirements. As the "computing core" of the system, its main function is to quickly preprocess, feature extract and data calculate the real-time pictures transmitted by the image acquisition module 12, and accurately output the depth parameters of the hammer head entering the electrolyte and the actual flow data of the material.

[0027] The data communication module 14 can be a RISC-V dual-core processor integrated with USB, RS485, CAN, Ethernet and other standardized communication interfaces. It can rely on stable hardware interface links to report the accurate results output by the image calculation module in real time and with low delay, provide key data support for the control logic operation, dynamic adjustment strategy formulation and decision execution of the tank control machine 15, and ensure efficient and accurate control of the target system by the tank control machine 15.

[0028] The high-temperature-resistant insulation structure 16 refers to a multi-layer composite protection device made of 304 stainless steel material, which has excellent high-temperature resistance and corrosion resistance. The double-layer stainless steel protection structure inside effectively isolates external high-temperature radiation and impact, greatly prolonging the service life of the equipment. At the same time, the shell and the internal circuit are reliably electrically isolated by high-temperature-resistant insulation pads such as mica sheets, avoiding interference of the strong electric field of the electrolytic cell on the equipment circuit; The supporting network cable is a high-temperature-resistant comprehensive cable with a temperature upper limit of 200°C, ensuring stable and smooth communication link in high-temperature environment.

[0029] The cooling device can use compressed air vortex cold pipe cooling technology. Compressed air is cooled by vortex cold pipe and uniformly flows through the device interlayer to achieve cooling, ensuring that the internal core components are in the appropriate working temperature range. In addition, the front cover is designed with a compressed air wind curtain structure, which forms an air curtain barrier in front of the device lens, effectively blocking dust from adhering to the electrolytic cell, ensuring clear camera imaging. The vortex cooling pipe has the characteristics of instantaneous cooling and rapid response, which can efficiently solve the problem of equipment heat dissipation in high-temperature environment.

[0030] Specifically, the laser line is projected on the working area of the crust breaking hammer head, and the camera collects dynamic images at a rate of 25 frames per second. The image processor locks the hammer head profile by color threshold segmentation, calculates the pixel area of the adhered electrolyte area, and converts the actual penetration depth combined with the calibration parameters. After the background difference processing of the material image at the discharge port, the pixel area accumulation value of the material passing through is counted, and the instantaneous flow rate is calculated combined with the time parameter. After receiving the depth and flow data, the slot control machine 15 compares the preset process parameters to generate control instructions, dynamically adjusts the stroke of the crust breaking mechanism and the opening frequency of the discharge valve. The high-temperature-resistant structure is isolated from external heat radiation by an air sandwich, and the cooling device maintains the internal temperature below 70°C to ensure the normal operation of electronic components.

[0031] The present application can simultaneously obtain multi-dimensional information such as hammer head position, wear state and electrolyte adhesion by visual feature analysis, and realize millisecond-level response and continuous monitoring capability. Under the same working condition, the application of the system in a certain electrolytic cell reduces the crust breaking depth control deviation, reduces the discharge metering error, and prolongs the equipment maintenance period.

[0032] Through the above technical solution, the present application effectively solves the problem of equipment reliability in a high-temperature strong electric field environment, and realizes accurate measurement of the crust breaking depth and the discharge flow. The system automatically maintains the stability of the process parameters through a closed-loop control mechanism, avoiding the lag of manual intervention and improving the efficiency of the electrolytic cell. Multi-dimensional data acquisition provides a reliable basis for process optimization, and the structural protection design significantly reduces the equipment failure rate.

[0033] In one specific implementation process, the system is installed as follows: The high-temperature-resistant camera is fixed on the electrolytic cell cover plate at the pre-set installation position through a special insulating support, and the camera angle and position are accurately calibrated through a three-dimensional adjusting support to ensure that the lens axis is perpendicular to the center area of the discharge port (deviation not more than ±5mm). Turn on the active detection module 11, adjust the laser line width to 8-10mm, and simultaneously operate the camera electric zoom function to make the crust breaking hammer head movement trajectory and the discharge port profile clearly presented in the preview picture (the target area pixel ratio is not less than 30%). After installation, use an insulating resistance tester to test the insulation performance between the camera shell and the fixed frame to ensure that the insulation resistance is ≥100MΩ to meet the safety operation requirements in the strong electric field environment of the electrolytic cell.

[0034] The system debugging is as follows: The high-temperature-resistant comprehensive cable is used to connect the camera and the industrial Ethernet switch in the plant to ensure that the net joint is well sealed and protected. The device configuration tool is run at the image processing and computing module 13 end, the camera device is identified through the IP address scanning function, the communication parameter matching (including IP address, subnet mask, gateway setting) is completed according to the TCP / IP protocol standard, and a stable data transmission link is established. After the communication connection is established, the picture test program of the image processing and computing module 13 is started, whether the real-time transmitted video stream exists frame loss, delay (the delay should be ≤200 ms) or color deviation phenomenon is checked, and the display effect of the active detection laser line in the image is verified to ensure that the identification degree of the key feature area meets the algorithm operation requirements. After the debugging is completed, the system configuration parameters are saved, and the device operation benchmark report is generated.

[0035] In some embodiments, the application further proposes an intelligent shell breaking and discharging system based on machine vision. The image processing and computing module 13 matches the shell breaking hammer head movement image with the preset LAB depth threshold template to obtain a first matching result. If the first matching result indicates that there is a matched adherent electrolyte feature area, the adherent electrolyte feature area is framed by a minimum enclosing rectangle, and the width, height and pixel area of the minimum enclosing rectangle are calculated. According to the conversion relationship between the pixel area and the actual size, the pixel area is converted into the actual depth value of the shell breaking hammer head entering the electrolyte, wherein the conversion relationship between the pixel area and the actual size is determined by the calibration parameters during system installation.

[0036] The LAB depth threshold template refers to an adherent electrolyte feature recognition model based on the LAB color space. Specifically, electrolyte images under different lighting conditions can be collected, the chroma and brightness distribution range of the LAB channel is extracted, and a color threshold template is generated. The LAB channel refers to the specific L channel brightness, A channel red-green color difference and B channel blue-yellow color difference. The template is used to distinguish the electrolyte adherent area from other background areas and improve the accuracy of image segmentation. Specifically, a system threshold editing tool can be started, and sample images under typical working conditions of aluminum electrolysis cells (including different thickness state images of shell breaking hammer head adherent electrolyte, and the number of samples is not less than 50 groups) are imported. Through the manual adjustment and automatic optimization functions of the tool, the LAB color space threshold range of the two types of targets is calibrated (the LAB threshold deviation of the hammer head adherent electrolyte feature is controlled within ±3), the feature recognition template (i.e. the LAB depth threshold template) is generated and saved to the image processing and computing module 13, and is used as the basis for subsequent real-time recognition.

[0037] The minimum circumscribed rectangle refers to the minimum rectangular frame that can completely enclose the adhesion electrolyte feature area, and can be specifically realized by extracting the feature area contour through an edge detection algorithm, and then calculating the minimum circumscribed rectangle thereof through a rotating jigsaw algorithm. The rectangle is used to quantify the geometric size of the feature area, avoiding measurement errors caused by irregular electrolyte shapes.

[0038] The conversion relationship between the pixel area and the actual size refers to the conversion ratio of the image pixel quantity and the real physical size, and can be specifically realized by placing a calibration object with a known size during the system installation stage, and calculating the ratio of the pixel area and the actual area. The ratio relationship is used to convert the pixel data in the image into an actual depth value, eliminating the effects of lens distortion and viewing angle errors.

[0039] Specifically, in the process of the crust breaking hammer head entering the electrolyte, the image acquisition module 12 captures images containing the hammer head and the electrolyte in real time. The image processing calculation module 13 first converts the image to the LAB color space, and identifies the adhesion electrolyte area by using a preset LAB depth threshold template. When a matching area is detected, the contour boundary thereof is extracted through an edge detection algorithm, a minimum circumscribed rectangle is generated, and the pixel area thereof is calculated. According to the mapping relationship between the pixel area and the actual depth pre-calibrated during the system installation stage, the pixel area is converted into an actual invasion depth value. For example, during the calibration stage, a rectangular calibration board with a known size can be placed on the working plane of the hammer head, and a nonlinear conversion function is established by measuring the ratio of the pixel area and the actual area of the calibration board in the image. In real-time calculation, the function is used to map the pixel area of the adhesion area to the actual depth value, so as to accurately reflect the invasion state of the crust breaking hammer head.

[0040] The present application directly quantifies the geometric features of the adhesion electrolyte area through image processing, realizes non-contact depth measurement by combining calibration parameters, avoids sensor drift and corrosion problems. At the same time, the use of the LAB color space template enhances the robustness of feature recognition under complex lighting conditions, and the minimum circumscribed rectangle algorithm effectively reduces the influence of irregular shapes on measurement accuracy.

[0041] Through the above technical solutions, the present application solves the problem that the traditional displacement sensor cannot accurately measure the adhesion thickness of the electrolyte, and realizes non-contact dynamic monitoring of the crust breaking depth. Through color space segmentation and geometric feature quantization, the anti-interference ability and repeatability of the measurement results are significantly improved. The introduction of the calibration parameter further eliminates lens distortion and installation errors, making the depth calculation more consistent with the actual working conditions, and providing a reliable data basis for closed-loop control. In addition, this method does not need to frequently replace the sensor, reduces the maintenance cost, and adapts to the high temperature and strong electric field environment of the aluminum electrolysis cell.

[0042] In some embodiments, the present application further proposes an intelligent crust-breaking and discharging system based on machine vision. When the system is installed, a calibration object with a known size is placed at the working position of the crust-breaking hammer head, and the image acquisition module 12 acquires the image of the calibration object to establish the conversion relationship between the pixel area and the actual size. Specifically, one or more calibration objects with known physical sizes can be used, which are placed on the typical working plane of the crust-breaking hammer head. The image acquisition module 12 acquires the calibration image containing the calibration object. The image processing and calculation module 13 identifies and calculates the pixel area occupied by the calibration object in the calibration image. Based on the known actual physical area of the calibration object and its corresponding pixel area, a nonlinear mapping function is established by fitting algorithm, taking the pixel area as the independent variable and the actual physical depth as the dependent variable. The mapping function and its parameters are stored in the system for subsequent real-time calculation of the actual penetration depth of the crust-breaking hammer head.

[0043] In a specific implementation process, the internal temperature of the high-temperature-resistant insulation structure 16 and the external tank temperature of the aluminum electrolysis tank can be obtained, and the pixel area correction coefficient at the current temperature can be calculated according to the pre-calibrated temperature and distortion coefficient correlation model. The hammer head profile in the hammer head motion image is extracted, the current diameter of the hammer head is calculated, and the difference between the initial diameter of the hammer head and the current diameter of the hammer head is obtained to obtain the wear amount of the hammer head. According to the pre-set correlation between the wear amount and the depth correction amount, the depth correction amount is determined. After the pixel area is corrected by using the pixel area correction coefficient, the preliminary depth is obtained according to the conversion relationship between the pixel area and the actual size, and the depth correction amount is superimposed to obtain the corrected depth. The corrected depths of the continuous multiple frames of the crust-breaking hammer head motion image are removed by using the Grubbs criterion to remove outliers, and the arithmetic mean of the remaining effective depths is taken to obtain the actual depth value.

[0044] In the formula, the calibration object refers to a reference object with a known actual physical size, which can be realized by using a metal block or a ceramic plate with a standard size, and the surface is coated with a high-contrast coating to enhance the image recognition effect. This feature provides a reference conversion basis for subsequent real-time calculation by establishing a mapping relationship between the pixel area and the actual depth.

[0045] In the formula, the nonlinear mapping function refers to a mathematical model describing the nonlinear relationship between the pixel area and the actual depth, which can be realized by using a square root calculation formula or a polynomial equation. This function can accurately reflect the complex correlation between image features and physical parameters by fitting calibration data, avoiding conversion errors caused by linear assumptions.

[0046] In the formula, the temperature and distortion coefficient correlation model refers to a mathematical model reflecting the influence of temperature change on the optical system of the image acquisition module 12, which can be established by experimentally calibrating the image distortion degree at different temperatures. This model is used to correct the pixel area measurement deviation caused by the thermal expansion of the lens due to high temperature, ensuring data accuracy.

[0047] The hammer head wear amount refers to the diameter reduction amount of the shell breaking hammer head caused by mechanical friction during use, and can be determined by extracting the hammer head contour and calculating the difference between the current diameter and the initial diameter through an image processing algorithm. This parameter is used to correct the depth measurement error caused by hammer head wear, ensuring the reliability of the calculation results.

[0048] The Grubbs criterion refers to a statistical outlier detection method, which can be implemented by calculating the sample mean and standard deviation to determine whether each data point is outside the pre-set threshold range. This criterion is used to eliminate abnormal depth values caused by image noise or transient interference, improving the stability of the final output results.

[0049] Specifically, during system installation, the calibration object is accurately placed on the working plane of the shell breaking hammer head, and the calibration image containing the calibration object is obtained by the image acquisition module 12. The image processing calculation module 13 identifies the contour of the calibration object and calculates its pixel area, and combines the known actual area to generate a nonlinear mapping function. During real-time monitoring, the system synchronously acquires temperature data, calculates the current pixel area correction coefficient according to the pre-calibration model, and simultaneously extracts the hammer head contour to calculate the depth correction amount corresponding to the wear amount. After the preliminary depth value is corrected by temperature and wear, abnormal data is removed through statistical methods, and the final actual depth value is output after multi-factor calibration. The entire process realizes accurate conversion from image features to physical parameters, and dynamically adapts to environmental changes and equipment wear and tear.

[0050] The present application introduces a nonlinear mapping function, a temperature correction model and a wear compensation mechanism to construct a multi-factor collaborative calibration system, effectively overcoming the measurement distortion problem under complex working conditions, and significantly improving the accuracy and stability of depth detection.

[0051] Through the above technical solutions, the present application can dynamically compensate for image distortion caused by high temperature environment and mechanical wear caused by long-term use of equipment, ensuring the accuracy of shell breaking depth measurement results under different working conditions. At the same time, by removing abnormal data through statistical methods, the reliability of the output parameters is further improved, providing a high-precision data basis for closed-loop control of the shell breaking process.

[0052] In some embodiments, the application further proposes that the image processing calculation module 13 is further used for counting a first recognition accuracy of the LAB depth threshold template according to a first preset period, the first recognition accuracy being a ratio of a number of successful matches to a total number of detections; if the first recognition accuracy in a current period is less than a first preset accuracy for a continuous time length reaching a first preset time length, adaptive learning is triggered; a first preset number of first effective images in the current period is collected, a threshold range of the LAB color space is recalibrated using the maximum inter-class variance method to generate an adjusted LAB depth threshold template, and a trial mode is entered; if a second recognition accuracy of the adjusted LAB depth threshold template in the trial mode is greater than or equal to the first preset accuracy, the original template is replaced, and if the second recognition accuracy is still less than the first preset accuracy, optimization is repeated until a stop condition is met.

[0053] The first preset period refers to a system set performance evaluation time interval of the template, which can be implemented by using a fixed period, for example, every 24 hours or a cumulative period of every 100 tons of aluminum ingots, for periodic monitoring of the applicability of the template. The first recognition accuracy refers to a probability index of successful template matching, which is calculated by counting a ratio of a number of successful matches to a total number of detections in a period, and is used to quantify the reliability of the template under the current working condition. The adaptive learning refers to an optimization process automatically started when the recognition accuracy continuously falls below a threshold value, which realizes dynamic updating of the template parameters by collecting effective images to reanalyze the color feature distribution. The maximum inter-class variance method refers to an algorithm for calculating the best segmentation threshold based on the image gray histogram, which is specifically used to find the segmentation point with the maximum inter-class variance by traversing the threshold combination of the LAB color channel, and is used to generate a new color space threshold range.

[0054] Specifically, during the operation of the aluminum electrolysis cell, changes in electrolyte composition or fluctuations in environmental light may cause the matching accuracy of the original LAB depth threshold template to decrease. The system automatically counts the recognition accuracy of the template every 24 hours, and triggers the optimization process when the accuracy is less than 85% for 8 consecutive hours. At this time, the system collects the last 50 effective crust breaking images, statistically analyzes the LAB color values of the electrolyte adhesion area in the images, and recalculates the threshold range of the L, A and B channels using the maximum inter-class variance method. For example, the new L channel threshold may be adjusted from the original [30-45] to [28-47], and the A channel may be expanded from [10-25] to [8-27]. The generated adjusted template runs in a trial mode for 4 hours, and if the accuracy during this period rises to more than 90%, the old template is formally replaced, otherwise new samples are continuously collected for iterative optimization until the maximum optimization number limit is reached.

[0055] The application can automatically start optimization at the early stage of template performance degradation through periodic accuracy evaluation and adaptive learning mechanism, effectively solves the false detection problem caused by changes in electrolyte crust morphology, and avoids production interruption caused by manual intervention.

[0056] Through the technical solution, the self-optimization maintenance of the LAB depth threshold template is realized, and the long-term stability of the crust depth detection is ensured. When the electrolyte crust morphology changes due to the change of process parameters, the system can automatically adjust the color recognition threshold range, avoid the depth calculation error caused by the invalid template, and ensure the reliability of the crust depth closed-loop control.

[0057] In some embodiments, the application further proposes that the image processing calculation module 13 is further used for matching the material image at the discharging port with the preset LAB flow threshold template to obtain a second matching result; if the second matching result indicates that there is a material feature area matched with the LAB flow threshold template, timing is started and the rectangular pixel area of the first detected material feature area is recorded; the material feature area at the discharging port is continuously monitored, the pixel areas of all the material feature areas are accumulated until the last material feature area disappears to stop timing to obtain the accumulated pixel area sum and the timing duration; the accumulated pixel area sum and the timing duration are multiplied to obtain the discharging flow, and the accumulated pixel area sum and the size conversion coefficient are multiplied to obtain the total amount of single discharging.

[0058] The LAB flow threshold template refers to a set of material feature recognition parameters established based on the LAB color space, and can be realized by calibrating the material color threshold range by using the maximum inter-class variance method to accurately distinguish the falling material and the background area. The process of obtaining the LAB flow threshold template is similar to the process of obtaining the LAB depth threshold template, and details can be referred to the related description above, which will not be repeated here.

[0059] The accumulated pixel area sum refers to the cumulative value of the pixel areas of all the material feature areas in the single discharging process, which can be realized by extracting the pixel number of the material area in the continuous frame through the image segmentation algorithm, and is used to represent the total amount of material falling. The size conversion coefficient refers to the conversion relationship parameter between the pixel area and the actual physical quantity, which can be realized by establishing a nonlinear mapping function through calibrating the material image analysis, and is used to convert the image data into the actual discharging amount.

[0060] Specifically, when the material starts to fall, the system recognizes the material area through the LAB color space feature and triggers the timer, and continuously tracks the material falling process. The pixel area of the material area is extracted and accumulated in each detection period until the material flow disappears to stop timing. The total pixel area is multiplied by the size conversion coefficient to obtain the total amount of single discharging, and the discharging flow per unit time is calculated in combination with the timing duration. This process realizes the whole process dynamic monitoring of the material falling process, wherein the size conversion coefficient eliminates the influence of perspective distortion and lens distortion through pre-calibration, and ensures that the measurement result is consistent with the actual physical quantity.

[0061] The present application avoids physical contact through visual measurement, eliminating systematic errors caused by material adhesion. The existing control system can only act according to the preset program, and the present application realizes accurate feedback of the actual discharge amount through real-time accumulation of pixel area, providing a data basis for closed-loop control.

[0062] Through the above technical solutions, the present application effectively solves the problem of inaccurate measurement caused by poor flowability of alumina powder, avoids sensor corrosion and wear through non-contact visual measurement, and realizes accurate quantification of the discharge amount. The dynamic accumulation computer mechanism can accurately capture instantaneous flow fluctuations, providing a reliable basis for optimizing the discharge interval and significantly improving the stability of material supply for electrolytic cells.

[0063] In some embodiments, the present application further proposes an optimization method for the image processing calculation module 13 in the intelligent shell cracking and discharging system based on machine vision, which includes: according to a second preset period, counting a third recognition accuracy of the LAB flow threshold template; if the third recognition accuracy in the current period is less than the second preset accuracy for a continuous time length reaching the second preset time length, triggering adaptive learning; collecting a second preset number of second effective images in the current period, and using the maximum inter-class variance method to recalibrate the LAB color space threshold range to generate an adjusted LAB flow threshold template and enter a trial mode; in the trial mode, if a fourth accuracy of the adjusted LAB flow threshold template is greater than or equal to a preset accuracy, replacing the LAB flow threshold template; if the fourth accuracy is still less than the preset accuracy, repeating the optimization until the second stop optimization condition is met.

[0064] Among them, the second preset period can be a time interval set according to the production rhythm, for example, statistics is performed once every 8 hours. The second preset accuracy can be a judgment threshold set according to the process requirement, for example, 90%. The maximum inter-class variance method refers to determining the optimal segmentation threshold by calculating the maximum inter-class variance of the image gray histogram, and the adaptive segmentation of the color space can be realized by using iterative calculation method. The trial mode refers to the stage of applying the newly generated threshold template to the real-time detection process and verifying the effect, and the synchronous testing of the new and old templates can be realized by setting parallel processing channels.

[0065] Specifically, when the system detects that the recognition accuracy of the LAB flow threshold template continues to be lower than the set threshold, the optimization mechanism is automatically triggered. By collecting effective image samples under the current working condition, the color distribution characteristics of the material area are recalculated using the maximum inter-class variance method, and a new LAB flow threshold template is generated. The new template runs in parallel with the original template during the trial period, and the detection accuracy of the two templates is compared to determine whether to replace the template. If the new template meets the preset accuracy requirement during the trial period, the template update is completed; if it does not meet the requirement, samples are continuously collected and the threshold range is optimized until the stopping condition is met. This process ensures that the threshold template is always adapted to environmental light changes, material color fluctuations, and other interference factors through a closed-loop feedback mechanism.

[0066] The present application realizes dynamic optimization of threshold parameters through periodic accuracy monitoring and adaptive learning mechanism, effectively solving the detection failure problem caused by environmental interference in traditional methods.

[0067] Through the above technical solutions, the present application can correct the color recognition deviation in the discharging flow detection in real time, ensuring the stability of material feature extraction. Through the automatic threshold optimization process, the frequency of manual intervention is reduced, and the accumulation of detection errors caused by environmental factor changes is avoided. Ultimately, the reliability of the discharging flow data is improved, providing accurate input parameters for closed-loop control, thereby optimizing the timeliness and accuracy of discharging interval adjustment.

[0068] In some embodiments, the present application further proposes that the tank control machine 15 establishes a correlation model between the pixel area and the set crust breaking depth; when it is monitored that the pixel area exceeds the preset upper threshold, it is determined that the crust breaking is too deep, and an adjustment instruction is sent to the crust breaking execution mechanism to decrease the set crust breaking depth by a gradient; when it is monitored that the pixel area is lower than the preset lower threshold, an adjustment instruction is sent to the crust breaking execution mechanism to increase the set crust breaking depth; wherein the preset upper threshold and the preset lower threshold are determined by aluminum electrolysis production process parameters.

[0069] The correlation model between the pixel area and the set crust breaking depth refers to establishing a mathematical correspondence between the pixel area data of the crust breaking hammer head invading the electrolyte area output by the image processing module and the set depth required by the production process, which can be obtained by linear regression or machine learning algorithm training, and is used to convert the visual detection result into a control parameter. The preset upper threshold refers to the maximum allowed adhesion area critical value set according to the electrolytic tank running state, which can be obtained by dynamic calculation of process parameters, and is used to determine whether the crust breaking depth exceeds the process safety range. The gradient decrease of the set crust breaking depth refers to gradually reducing the crust breaking set value by a fixed step, which can be achieved by adjusting the original set value by 5%-10% each time, avoiding control quantity mutation causing execution mechanism oscillation.

[0070] Specifically, during the operation of the aluminum electrolysis cell, pixel area data of the penetration of the crust breaking hammer head into the electrolyte area is obtained in real time by the image processing module, and the data is input into the correlation model to be converted into an equivalent depth value. When it is detected that the pixel area continuously exceeds the preset upper threshold value, it indicates that the actual crust breaking depth exceeds the process requirement, at which time the cell control machine 15 sends a control instruction to the actuator to reduce the crust breaking depth by a preset gradient, for example, reducing the set depth by 2 cm each time. Conversely, when the pixel area is lower than the lower threshold value, the adjustment is made in the opposite direction. This control method forms a closed loop by combining the visual detection data with the actuator action, so that the crust breaking depth can dynamically follow the changes in process requirements.

[0071] The present application realizes closed-loop control based on actual working conditions by establishing a dynamic correlation model of pixel area and depth and generating a control instruction in combination with real-time visual detection data. In the prior art, the displacement sensor can only measure the mechanical displacement, and cannot detect the electrolyte adhesion state, while the present application can accurately identify the actual penetration depth through image analysis, effectively avoiding the control deviation caused by hammer head wear or electrolyte state changes.

[0072] Through the above technical solutions, the present application solves the problem of over-deep or over-shallow in traditional crust breaking control due to the lack of real-time feedback, and can dynamically adjust the execution parameters according to the actual detection data. In the case of electrolyte state mutation or hammer head wear, the system can automatically correct the control amount to ensure that the crust breaking depth is always within the process allowable range. At the same time, the gradient adjustment method avoids frequent action of the actuator, prolongs the service life of the equipment while ensuring the control accuracy.

[0073] In some embodiments, the present application further proposes that the cell control machine 15 collects aluminum electrolysis production process parameters in real time, the aluminum electrolysis production process parameters including cell temperature, electrolyte molecular ratio, electrolyte level and aluminum level, and substitutes the cell temperature, electrolyte molecular ratio, electrolyte level and aluminum level into a preset nonlinear mapping formula to calculate the preset upper threshold value and the preset lower threshold value. The nonlinear mapping formula can include a preset upper threshold value formula and a preset lower threshold value, the preset upper threshold value is determined by the formula , and the preset lower threshold value is determined by the formula . In the formula, and are proportional coefficients, and are cell type base correction values, in the formula, the positive index indicates that the increase of the parameter leads to the increase of the target value, and the negative index indicates that the increase of the parameter leads to the decrease of the target value, represents the cell temperature, represents the electrolyte molecular ratio, represents the electrolyte level, represents the aluminum level.

[0074] wherein the bath temperature refers to the actual temperature of the electrolyte in the aluminum reduction cell, which can be collected in real time by a thermocouple or an infrared temperature measuring device, and the negative index relationship means that the preset threshold decreases when the bath temperature rises, so as to adapt to the demand of the crust breaking depth caused by the liquidity change of the electrolyte in the high temperature environment. The electrolyte molecular ratio refers to the molar ratio of sodium fluoride and aluminum fluoride in the electrolyte, which can be obtained by an online analyzer or laboratory testing, and the positive index relationship reflects that the electrolyte viscosity increases when the molecular ratio rises, and the preset threshold needs to be increased to match the working condition of the increased adhesion area. The electrolyte level refers to the height of the electrolyte melt, which can be measured by a laser ranging or a float type sensor, and the positive index relationship corresponds to the increase of the crust breaking resistance when the electrolyte level rises, and the preset threshold needs to be adjusted. The aluminum level refers to the height of the liquid aluminum layer in the cell, which can be detected by an electromagnetic measuring device, and the negative index relationship reflects the influence of the change of the anode immersion depth on the crust breaking effect when the aluminum level rises. The index parameters in the nonlinear mapping formula can be determined by fitting the historical production data, for example, the index value is optimized by using a multivariate regression analysis method, so that the preset threshold dynamically adapts to the process characteristics of different cell types.

[0075] Specifically, the cell controller 15 periodically obtains the real-time measurement values of the bath temperature, the molecular ratio, the electrolyte level and the aluminum level through the data acquisition interface, and inputs these parameters into the preset nonlinear calculation formula. For example, when the bath temperature rises, the term in the formula makes the calculation result decrease, so as to automatically reduce the preset upper threshold, and prevent the crust breaking from being too deep due to the enhanced electrolyte liquidity in the high temperature environment. At the same time, the positive index term of the electrolyte level makes the calculation result increase when the electrolyte level rises, so as to match the natural increase trend of the adhesion area caused by the rise of the electrolyte level. The and obtained by calculation are input into the crust breaking depth control module as dynamic thresholds, instead of the traditional fixed threshold, so as to realize the automatic adjustment of the crust breaking set depth with the process condition fluctuation. and can be set as initial values according to the size specifications of different electrolytic cells, and are gradually optimized in the production process through a self-learning algorithm.

[0076] The present application establishes a nonlinear calculation model coupled with multiple parameters, so that the preset threshold can reflect the change of the running state of the electrolytic cell in real time, for example, when the aluminum level abnormally rises, the term in the formula automatically adjusts the , so as to avoid the insufficient actual insertion depth of the hammer head caused by the rise of the aluminum liquid. This dynamic threshold adjustment mechanism solves the problem of strong hysteresis and poor adaptability of the manually experienced set threshold.

[0077] By the technical scheme, the present application realizes accurate matching of the crust breaking depth control parameter and the real-time process state of the electrolytic cell, effectively avoids the phenomenon of over-deep or over-shallow crust breaking caused by sudden change of cell temperature or fluctuation of electrolyte composition. For example, when the electrolyte molecule ratio suddenly increases, the system automatically increases the preset threshold value to prevent misjudgment of the adhesion area caused by the increase of electrolyte viscosity, thereby improving the execution precision of the crust breaking action and reducing the problems of crust hardening or electrolyte spattering caused by improper threshold setting.

[0078] In some embodiments, the present application further proposes a control method for dynamically adjusting the discharging interval time of the tank control machine 15 according to the comparison results of the discharging flow rate and the preset first process requirement value, and the comparison results of the total discharging amount and the preset second process requirement value. Specifically, when the discharging flow rate is lower than the first process requirement value or the total discharging amount does not reach the second process requirement value, the discharging interval time is shortened by a deviation proportion, but the maximum shortening rate is limited by a preset threshold value; when the discharging flow rate or the total amount exceeds the requirement value, the discharging interval time is lengthened.

[0079] The discharging flow rate refers to the volumetric flow rate of the material through the discharging port per unit time, which can be obtained by multiplying the quotient of the total accumulated pixel area calculated by the image processing module and the timing duration by the size conversion coefficient. The total discharging amount refers to the total volume of the material in a single discharging process, which is determined by the product of the total accumulated pixel area and the size conversion coefficient. The deviation proportion dynamic adjustment mechanism refers to adjusting the control parameter in proportion to the deviation degree of the real-time detection value and the process requirement value, for example, when the flow deviation reaches 10%, the interval time can be shortened by 5%. The maximum shortening rate limit is to avoid system oscillation caused by excessive control action, for example, the single adjustment amplitude is limited to not more than 20% of the original interval time.

[0080] Specifically, the control method realizes closed-loop feedback control of the discharging flow rate and total amount data with the process requirement value in real time. When it is detected that the actual discharging amount is insufficient, the system automatically calculates the current deviation proportion, generates a command to shorten the interval time according to the preset algorithm, but is limited by the maximum shortening rate to prevent excessive adjustment. For example, when the flowability of alumina powder decreases to cause poor discharging, the system can gradually shorten the discharging interval until the flow rate recovers, while avoiding material accumulation caused by excessive adjustment amplitude. Conversely, when excessive discharging is detected, the system lengthens the interval time to reduce material waste. The process is realized by the control algorithm built in the tank control machine 15, and the mapping relationship between the deviation proportion and the adjustment amplitude can be adapted to different electrolytic cell conditions by parameterized configuration.

[0081] In some embodiments, the deviation ratio calculation can adopt a piecewise function process, for example, maintaining the original parameters when the deviation is less than 5%, adopting linear adjustment in the interval of 5%-15%, and starting the intensive adjustment mode when exceeding 15%. The maximum shortening rate can be set as a dynamic parameter, which is automatically adjusted according to the current thermal balance state of the electrolytic cell, for example, appropriately reducing the maximum shortening rate threshold when the cell temperature rises.

[0082] Compared with the weighing sensor scheme, the present application avoids the metering error caused by material wall hanging, and the control precision is significantly improved.

[0083] Through the above technical scheme, the present application realizes intelligent adjustment of the discharging process, can automatically compensate the interference caused by the change of material flow characteristics, and ensures the accurate matching of the actual discharging amount and the process demand value. The control method effectively eliminates the hysteresis of manual experience adjustment, prevents the hardening of the shell surface or the uneven discharging phenomenon, and at the same time, guarantees the stability of the control system through the maximum shortening rate limitation, avoiding the equipment wear problem caused by frequent large adjustment.

[0084] In some embodiments, the present application further proposes a maintenance management device for outputting a maintenance strategy; the maintenance strategy includes daily maintenance items, air cooling device and temperature monitoring items, weekly special maintenance insulation performance re-measurement items, communication link checking items, maintenance records and abnormal treatment items.

[0085] Among them, the daily maintenance items include using a strong light flashlight to irradiate the lens surface of the image acquisition module 12, observing whether there are dust and electrolyte splashes and other stain accumulations on the lens surface; if the lens transparency decreases to affect the picture definition, using a special lens paper to dip anhydrous ethanol to gently wipe the lens surface, and prohibiting the use of hard objects or ordinary paper towels to wipe; after cleaning, preview the picture through the image processing calculation module 13 to confirm that the lens has no stain residue and the imaging is clear.

[0086] The air cooling device and temperature monitoring items include checking the pressure gauge reading of the cooling device to ensure that the air pressure is stable within the preset pressure range, if the air pressure is lower than the lower limit pressure of the preset pressure range, contacting the air compressor station to adjust the air supply pressure; using an infrared temperature detector to detect the temperature of the outer shell of the active detection module 11 and the machine body of the image acquisition module 12 to ensure that the surface temperature is less than the first preset temperature; if the surface temperature exceeds the second preset temperature, checking whether the vortex cooling pipe is blocked and whether the air curtain outlet is unobstructed, and disassembling and cleaning the impurities in the cooling pipe; Every week, special maintenance insulation performance retest project, including the use of insulation resistance tester again to detect the camera shell and fixed frame insulation resistance, to ensure that the insulation resistance is greater than the preset resistance, if the insulation resistance is less than or equal to the preset resistance, check whether the insulation support and gasket are aging and damaged, and replace the failed components in time; Communication link checking project, including logging into the image processing computing module 13 background, checking whether the camera data transmission packet loss rate is less than the preset packet loss rate; at the same time, check the sealing condition of the network cable joint, if the sealing glue is found to be aged and fallen off, reapply high-temperature resistant sealing glue for protection; Maintenance records and abnormality handling project, including establishing system maintenance records, recording the time, maintenance project, test data and abnormality of each maintenance in detail; if the lens is found to be seriously scratched, the equipment temperature continues to exceed the standard and cannot be solved on site, stop the machine and report to the equipment management department, and at the same time, use the standby monitoring equipment to avoid affecting the electrolytic production.

[0087] Specifically, the maintenance strategy covers the key links of equipment maintenance through preset standardized operation processes. In daily maintenance, the degree of lens contamination is quickly judged by strong light irradiation, and the interference of electrolyte splashes on the imaging quality is eliminated by wiping with anhydrous ethanol. The air pressure monitoring ensures the stability of the cooling device air source in the wind cooling device maintenance, and the infrared temperature measurement identifies the overheating risk of the equipment to trigger the cleaning process of the cooling pipe. The insulation performance retest uses quantitative resistance testing instead of manual visual inspection to accurately identify the insulation failure position. The communication link checking combines background data monitoring with physical sealing protection to prevent high-temperature dust from entering and causing signal interruption. The maintenance records record the detailed data of each maintenance, providing basic data for equipment state analysis, and automatically switching to standby equipment in abnormal conditions to avoid production interruption.

[0088] Through the above technical solutions, the present application realizes systematic maintenance management of the aluminum electrolysis cell monitoring equipment, effectively reduces the imaging distortion risk caused by lens contamination, improves the running stability of the cooling device, avoids the equipment short circuit caused by insulation failure, and reduces the control delay caused by communication link interruption. Through the maintenance data recording and abnormality handling mechanism, the device availability is significantly improved, and the continuous and stable operation of the crust breaking and discharging closed loop control is ensured.

[0089] The device embodiments described above are only schematic, wherein the units shown as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e., can be located in one place, or can be distributed on multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the present embodiment. Those skilled in the art can understand and implement without creative labor.

[0090] It should be pointed out finally that the above embodiments are only used to illustrate the technical solutions of the present application, but not to limit the same; and although the present application has been described in detail with reference to the foregoing embodiments, it should be appreciated by those skilled in the art that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features thereof can be replaced equivalently; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A machine vision-based intelligent shell-breaking and blanking system, characterized in that, include: An active detection module is used to create a reference baseline in the monitoring area of ​​the aluminum electrolysis cell; The image acquisition module is used to acquire images of the movement of the shell-breaking hammer inside the aluminum electrolysis cell and images of the material at the feeding port; The image processing and calculation module is used to preprocess, extract features, and calculate data from the images of the shell-breaking hammer's motion and the material at the discharge port to obtain the depth parameters of the shell-breaking hammer entering the electrolyte and the actual flow rate data of the material at the discharge port. The data communication module is used to transmit depth parameters and actual flow data to the tank control machine; The slot control machine is used to adjust the shelling setting depth according to the depth parameter and optimize the feeding interval according to the actual flow data, forming a closed-loop control of shelling and feeding; High-temperature resistant insulation structure is used to provide physical protection, high-temperature protection, and electrical isolation for the image acquisition module, image processing and computing module, and data communication module; The cooling device is used to reduce the internal temperature of the high-temperature resistant insulation structure and to prevent dust from the aluminum electrolysis cell from adhering to the lens of the image acquisition module.

2. The intelligent shell-breaking and blanking system based on machine vision according to claim 1, characterized in that, The image processing and calculation module is specifically used for: The motion image of the shell-breaking hammer is matched with a preset LAB depth threshold template to obtain a first matching result; If the first matching result indicates the existence of an adhesive electrolyte feature region that matches the LAB depth threshold template, the adhesive electrolyte feature region is defined using the minimum bounding rectangle, and the width, height, and pixel area of ​​the minimum bounding rectangle are calculated. Based on the preset conversion relationship between pixel area and actual size, the pixel area is converted into the actual depth value of the shell-breaking hammer entering the electrolyte. The conversion relationship between pixel area and actual size is determined by the calibration parameters during system installation.

3. The intelligent shell-breaking and blanking system based on machine vision according to claim 2, characterized in that, The image processing calculation module is also used for: During system installation, a calibration object of known size is placed in the working position of the shell-breaking hammer, and the image acquisition module acquires an image of the calibration object to establish a conversion relationship between pixel area and actual size. The internal temperature of the high-temperature resistant insulation structure and the external temperature of the aluminum electrolytic cell are obtained, and the pixel area correction coefficient at the current temperature is calculated based on the pre-calibrated temperature and distortion coefficient correlation model. Extract the hammer head contour from the hammer head motion image, calculate the current diameter of the hammer head, and calculate the difference between the initial diameter and the current diameter of the hammer head to obtain the hammer head wear amount. Based on the preset correlation between the wear amount and the depth correction amount, determine the depth correction amount. After correcting the pixel area using a pixel area correction coefficient, a preliminary depth is obtained based on the conversion relationship between pixel area and actual size. Then, a depth correction amount is added to obtain the corrected depth. The corrected depth for multiple consecutive frames of shell-breaking hammer motion images is obtained by using the Grubbs criterion to remove outliers and taking the arithmetic mean of the remaining effective depths.

4. The intelligent shell-breaking and blanking system based on machine vision according to claim 2, characterized in that, The image processing calculation module is also used for: According to the first preset period, the first recognition accuracy of the LAB depth threshold template is calculated; If, within the current period, the duration for which the first recognition accuracy is less than the first preset accuracy reaches the first preset duration, adaptive learning is triggered. Collect the first valid image of the first preset number within the current period, and recalibrate the threshold range of the LAB color space using the maximum inter-class variance method, generate an adjustment template for the LAB depth threshold, and enter trial mode; Within the usage mode, if the second recognition accuracy of the adjusted LAB depth threshold template is greater than or equal to the first preset accuracy, replace the LAB depth threshold template. If the second recognition accuracy is still less than the first preset accuracy, repeat the optimization until the first stop optimization condition is met.

5. The intelligent shell-breaking and blanking system based on machine vision according to claim 1, characterized in that, The image processing calculation module is also used for: The material image at the discharge port is matched with a preset LAB flow threshold template to obtain a second matching result; If the second matching result indicates the existence of a material feature region that matches the LAB flow threshold template, start timing and record the rectangular pixel area of ​​the first detected material feature region; Continuously monitor the material feature areas at the feed port, accumulate the pixel areas of all material feature areas, and stop timing when the last material feature area disappears, to obtain the total accumulated pixel area and timing duration; The material flow rate is obtained by dividing the sum of the accumulated pixel areas and the timing duration, and then multiplying the product with the size conversion factor. The total material flow rate for a single feeding is obtained by multiplying the sum of the accumulated pixel areas and the size conversion factor.

6. The intelligent shell-breaking and blanking system based on machine vision according to claim 5, characterized in that, The image processing calculation module is also used for: According to the second preset period, the third recognition accuracy of the LAB traffic threshold template is calculated. If, within the current period, the duration for which the third recognition accuracy is less than the second preset accuracy reaches the second preset duration, adaptive learning is triggered. Collect the second set number of second valid images within the current period, and recalibrate the LAB color space threshold range using the maximum inter-class variance method to generate an LAB flow threshold adjustment template and enter trial mode; Within the usage mode, if the fourth accuracy of the LAB traffic threshold template is greater than or equal to the preset accuracy, replace the LAB traffic threshold template. If the fourth accuracy rate is still less than the preset accuracy rate, repeat the optimization until the second stop optimization condition is met.

7. The intelligent shell-breaking and blanking system based on machine vision according to claim 1, characterized in that, The tank control machine is specifically used for: Establish a correlation model between pixel area and set shelling depth; When the pixel area is detected to exceed the preset upper limit threshold, it is determined that the shelling is too deep, and an adjustment command is sent to the shelling execution mechanism to reduce the set shelling depth in a gradient manner; the preset upper limit threshold is determined by the aluminum electrolysis production process parameters. When the detected pixel area is lower than the preset lower limit threshold, an adjustment command is sent to the shelling execution mechanism to increase the set shelling depth; the preset lower limit threshold is determined by the aluminum electrolysis production process parameters.

8. The intelligent shell-breaking and blanking system based on machine vision according to claim 1, characterized in that, The tank control machine is also used for: Real-time acquisition of aluminum electrolysis production process parameters; the aluminum electrolysis production process parameters include cell temperature, electrolyte molecular ratio, electrolyte level and aluminum level; Substitute the bath temperature, electrolyte molecular ratio, electrolyte level, and aluminum level into the preset nonlinear mapping formula to calculate the preset upper limit threshold and the preset upper limit threshold.

9. The intelligent shell-breaking and blanking system based on machine vision according to claim 1, characterized in that, The tank control machine is also used for: The material flow rate is compared with the preset first process requirement value, and the total material flow rate is compared with the preset second process requirement value. If the material flow rate is lower than the first process requirement value, or the total material flow rate does not reach the second process requirement value, the material flow interval time is shortened; wherein, the shortening rate is dynamically adjusted according to the first deviation ratio between the material flow rate and the first process requirement value, and the second deviation ratio between the total material flow rate and the second process requirement value, and the maximum shortening rate is less than the preset shortening rate. If the material flow rate is higher than the first process requirement, or the actual material flow exceeds the second process requirement, extend the material flow interval.

10. The intelligent shell-breaking and blanking system based on machine vision according to any one of claims 1-9, characterized in that, Also includes: Maintenance management device, used to output maintenance strategies; The maintenance strategy includes: Routine maintenance includes shining a strong flashlight on the lens surface of the image acquisition module to observe whether there is any accumulation of dirt such as dust and electrolyte splashes; if the lens transparency decreases and affects the image clarity, gently wipe the lens surface with special lens paper dipped in anhydrous ethanol, and do not use hard objects or ordinary paper towels to wipe it; after cleaning, preview the image through the image processing and computing module to confirm that there are no dirt residues on the lens and the image is clear. The air-cooling unit and temperature monitoring project includes checking the pressure gauge reading of the cooling unit to ensure that the air pressure is stable within the preset pressure range. If the air pressure is lower than the lower limit of the preset pressure range, contact the air compressor station to adjust the air supply pressure. Use an infrared thermometer to detect the temperature of the active detection module shell and the image acquisition module body to ensure that the surface temperature is lower than the first preset temperature. If the surface temperature exceeds the second preset temperature, check whether the vortex cooling pipe is blocked and whether the air curtain outlet is unobstructed. Disassemble and clean the impurities inside the cooling pipe. The weekly special maintenance insulation performance retest project includes using an insulation resistance tester to retest the insulation resistance between the camera housing and the fixed frame to ensure that the insulation resistance is greater than the preset resistance. If the insulation resistance is less than or equal to the preset resistance, check whether the insulation bracket and gasket are aged or damaged, and replace the failed parts in time. The communication link inspection items include logging into the image processing and computing module backend to check whether the camera data transmission packet loss rate is less than the preset packet loss rate; at the same time, check the sealing of the network cable connectors, and if the sealant is found to be aged and peeling off, reapply high-temperature resistant sealant for protection. Maintenance records and anomaly handling items include establishing a system maintenance log, which records in detail the time, maintenance items, test data and anomalies of each maintenance; if serious scratches on the lens or equipment temperature that is continuously exceeded and cannot be resolved on-site is found, the system is shut down and reported to the equipment management department, and backup monitoring equipment is activated at the same time.

Images