Carbon block number identification system

Abstract

The invention relates to a carbon block number identification system which is used for automatically identifying the number of an anode carbon block in electrolytic aluminum production. The system comprises an image acquisition module used for acquiring an image of a carbon block surface number; the recognition algorithm module is used for processing the image by using technologies such as deep learning to recognize the number of the carbon block; the platform linkage module is used for interacting an identification result with a factory data platform to realize data uploading and association; the result pushing module is used for sending the identified number information to the user terminal or the display equipment in real time; and the data export module is used for storing the identification data and supporting export analysis. According to the system, the whole process of image acquisition, intelligent identification, platform synchronization, result pushing and data export is automatic, the number of the carbon block can be accurately identified and the data can be immediately shared in the field environment of high temperature dust and the like, the management efficiency of the electrolytic aluminum anode carbon block is remarkably improved, and the system has the advantages of high identification accuracy, strong real-time performance, industrial application and the like.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of industrial image recognition and production process automation, and particularly relates to a system for automatic recognition of anode carbon block numbering in electrolytic aluminum production. The system comprehensively uses computer vision and data communication technology to realize automatic reading and information processing of the numbering of carbon blocks (anode carbon blocks) for electrolytic aluminum. BACKGROUND

[0002] In the process of electrolytic aluminum production, a large number of anode carbon blocks are used in electrolytic cells and produce corresponding production and quality information, and each carbon block needs to be numbered and tracked for subsequent production links and flow processes. After forming, the anode carbon block needs to go through high-temperature baking (about 1200℃) and cleaning and assembly processes. Because the baking environment temperature is extremely high, common printing and labeling, laser etching, RFID radio frequency identification and other methods cannot be used in this environment. In industry, a steel stamp is usually used to press and extrude a number mark on the surface of the carbon block. Although this pressure marking method is resistant to high temperature and does not affect the performance of the carbon block, due to the low reflectivity of the carbon block material itself and the possible contamination of the mark surface in the cleaning process, it is difficult to achieve automatic recognition of the pressure marking number under existing conditions. The traditional planar image recognition method has strict requirements on the lighting conditions and has poor recognition reliability in the complex light field environment of the electrolytic aluminum carbon block site. Therefore, at present, the recognition of the carbon block number mainly relies on manual reading, which is low in efficiency and prone to errors, and cannot meet the real-time data acquisition needs of modern large-scale electrolytic aluminum production.

[0003] In order to improve the automation level of anode carbon block numbering recognition, some technical solutions have been proposed. For example, some research uses three-dimensional vision combined with deep learning to recognize the steel stamp number on the surface of the carbon block: by spraying developing agent on the mark to improve the contrast, using a depth sensor to obtain three-dimensional point cloud data of the steel stamp, and then inputting the processed projection image into a convolutional neural network model to recognize the number. The offline trained deep learning model can achieve high recognition accuracy and stability, meeting the requirements of industrial production. However, such a solution is complex in equipment and process, high in cost, and has not yet formed a complete system with high integration and easy deployment in the production site. The existing technology lacks a solution that closely integrates the electrolytic aluminum carbon block recognition scene and organically integrates image acquisition, intelligent recognition, platform data linkage and result output. Therefore, it is necessary to provide a new carbon block numbering recognition system to realize automatic acquisition, recognition and synchronous sharing of carbon block numbering information, and improve the informatization level of electrolytic aluminum production. SUMMARY

[0004] The present application aims to overcome the shortcomings of low efficiency and insufficient automation in the prior art, and provides a carbon block number recognition system. The system can automatically collect carbon block number images in the electrolytic aluminum production site, accurately recognize carbon block numbers, and push and upload the recognition results to the management platform in real time, thereby realizing the automatic management of the life cycle data of anode carbon blocks.

[0005] To achieve the above-mentioned purpose, the present application provides a carbon block number recognition system, comprising: An image acquisition module: acquires images of the surface numbers of carbon blocks. An industrial camera or special imaging device is used to take pictures of carbon blocks (such as electrolytic aluminum anode carbon blocks) and capture images of their number areas. This module can be fixedly installed at a suitable position on the production line or realized by a mobile device carried by a person, and the focal length, shutter, and light supplement are adjusted to adapt to the on-site environment.

[0006] An identification algorithm module: processes and analyzes the collected images to identify carbon block number information. This module uses optical character recognition (OCR) technology and deep learning algorithms to identify digital characters in the images. To improve accuracy, it can include a pre-processing unit (such as gray scale enhancement, noise filtering, image correction, etc.) and a trained convolutional neural network model to cope with challenges such as low reflectivity and poor contrast on the surface of carbon blocks. The identification algorithm is trained offline with a large number of samples, and has the stability and accuracy required in industrial sites.

[0007] A platform linkage module: an interface module for data interaction with the background data management platform. This module uploads the carbon block numbers extracted by the identification algorithm module to the existing information system or database of the factory, and can query or update the production and quality information corresponding to the carbon blocks according to the numbers. For example, the platform linkage module can call the API of the manufacturing execution system (MES) or other databases to associate the newly identified carbon block numbers with their production batch, furnace number, etc., and realize the linkage of the production process.

[0008] A result pushing module: a module that sends the recognition results to relevant users or devices in a timely manner. The recognized number information is pushed to the monitoring center interface, the operator's handheld terminal, or the on-site display screen via wired or wireless networks, realizing real-time notification. For example, when the system recognizes the carbon block number, the result pushing module can pop up a prompt on the screen in the control room, or prompt the staff through sound and light alarm that a specific carbon block has been identified for subsequent operation.

[0009] Data export module: a module that stores, aggregates, and provides export functions for the identification data. This module saves all identification records to local storage or a cloud database and supports retrieval queries based on time periods, batches, and other conditions. Users can export the carbon block number recognition results within a specified range to a report or a standard format file (such as CSV or Excel) for production statistical analysis or quality traceability through the data export module.

[0010] Through the organic combination of the above modules, the system of the present application constructs a complete process from data acquisition to information processing, and then to result output and archiving. In terms of working principle, the image acquisition module obtains the image of the carbon block number, which is then subjected to intelligent recognition by the recognition algorithm module to obtain the digital number; the platform linkage module enables the identification data to be synchronized with the existing platform database; the result pushing module notifies the relevant personnel in real time of the key results; and the data export module ensures long-term data storage and flexible extraction. This system can reliably operate in harsh industrial site environments, achieving automatic acquisition and sharing of carbon block number information, and greatly reducing manual intervention. Compared with the prior art, the present application has the following beneficial effects: Improving recognition efficiency: the system automatically completes the reading of the carbon block number, greatly improving the data acquisition efficiency compared with manual one-by-one recording.

[0011] High recognition accuracy: the well-trained deep learning algorithm can accurately recognize low-contrast carbon block numbers, reducing the misreading rate.

[0012] Strong environmental adaptability: the imaging scheme and algorithm are robust and can adapt to high-temperature, dust, and light changes in the aluminum plant site environment to achieve stable operation.

[0013] High platform integration: the recognition results are automatically uploaded to the production management platform, eliminating manual input, realizing data linkage, and facilitating subsequent analysis and decision-making.

[0014] Timely information feedback: through the result pushing function, the on-site personnel can obtain the identification information in real time, handle abnormal situations in a timely manner, and improve safety and management efficiency.

[0015] Data traceability: all identification data are traceable and can be easily exported and backed up, meeting the needs of production traceability and quality management.

[0016] Preferably, the image acquisition module includes an industrial digital camera and a light supplement device, the camera is installed on a fixed bracket above the carbon block conveying path or above the side, and automatically captures the image of the number area when the carbon block passes through the identification position; the light supplement device is used to improve the contrast and edge visibility between the shallow relief / embossed characters and the background.

[0017] By reducing perspective distortion at vertical or near-vertical viewing angles, using a global shutter to suppress motion blur; using a pulsed stroboscopic flash synchronized with a trigger to momentarily increase the illumination within the camera's exposure window, and using a polarizer to suppress reflections. Optionally, an external trigger (photoelectric sensor / encoder) can be used to trigger the exposure when the carbon block enters the recognition window, ensuring that the shooting position and focal plane are stable.

[0018] This scheme significantly improves the clarity and stability of the image under adverse conditions such as dust, reflections, and low contrast, ensuring the distinguishability of the numbered strokes, providing high-quality input for subsequent algorithm recognition, and reducing the probability of false and missed collection.

[0019] Preferably, the recognition algorithm module uses a pre-trained deep learning OCR model, preferably a two-stage detection + sequence recognition: first use the detection network to locate the numbered ROI, then use the convolution-recurrent (or attention) structure to perform end-to-end sequence recognition; and set up an image preprocessing submodule (denoising, contrast enhancement, geometric correction).

[0020] Through data augmentation (blur, low contrast, stains, partial occlusion, etc.) on numbered samples, offline training obtains parameters that are robust to on-site interference; when online, perform adaptive histogram equalization / gamma correction on the collected images to eliminate uneven lighting; the detection network outputs the numbered boundary, and the recognition network decodes the character sequence in CTC / Attention mode, and performs compliance verification in combination with the character set and regular template.

[0021] Maintain high recognition accuracy and stability in low-contrast, reflective, and lightly soiled scenes; the two-stage architecture enhances the interpretability of positioning-recognition, facilitating online maintenance and special optimization.

[0022] Preferably, the platform linkage module interfaces with the production management system (such as MES) through wired Ethernet or industrial wireless, uses standard protocols / interfaces (REST API, OPC, MODBUS, any one or combination) to complete recognition result upload, numbered uniqueness verification, and bidirectional association and backfilling with process / quality data.

[0023] The platform linkage module generates event messages (number, timestamp, device ID, picture URL / binary) on the edge side, writes them into the platform in a transactional manner; the platform returns associated information (such as batch, specification, quality inspection conclusion), and the platform linkage module binds the backfilling information to the event and archives it. To ensure reliability, support breakpoint caching / retransmission and signature authentication.

[0024] Build a closed-loop data channel between recognition and business systems to automatically associate and synchronize numbered and production data throughout the entire process, reduce manual entry errors, and improve traceability and auditability.

[0025] Preferably, the result push module includes an HMI interface and an alarm unit. After successful recognition, the number and its backfilled information are displayed in real time on the HMI; when the recognition confidence is low, the rule verification fails, or there is no corresponding record in the database, an audible / visual / voice alarm is triggered, and an entry point for manual review and a reason code selection are provided on the interface.

[0026] The results push module renders the UI according to event level: normal items are green, items requiring review are yellow, and abnormal items are red; at the same time, deletion / correction operations and reason codes are entered into the database. The alarm unit drives a tri-color light / buzzer through the DO port.

[0027] It enables second-level feedback and operable verification, moves the risk of misidentification / omission to the site for closure, shortens the time for handling anomalies, and improves operational safety and cycle stability.

[0028] Preferably, the data export module will structure and store the identified records according to dimensions such as time / production line / shift (either relational database or time series database is acceptable), and provide CSV / Excel batch export and standard data interface for external analysis systems to call.

[0029] The data export module performs column mapping and deduplication verification on the query results, supports scheduled tasks (daily / weekly / monthly reports) and incremental export, and verifies / signs the exported files. Optional encrypted transmission to a specified email address or shared directory is also available.

[0030] Standardized data assets facilitate statistical analysis, reconciliation, and quality traceability, reduce manual processing costs, and improve management transparency. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of an embodiment of the present invention. Detailed Implementation The following detailed description illustrates the specific implementation method: like Figure 1 As shown, the carbon block numbering and identification system of the present invention includes: an image acquisition module, an identification algorithm module, a platform linkage module, a result push module, and a data export module. The functions of each module are as described in the invention content section above. The following describes in detail the process of the collaborative work of each module in practical applications: Image Acquisition: When a carbon block to be identified and numbered enters the identification location, the image acquisition module is activated. This module uses an installed industrial camera to photograph the numbered area on the surface of the carbon block. To ensure image quality, the acquisition device can be equipped with supplementary lighting or employ measures such as filtering and dust prevention to overcome adverse factors such as the low reflectivity of the carbon block surface and complex ambient lighting. The acquired raw image data is temporarily stored in the local processing unit, awaiting further processing.

[0032] Number recognition: After image acquisition is completed, the recognition algorithm module receives the image data and begins image recognition processing. First, the recognition algorithm module pre-processes the image, including grayscale conversion, contrast enhancement, denoising, and distortion correction, etc., to highlight the texture features of the carbon block number. Then, the pre-processed image is input into the pre-trained OCR recognition model for analysis. Preferably, the present application uses a convolutional neural network model of deep learning to recognize digital characters. This model is trained on carbon block imprint digital samples and can accurately and stably output the number recognition result (e.g., recognizing the number as "A123" or "B-56" in the form of a marker). The recognition algorithm module can further check the result, such as filtering abnormal recognition values according to the number rule (containing only numbers or letter combinations), or comparing with the carbon block production batch information to improve accuracy.

[0033] Platform linkage: When the recognition algorithm module obtains the carbon block number, the platform linkage module immediately transmits the number information to the data management platform in the background through wired network or wireless communication. The platform linkage module acts as an intermediate interface, enabling the system to integrate with existing factory information systems. For example, the platform linkage module calls the interface of the factory MES system, uploads and registers the newly recognized carbon block number into the database; at the same time, the platform returns the carbon block attribute data corresponding to the number (such as production time, carbon block specifications, quality test results, etc.) for the system to record or display. If the platform requires to review the recognition result, the platform linkage module can also receive the feedback information from the platform (such as number validity confirmation instruction), thereby realizing platform interactive linkage. The entire process is completed in real time in a network environment, ensuring that the carbon block number data is synchronized into the production management process.

[0034] Result pushing: At the same time of data uploading, the result pushing module pushes the recognition result of the carbon block number to the relevant users in real time. Specifically, this module can notify different roles of the number and associated information through various ways: for the on-site operator, a prompt can be popped up on the handheld terminal APP, displaying "carbon block number X recognition successful"; for the control room monitoring screen, the number can be added to the real-time list or highlighted; for the management personnel, they can receive the push notification through SMS, email, etc. In this embodiment, if the recognized number does not match the expected one (e.g., the system cannot find the number in the database), the result pushing module will also send an alarm prompt to correct it in time. Through this module, the recognition result can reach the personnel within seconds, greatly improving the timeliness and transparency of information transmission.

[0035] Data export: Carbon block number and related identification data are recorded and stored by the data export module in the background. This module adds metadata such as timestamps, device IDs, etc. to each identification record to form a complete log record in internal storage or a cloud database. Users can query historical identification data according to their needs through the interface provided by the system, filtering conditions such as date range, furnace, shift, etc. and exporting the query results to Excel or CSV files by the data export module. This process is usually triggered manually by the user at the end of the shift or day to generate a carbon block identification report for the current shift. However, the system also supports automatic timed export functions, such as automatically summarizing all identification data for the day and sending it to a designated email at midnight. The data export module ensures the traceability and secondary use value of carbon block number identification information, providing convenience for subsequent production statistics and quality analysis.

[0036] Example 2: Device and installation of image acquisition module Camera and lens: Select a 2-5 million pixel industrial camera (global shutter, frame rate ≥25 fps), C port lens focal length 8-16 mm; install 3-5 m from the surface of the carbon block.

[0037] Light compensation and polarization: Linear array / surface array LED light compensation, supports stroboscopic (10-50 μs), and camera exposure is synchronized by external trigger; adjustable polarizer is added at the lens end to suppress reflection.

[0038] Trigger method: Set a light emitting diode or use a conveyor encoder pulse in front of the identification position, trigger exposure when the leading edge of the carbon block reaches the threshold position, and achieve equidistant sampling.

[0039] Installation and protection: Adjustable bracket with pitch / yaw scale and locking; covered with a dustproof cover, the transparent window uses tempered glass + sealing ring; cable bridge and shielded ground.

[0040] Parameter setting: Exposure time 0.2-2 ms, gain ≤6 dB; white balance fixed; resolution not less than 0.2 mm / pixel; field of view covers the complete numbering area.

[0041] Reproducibility explanation: According to the above configuration, clear, stable, and low trailing image acquisition of the numbering is achieved in a typical indoor line.

[0042] Example 3: Flow of identification algorithm module Preprocessing: Distortion correction, ROI cropping, CLAHE adaptive enhancement, and non-local mean denoising are performed on the collected images.

[0043] Number positioning: Use a lightweight detection network (such as YOLO-tiny / PP-PicoDet) to output number box coordinates; set a minimum box height threshold and aspect ratio constraint to eliminate non-targets.

[0044] Sequence recognition: Normalize ROI to HxW (e.g., 32x192), input CRNN / Attention model; decode to get character sequence, character set configured by project rules (number or letter + number).

[0045] Rule verification: Generate Top-K candidates with regular template (e.g., [A-Z]{1,2}\d{5,6}) and minimum edit distance; low confidence or mismatch is marked as “need review”.

[0046] Deployment and update: Model is deployed in container mode, exposing HTTP gRPC interface; supports gray release and A / B comparison.

[0047] Reproducibility explanation: Offline training with no less than 50,000 enhanced numbered samples, online inference latency single frame ≤ 60 ms (industrial x86 / ARM edge box).

[0048] Example 4: Docking of platform linkage module Interface protocol: Prefer REST API: submit event (number, time, device ID, picture URL / binary); platform returns, etc.

[0049] OPC / MODBUS (optional): Write event status word (NEW / OK / REVIEW / FAIL) in register, platform can subscribe to changes.

[0050] Reliable transmission: Enable local cache queue, persist when network is abnormal; retransmit by timestamp after recovery; request header carries Token / signature.

[0051] De-duplication logic: Same device ID + time window ± 3 s + same number of events are considered repeated, merged and reference count is recorded.

[0052] Reproducibility explanation: Communicate with MES in HTTPS+Token mode in enterprise intranet, still stable when daily event volume > 50,000.

[0053] Example 5: Result pushing module and HMI / alarm HMI interface: List displayed in descending order of time (number, confidence, batch, quality inspection); color coding: green (normal), yellow (need review), red (abnormal); click to enlarge original picture and ROI.

[0054] Alarm strategy: Confidence < threshold, regular does not pass or MES has no such number, trigger three-color light + buzzer (1-3s), and pop up review dialog box (manual correction + reason code).

[0055] Multi-terminal push: push to the central control large screen and APP through WebSocket / message queue; abnormal events can be notified by SMS / email.

[0056] Reproducibility explanation: on-site personnel obtain feedback in seconds and can review and revise on-site, review operations and reason codes enter the event audit chain.

[0057] Example 6: Data export module and report Data storage: MySQL / TimescaleDB structured table design; key fields include block_no, device_id, ts, batch, qc, status, operator.

[0058] Export capability: front-end selects time period / production line / shift for screening, and exports as CSV / Excel (.xlsx); column headers include number, time, device, batch, quality inspection status, whether to review, and operator.

[0059] Scheduled task: configure automatic export of daily / weekly / monthly reports; file is signed (SHA256) and delivered to a shared directory / mailbox.

[0060] Interface opening: GET / api / v1 / blocks / report?from=...&to=...&line=... returns CSV stream for external BI / statistical system to pull.

[0061] Reproducibility explanation: can seamlessly access BI systems in a regular enterprise IT environment, meeting statistical and traceability requirements.

[0062] In summary, the carbon block number recognition system of the present application realizes automatic, high-precision recognition of electrolytic aluminum anode carbon block numbers, as well as instant sharing and long-term storage of data through the cooperative work of the above-mentioned modules. The system has clear structure and complete process, and can run stably in actual industrial environment, significantly improving the informatization and intelligent level of the electrolytic aluminum production process. The embodiments of the present application have been described above in combination with the drawings, but the present application is not limited to the specific examples described. Those skilled in the art can make various modifications or equivalent replacements without deviating from the principles of the present application, and these should be considered as falling within the scope of protection of the present application.

Claims

1. A carbon block numbering and identification system, comprising an image acquisition module, an identification algorithm module, a platform linkage module, a result push module, and a data export module, characterized in that: The image acquisition module is used to acquire image information of the surface number of the carbon block of the electrolytic aluminum anode; The recognition algorithm module is used to process and analyze the image information to identify the number characters of the carbon blocks; The platform linkage module is used to communicate and interact with the identified numbering information and the backend data platform to realize the uploading and association of carbon block numbering data; The result push module is used to send the recognition results to the user terminal or on-site display device in real time for prompting. The data export module is used to store the identification result data and provide an export interface to output the carbon block number identification record in a predetermined format.

2. The carbon block numbering and identification system according to claim 1, characterized in that, The image acquisition module includes an industrial digital camera and a supplementary lighting device. The camera is installed next to the carbon block conveying path and can automatically capture images of the numbered areas of the carbon blocks as they pass by. The supplementary lighting device improves the clarity and contrast of the acquired images.

3. The carbon block numbering and identification system according to claim 1, characterized in that: The recognition algorithm module uses a pre-trained deep learning model for character recognition. The model is a convolutional neural network (CNN) architecture, which is trained offline on digital sample data imprinted on the surface of electrolytic aluminum carbon blocks. It can maintain recognition accuracy under conditions of light changes and dirt.

4. The carbon block numbering and identification system according to claim 1, characterized in that: The platform linkage module connects to the carbon block production management system via wireless network or wired Ethernet, and uses standard protocols to upload the identified number to the Manufacturing Execution System (MES) database, and retrieves the corresponding process and quality information for display and storage.

5. The carbon block numbering and identification system according to claim 1, characterized in that: The result push module includes a human-machine interface and an alarm unit. When the carbon block number is successfully identified, the human-machine interface displays the number and its associated information in real time. The alarm unit issues a warning signal when the identification result is abnormal to remind the on-site operator to check.

6. The carbon block numbering and identification system according to claim 1, characterized in that: The data export module can export the stored carbon block number identification data in CSV file or spreadsheet format, and supports interface with external data analysis systems through standard data interfaces, which facilitates statistical analysis of carbon block production and quality data.

Images