With the development of intelligent technology, machine vision applications have penetrated various industries and become the enabling technology of smart factories. At present, metallurgical enterprises are also actively exploring and vigorously promoting the application of machine vision. With the help of multi-scenario machine vision smart applications, they can realize intelligent production and improve quality and efficiency, and accelerate their own transformation to intelligent manufacturing.
The machine vision intelligent perception platform, a series of vision applications, has been put into use in the long product workshop application scenarios of many steel companies such as Linyi Iron and Steel Investment Special Steel, Kunming Iron and Steel, Lingyang Iron and Steel, and Yixin. Through machine vision and basic automation depth Integration helps enterprises to build smart workshops, improve production automation, reduce labor intensity and improve production efficiency.
Edge visual perception platform architecture
The machine vision smart perception platform builds machine vision smart applications based on edge technology and creates an integrated set of edge-end machine vision products of “hardware + software + algorithm” on the field application side to achieve hardware standardization, software modularization and algorithm customization. Among them, hardware includes image acquisition equipment (industrial cameras, security cameras, lidar, stereo cameras, etc.) ) preprocessing, the release of analysis results, issuance of recognition control instructions, data localization and other core support functions. Algorithms include image processing algorithms, deep learning algorithms, point cloud analysis algorithms, etc. for visual recognition in specific business scenarios. calculation function. The platform adopts a modular design concept for kernel development, supports secondary development of the application layer, and customizes the application of machine vision in specific scenarios.
Edge-end machine vision products can achieve faster data processing response, calculation and analysis response, and network response to meet the business needs of low latency and high response in metallurgical production. Debugging and rapid application greatly shorten the landing cycle and application cost of machine vision smart applications.
The edge-end machine vision products in different application scenarios can be connected to the cloud server, and the analysis data of the vision edge-end can be aggregated to the server, supporting the construction of a dynamically expandable workshop-level intelligent monitoring/control application system.
| Equipment | Stereo Camera | Laser Radar | Security Camera | Industrial Camera | ||||||
| Access Layer | UDP | TCP | SDK | GIGE | RTSP | |||||
| Data Layer | Point Cloud Data | Video Data | Image Data | |||||||
| Algorithm Layer | Deep Learning | Point Cloud Analysis | Image Processing | |||||||
| Application Layer | Bent Steel Inspection | Billet Number Identification | Stack Steel Detection | Scale Alignment Recognition | Collection Area Visual Tracking | |||||
| C-type Hook Number Recognition | Billet Measurement | Skid Recognition | billet Detachment Detection | Intelligent Crane Identification System | ||||||
| Release Layer | RestAPI | Socket | MQ | OPC | ||||||
Application in the whole process and multiple production links
The machine vision intelligent perception platform is deeply integrated with basic automation to realize billet number recognition, bending steel/off-square/length measurement detection, slip detection, piled steel detection, alignment recognition of bar fixed-length baffles, visual tracking of bar collection areas, C Hook recognition and other functions are widely used in various production links to improve the level of production automation, effectively reduce production costs, save labour, improve production efficiency, and reduce product defective rate.
(1) billet number identification
Through the target detection and OCR character recognition algorithm, the casting slab spray number characters on the roller table in front of the furnace are recognized, and the information is checked with the rolling plan according to the spray number recognition results to ensure that the billet entering the furnace is consistent with the production plan, realizing steelmaking and The rolling information is correct and true, and the purpose of material tracking by branch and root is achieved.
(2) Slab bending/detachment/length measurement detection system
Through the machine vision recognition method, calculate the bending rate, cross-sectional shape and size, and the length of the steel billet in the two dimensions of the length direction of the steel billet to be furnaced. If there are bending steel, off-square, and over-length conditions, an alarm will be issued in time, and it will be linked with the PLC control system. , realize the automatic removal of abnormal billets, and avoid unqualified billets from entering the heating furnace. Among them, when the bending degree of the billet is greater than 20mm per meter and the total bending degree is greater than 2% of the total length, the detection accuracy rate is greater than 99%. Labor intensity, to avoid production accidents.
(3) Slab slip detection system
During the production process of long products, billet slipping occasionally occurs during rough rolling, and with the promotion of centralized production control, when billet slips, it is difficult for operators to detect and deal with it in time. The system uses industrial cameras to monitor the rough rolling entrance area, and based on the self-developed algorithm, conducts a real-time analysis of the steel billet’s biting image frame, quickly detects the slippage of the billet, and promptly notifies the production personnel to intervene, significantly reducing the operator’s work Strength, improve the timeliness of exception handling.
(4) Heap steel detection system
When a stacking steel accident occurs in long product production, especially when flying steel occurs, it will need to be shut down for processing, which will seriously affect the production rate.
The system uses industrial cameras to monitor the frame area and the looper, and based on the self-developed algorithm, it conducts a real-time analysis of the image frames of the rolling mill and its front and rear areas and the looper and combines the self-developed production line early warning and diagnosis expert system to realize Steel stacking warning and stacking alarm are linked with the PLC control system. When the system judges the trend of stacking steel, the flying shears will start breaking to avoid accidents and significantly improve the production rate.
(5) Alignment recognition system for bar sizing baffles
During the bar production process, at the technological position where the fixed-length baffles collide, real-time and automatic alignment status detection cannot be performed, resulting in the inability to realize fully automatic operation in the shearing area, and manual visual inspection and on-site command must be relied on. The system detects the alignment status of the bars in real-time through machine vision + AI, confirms whether the alignment is completed, and transmits the recognition results to the PLC control system in real-time to realize automatic cutting of double-scale incoming materials and improve the production rhythm of the cutting line. Reduce the short-length problem caused by the misaligned cutting of the head, and solve the bottleneck problem that restricts the automatic production of this process section.
(6) Visual tracking of the bar collection area
Due to the large number of finished products, unstable detection components, and manual intervention in the bar collection area, it is often difficult to achieve accurate and fully automatic material tracking. Through deep learning + AI algorithm, the system dynamically identifies the position coordinates of the rolled piece on the stand and combines with the process control system to accurately track the bar materials in the finishing area, in order to realize the fully automatic operation between different subsystems in the finishing area. Integration (such as automatic number spraying, automatic listing, automatic steel sorting, etc.), strict mixed steel control, automatic or unmanned weighing and other applications provide basic guarantees.
(7) C-hook recognition
Using deep learning to realize the identification of the C-hook number is helpful to realize the whole process of material tracking, realize the accurate traceability of production logistics information, and reduce the manual confirmation and operation frequency of personnel. The system can recognize standard inkjet printing and handwriting and supports variable-angle recognition. The recognition accuracy of fonts that meet the specifications can reach 100%.
Significant quality and efficiency improvements
After the machine vision intelligent perception platform is put into use, the steelmaking and rolling data are accurately connected, and the production tracking management and quality tracking are truly realized; the labour intensity of the staff is greatly reduced, and the timeliness and accuracy of the operation are improved; the corresponding positions are streamlined At the same time, it solves the bottleneck factors that affect the improvement of operating rate and fault control under traditional conditions, and conducts capacity mining from specific scenarios, which greatly reduces production costs; improves the intelligence level of the production line, and realizes real-time monitoring through the application of machine vision + AI. Detection, intelligent early warning, instant alarm, auxiliary control, etc., in the case of unchanged or reduced production line capacity, personnel work efficiency is higher and production line control is better, realizing staff reduction and efficiency increase, providing technology for building an intelligent workshop ensure.
In addition, the machine vision intelligent perception platform has been successfully used in many application scenarios such as thick plate workshops and intelligent cranes. In the future, it will continue to focus on the research and development of intelligent technology in the industry and make outstanding contributions to the digital and intelligent development of iron and steel enterprises.