As the last process of the iron and steel process, cold rolling has the characteristics of various types of products, high product quality requirements, long process flow and complex process flow. There is an urgent need for cold rolling plants to increase production, improve quality, facilitate operation and maintenance, flexible production, and reduce costs. , to ensure safety and environmental protection needs.
Compared with human vision, machine vision has obvious advantages in accuracy, adaptability, objectivity, repeatability, reliability, efficiency and information integration. It has been widely used in consumer electronics, automobiles, medicine and steel and other fields , resulting in significant benefits.
Applying machine vision to cold rolling equipment can realize the unmanned cold rolling unit, which can optimize the operation and maintenance of the unit, reduce operating costs and better ensure safety.
1. Cold rolling scene
Cold rolling mills can be divided into various types according to steel types and processes. Taking silicon steel normalized pickling unit as an example, its typical production process is coiling→uncoiling→straightening→head cutting→welding (edge cutting)→entry work Jacket→preheating and non-oxidation heating→soaking→water jacket cooling→water spray cooling→water spray cooling→drying while blowing→shot peening descaling→pickling→rinsing→drying while blowing→export looper→side Internal heating→edge cutting→crushing→inspecting→slitting→coiling→weighing→baling→storage.
At present, the process section of the unit has been unmanned, and the demand for unmanned operation is mainly concentrated in the entrance section and exit section of the unit.
At the unit entrance section, raw steel coils are transported by car from the hot rolling mill to the new normalized pickling bay for storage. Coils are sent to No.1 or No.2 uncoilers. The head of the unit is equipped with a five-roll pinch roll leveler, and the steel strip is introduced into the pinch roll leveller through the uncoiler, and the strip head is pinched and straightened. Then the strip is sent to the entrance double-layer shear, and the head and tail of the strip are cut off, and the head of the strip after shearing reaches the welding machine, and is welded with the tail of the previous strip.
At the exit section of the unit, the steel strip is first trimmed by disc shears, and the cut waste is cut into small pieces by the scrap, and then transported to the operating side of the unit by the belt conveyor, and the waste falls into the waste collection device. After trimming, the strip is sheared head to tail and then enters the exit and turns to pinch rolls for tension coiling. The rolled steel coil is unloaded by the export steel coil trolley, weighed and bundled on the saddle, and the steel coil is hoisted to the Changhua back warehouse by the workshop crane.
2. Machine vision application realization path
2.1 Application category
Combing the unmanned application categories of cold rolling mills, the above application scenarios can be classified into four categories
(1) Image classification
Weld seam quality inspection, entrance belt jamming inspection, broken edge shearing edge inspection, strip surface quality online inspection, exit coil unloading core pulling inspection
(2) Target detection
Strip steel incoming quality inspection, saddle steel coil offset inspection, width and height alignment inspection, weld quality inspection, strip deviation inspection, disc shear quality inspection
(3) Image segmentation
Width and height centring detection, disc shear quality detection, strip steel surface quality online detection
(4) Text recognition
Strip steel incoming quality inspection
2.2 Implementation Architecture
The application is mainly composed of detection objects, actuators, imaging components, computer hardware, camera SDK, middleware and unmanned machine vision application software.
In the unmanned application scenario of the cold rolling mill, the detection object is mainly strip steel, including steel coils, inner rings of steel coils, strip welds, and strip edge wires. The actuator is typical equipment in the cold rolling mill, such as the coiling trolley, the double-layer shear, the welding machine, the correction roller and the disc shear, etc. Based on the machine vision detection results, the actuator acts on the detection object to realize efficient, stable and safe production of the unit.
Imaging components mainly include a light source, camera and lens. The main function of the light source is to illuminate the detection object, reveal defects from the background image, overcome ambient light interference, and ensure image stability. The main function of the lens and the camera is to convert the optical signal into an electrical signal, and then into a digital signal image. Cameras are divided into 2D cameras and 3D cameras, and 2D cameras are further divided into line scan cameras and area scan cameras.
Computer hardware includes image processing and storage server and client computer. The server is equipped with a high-performance graphics card for image processing and a large-capacity hard disk for historical image storage. Through the camera SDK, realize image data acquisition and configure camera parameters. Middleware performs image processing and storage. For image processing, common middleware includes OpenCV, HALCON, PaddlePaddle, TensorFlow, PyTorch, and MindSpore. For data storage, common databases include MySQL, SQL Server, and Oracle. Unmanned machine vision application software is a human-computer interaction interface, which realizes visual operation and interaction.
2.3 Application practice
At present, in the field of cold rolling, China Metallurgical South has realized the typical application of unmanned cold rolling mills and achieved good results.
1) Detection of broken edge shearing and blocking edge. Broken edge shearing and edge detection uses high-definition industrial cameras to capture the shape of edge wires in real time and detects them from multiple aspects such as spatial position, shape, and motion trajectory. After detecting stacking, escape, blockage, etc., an abnormal signal is sent to the PLC through I/O, and a voice alarm is issued to realize chain control.
2) Online detection of strip surface quality. On-line detection of strip steel surface quality reads images from the camera, preprocesses images, segments defects, extracts features, classifies typical defects and stores defects to the server, and expresses the defect data stored in the server in a visual way, real-time and intuitive Information such as the location and category of defects on the upper and lower surfaces of the strip is displayed, and suggestions for process improvement and alarm prompts are given for the defects that are focused on.
For the cold rolling scene, on the basis of sorting out the production process of the unit and the requirements of unmanned application scenarios, machine vision technology is used to realize height alignment detection, looper deviation detection, disc shear strip shear ratio detection, Applications such as broken edge shearing edge detection and strip surface quality online detection optimize unit operation and maintenance, reduce operating costs and better ensure safety, with significant benefits, creating a new era for unmanned operation and maintenance and remote centralized control of cold rolling units good conditions.