The production process of modern iron and steel industry is a complex and huge production system, covering hundreds of processes.
In the course of more than 70 years of development, various technologies such as steelmaking, steel rolling, continuous casting, energy saving and emission reduction have continuously evolved, all of which confirm the continuous breakthroughs made by China Iron and Steel on the road of technological innovation. Today, Baowu Zhongnan Iron and Steel (Baowu Group Zhongnan Iron and Steel Co., Ltd.) has achieved another innovative breakthrough through in-depth insight into a detail.
In the process of producing steel bars in iron and steel factories, computer control technology will be used to automatically measure and cut the steel bars. This process seems simple but extremely lethal. The robot arm in the subsequent process caused serious damage.
In response to this problem, Baowu Zhongnan Iron and Steel introduced AI technology to the person in charge of the intelligent project, and created a "reinforcement bar length overrun monitoring solution" based on the flying paddle image segmentation development kit Paddle Seg, which will be put into use in February 2023 It has been running stably in multiple production lines till now. This solution replaces the traditional method of manual verification by workers and masters, significantly improves production safety and production efficiency, and saves more than 500,000 yuan in cost for a single production line.
The person in charge of the intelligent project said, "After the project was launched, various indicators have excellent effects, the semantic segmentation evaluation index (mIoU) > 99%, and the abnormal/normal binary classification accuracy (Accuracy) is greater than 99.5%. We will consider using it in the future. Flying paddle -related technologies will continue to explore more intelligent scene solutions in the direction of safe production and intelligent identification quality inspection.”
Figure: Fully automatic steel bar overrun monitoring production line site
"Fire eyes and golden eyes", Baowu Zhongnan Iron and Steel has insight into the breakthrough of technological innovation and application
In the process of steel bar production, due to the randomness of the industrial environment, it is impossible to guarantee that each bundle of steel bars on the production line is within the specified length. The steel bars whose length exceeds the limit will cause damage to the mechanical arm on the production line, and the direct equipment loss and the indirect cost of lost work may cause huge economic losses.
The traditional method is that the steel factory arranges workers and masters to judge visually on the production line, and intervene in time if the length of the steel bar is found to exceed the limit. This method relies purely on manpower, requires high business proficiency, and has misjudgment phenomena, which cannot fundamentally solve the problem. At the same time, workers and masters cannot follow the production line to work 24 hours a day, which restricts production efficiency to a certain extent.
It is reported that for a single production line, with one worker and one robotic arm, the annual labor cost plus the operation and maintenance cost of the robotic arm exceeds 500,000 yuan.
The person in charge of the intelligent project said, "Perhaps from the perspective of the whole process of the iron and steel industry, the urgency of such problems is not so high. The task of using visual sensing equipment to monitor whether the length of steel bars exceeds the limit in real time seems relatively simple, but Traditional image processing methods cannot achieve sufficient accuracy and generalization. Now the new AI technology can solve this kind of problem very well, even if it is only applied in this scenario, it can save a considerable amount of money for the production line. Cost, and greatly improve production efficiency, create economic value."
Figure: Visual image of the fully automatic steel bar overrun monitoring solution
Based on the paddle image segmentation development kit Paddle Seg, "get" industrial vision intelligence
The accuracy and generalization of traditional image processing solutions cannot meet the requirements. Accuracy determines the practical application value of the solution, and generalization determines whether the solution can be copied to other scenarios. The person in charge of the intelligence project chose to introduce AI technologies such as deep learning to solve this problem. After investigating and comparing related technologies at home and abroad, the project leader believes that Baidu Flying Paddle's industrial-level deep learning platform has unique advantages.
The overall technical solution can be summarized as the following steps:
-
Install a camera on one side of the steel bar to take images;
-
Using image segmentation technology to extract steel mask;
-
Determine the length limit according to the camera position and angle;
-
Finally, judge whether this batch of steel bars exceeds the limit according to the geometric relationship between the length limit and the steel bar division range;
The problem of steel overrun monitoring can be transformed into a geometric judgment problem after image segmentation. Paddle provides a full-process segmentation solution - the image segmentation development kit Paddle Seg . Simply modify the configuration file to perform model training and obtain high-precision segmentation results. "Only about four or five hundred sheets were marked for data labeling, and then the model was developed. The effect is quite good and can meet the needs of the production line." Furthermore, the PP-LiteSeg model that uses the balance of accuracy and speed is selected to ensure that it meets the requirements of industrial deployment while achieving high precision.
What is important is that from the perspective of industry characteristics, the complex industrial environment makes it difficult to guarantee uniform standards and quality of graphic sampling and corresponding data. The built-in data enhancement algorithm of Paddle Seg can realize the expansion of data and overcome the imbalance of sampling samples, For problems such as blurred pictures, large differences in background transformation, inconsistent steel bar directions, and less data volume, quickly iterate solutions to shorten the model development cycle.
At present, Baowu Zhongnan Iron and Steel has applied the "full-automatic steel bar length overrun monitoring solution" developed based on flying paddles to two production lines to solve the automatic monitoring problem of steel bar length overrun and realize end-to-end fully automatic steel bar length overrun monitoring . Limited monitoring, semantic segmentation mIoU > 99%, abnormal/normal binary classification accuracy (Accuracy) greater than 99.5%, high precision and high accuracy, helping enterprises reduce production costs and increase efficiency in actual industrial scenarios.
Figure : Paddle Seg image segmentation development process
The large-scale application of AI in industrial scenarios is expected
Based on the iron and steel industry, Baowu Zhongnan Iron and Steel Co., Ltd. develops industry, science and trade simultaneously, and operates diversified operations. It is an important iron and steel production base in Guangdong Province and a promoter of smart manufacturing and 5G factories in the iron and steel industry. The products mainly include special rods, industrial wires, building materials (common materials), and plates, which are widely used in automobiles, machinery manufacturing, energy transportation, aerospace, nuclear power and other industries.
This time, Baowu Zhongnan Iron and Steel took the lead in introducing new AI technology, and applied flying paddles in the automation and intelligent upgrading process of the iron and steel industry. Paddle continues to explore more intelligent scenarios and solutions."
"The fault tolerance rate in industrial scenarios is very low. We can only put it online when we ensure that all aspects of the performance of AI platform technology applications meet our actual production standards. Flying Paddle fully meets our requirements, allowing us to rest assured that more Entrusting tasks to machines also makes us more confident in the large-scale application of artificial intelligence in the industrial field." The person in charge of the intelligent project said.
It is an inevitable trend for traditional industries to evolve to high-end and intelligent. For enterprise developers, the "tools" at hand can help them quickly and efficiently overcome the complexity of the industrial environment, and give birth to more "AI+industry" applications. Flying Paddle will continue to provide effective support for the intelligent upgrading of the steel industry, and work together to expand the road of technological innovation.
Related Reading