From idea to method - Jianlong independent research and development of scrap intelligent inspection system application road

"In 2018, we began to do the feasibility study of the scrap automatic detection and classification project, and officially began to enter the research and development stage in the same year. It can be said that we are the first to carry out intelligent identification of scrap steel with machine vision and artificial intelligence in China, and to use this method for actual scrap steel quality inspection enterprises." Wang Baohong said that the original intention of developing this system is to reduce the labor intensity of workers and avoid the impact of human subjective factors on quality.
Over the years, the manual steel scrap inspection, generally before, during and after the vehicle unloading with the naked eye or caliper measurement, the scrap to do 3 categories, grade determination, often a level of the gap will lead to 30 yuan per ton, which brings some quality personnel no small gray income space, will also make the quality of scrap more likely to fluctuate, affecting the subsequent smelting production.
"When I first received this project three years ago, I wondered if it was possible to solve the pain point of scrap inspection through image recognition or other non-human means." Wang Baohong said that through analysis and screening, the company finally determined the idea of solving the problem through image recognition and artificial intelligence. The system takes multiple photos from multiple angles, using hundreds of thousands of photos to finally achieve extremely high decision accuracy. After comparing with manual quality test for more than 1 year, the judgment of the system is finally consistent with manual quality test.
However, it is not easy to build a system that is usable, convenient, and can save businesses costs. According to reports, the shooting quality of scrap steel photos and the distance and Angle of the camera and scrap steel are closely related, and the problem of near large far small needs a specific algorithm to solve; After exploring, the parking position of the car is very sensitive to the observation results of the camera, although it now seems that the conclusion is very simple, but to solve these problems, it needs to go through a lot of industrial knowledge accumulation and computer modeling process, especially the difficulty of obtaining real data. At the same time, AI requires a lot of data and time to self-learn. "These are the technical difficulties and the cost of making a system." Wang Baohong said.
Not only that, because the system can avoid the fluctuation and the possibility of cheating brought about by manual operation, in the early days of promotion, the system was resisted by many scrap suppliers. "However, the relevant person in charge of a steel company user is very supportive of us and let us resist." We have also done a lot of work, and after a while, suppliers have gradually realized that with this system, quality testing is fair and fair, and no longer have to be secretive."
In July 2019, Beijing Tongchuang Information Communication Technology Co., Ltd. obtained the "Scrap Intelligent Detection System V1.0" computer software copyright registration certificate of the National Copyright Administration.

"There is a large number of data and industry personnel to participate in support, is the advantage of our company scrap intelligent detection system V1.0." Wang Baohong introduced that so far, the system has accumulated a large number of enterprise data, the system built for Hegang Shi Steel has entered the acceptance stage, and the contract has been signed with three steel Group, Changzhou Dongfang Special Steel has signed the contract and paid the advance payment... "Dozens of enterprise data and hundreds of thousands of pictures can greatly ensure the level of systematic discrimination." It is understood that the classification accuracy of the system to judge the type and classification of scrap is 95%, especially the classification accuracy of heavy steel scrap can reach 2 mm.
"The project avoids the subjectivity of manual processing, will greatly reduce the labor intensity of on-site employees, improve the working environment of Shigang employees, and improve on-site work efficiency, I believe that this system will create huge economic benefits for us in terms of fine management." According to the relevant person in charge of Hegang Steel, the system passed the pressure test in Shigang from June 17 to 21 this year, the automatic photo rate is 99%, the intelligent identification accuracy rate of scrap steel grade is more than 90%, and the automatic generation of scrap steel grade judgment report.
IT is also understood that the main research and development team members of the system are about 30 people, mainly IT personnel and business personnel. "This system can't be done with just the IT people." Wang Baohong said with emotion that industrial informatization and intelligent transformation and the participation of industrial personnel are the basic conditions. In the R & D team of the system, there are many senior scrap inspectors and scrap standard makers.
According to the company's research shows that many competitive systems are currently in a semi-stagnant state, the reason is that only rely on capital investment and algorithm optimization, without industry knowledge, so built up the system, can not be well adapted to the actual production of enterprises, embedded in the enterprise workflow.

It is understood that in November 2020, the China Iron and Steel Association organized relevant experts, including Xu Jinwu, former president of the University of Science and Technology Beijing, to form an expert group to identify this method. The expert group highly praised this achievement: "The first scrap grade intelligent detection algorithm technology, the development of automatic unmanned scrap grade intelligent detection system and complete sets of equipment, completely replacing manual quality inspection." The overall technical achievements have reached the international advanced level, and the accuracy rate of scrap grade recognition has reached the international leading level."
At the same time, the identification certificate affirmed the three main innovation points of this achievement: First, the development of automatic unmanned scrap intelligent inspection system based on machine vision and deep learning, the first time to realize the unmanned scrap grade intelligent identification and automatic inspection of the whole process of scrap acceptance; Secondly, the camera Angle adaptive adjustment algorithm based on scan path optimization is developed to realize the efficient and accurate classification of scrap images. Third, based on the historical database, the multi-dimensional convolutional neural network model for intelligent identification of scrap grade and scrap classification detection system in collection and storage are developed to realize the automatic impurity retention function of scrap materials and improve the scientific determination of scrap grade. In May this year, the China Iron and Steel Industry Association, the Chinese Institute of Metal Metallurgy Science and Technology Award awarded 115 projects, of which, by Beijing Jianlong Heavy Industry Group, Shanxi Jianlong Industrial Company and Beijing Tongchuang Information Technology Company completed the "machine vision-based scrap grade intelligent identification technology and application" won the second prize.
"In principle, the establishment of scrap detection system is relatively simple, that is, through the scrap intelligent detection system automatically distinguish scrap grade." Therefore, different aspects of the enterprise smell this business opportunity, there are some imitations in the market. Due to the reasons mentioned earlier, capital has poured into this field without patent license and technology is not mature, which has a certain negative impact on scrap testing and steel company customer utility." Mr. Wang expressed his concern.
Ms Wang said many of the practices had confused users. For example, based on the usual experience of artificial intelligence, self-learning, etc., plus some hardware investment, the normal implementation cycle takes at least six months. And some companies charge very low or even free, saying that in a short time can make the system function normally, there is a suspicion of disrupting the market.
"The principle of scrap visual intelligent identification systems that have emerged in the past two years is the same, and in terms of technology, each family has made some corresponding adjustments." "However, we have applied for an invention patent for this method," Wang said. It is understood that in early October 2019, the company applied for relevant patents. From April to May 2021, the company's "A method for Establishing a neural network model for scrap grade classification" (patent No. 201910958076.8) and "a method for Scrap Grade Classification Detection in Storage" (patent No. 201910958076.9) successively obtained patent certificates. "It is safe to say that we are the only company that has obtained a patent certificate for this method in China." Wang Baohong said.
It is understood that at present, the company has collected evidence of more than 10 similar systems in China, generally by installing cameras in the material yard to collect pictures, and using image recognition technology and artificial intelligence, neural network methods to let the system learn, according to the definition of scrap categories, in order to achieve the purpose of automatic recognition.
"Given that we have applied for a patent for this method, it is strictly necessary to pay us a certain amount of intellectual property fees for using this method, otherwise it is an infringement." Wang Baohong said seriously, "To respect the original, but also in accordance with the market rules to do things." Now, there is indeed such a situation in the market, but also to remind the majority of steel consumers, do not be confused by these so-called phenomena. Yesterday, we commissioned the firm to send letters to two of them. Of course, we hope that if possible, we can work together to find opportunities to work together, to serve our customers well, and to contribute positively to the industry."
Talking about the company's independent research and development system, Wang Baohong confidently said: "At present, our system has started another round of technical reserves, after the completion of all, can be used for scrap products for more dimensions of discrimination, discrimination accuracy can be further improved, conservative estimates can be on the current basis for the enterprise to save 40% of the cost."