EVERSTEEL Co., Ltd.
EVERSTEEL completes on-site verification of steel scrap automatic analysis AI system with Asahi Kogyo
EVERSTEEL Co., Ltd. (Headquarters: Bunkyo-ku, Tokyo; President: Keijiro Tajima), which develops an AI system for iron scrap automatic analysis, and Asahi Industry Co., Ltd. (Headquarters: Toshima-ku, Tokyo; President: Noriyuki Nakamura), We have completed a five-month on-site verification of the iron scrap inspection AI development for the purpose of improving the efficiency and accuracy of the iron scrap inspection process. From November 2022, we have introduced cameras to all receiving ports for unloading at the yard, and we have started the development of AI for this operation, which is capable of making assessments at the same level as veteran acceptance inspectors in determining product types and grades and detecting non-conforming products. .
Background of on-site verification
In order to realize a decarbonized society by 2050, efforts to reduce emissions in the “steel sector,” which accounts for approximately 14% of Japan’s total CO2 emissions, are essential. There is a need to expand steel production using the electric furnace method, which can reduce CO2 emissions by 75% or more compared to the conventional method. Asahi Kogyo Co., Ltd. is an electric furnace manufacturer that manufactures rebars and structural steel using steel scrap as the main raw material while taking the environment into consideration. Based on the recognition that “we want to be a leader in creating a
resource-recycling society,” we are working on the effective use of limited resources with the aim of realizing a resource-recycling society. The key to popularizing electric furnace steel is the quality control of the raw material, iron scrap. For steel scrap, quality of raw materials is guaranteed through advanced inspection work by on-site acceptance inspectors at the time of receipt. Accurate inspection work for the large amount of scrap that is brought in every day, however, requires skilled skills, but on the other hand, there are few structural issues that business operators are forced to address, such as labor shortages and the aging of skilled workers. There is none. In order to break through such a situation, EVERSTEEL has started a field verification experiment of an AI system for automatic analysis of steel scraps with Asahi Industry in order to objectify personal skills with AI and pass on technology.
Details of on-site verification
1. On-site stay and quantification of acceptance inspector skills Tajima, the representative of EVERSTEEL, stayed directly at the factory for two weeks and conducted a local survey of its own acceptance inspection standards and non-conforming products, as well as an understanding of operations, in order to develop an acceptance inspection AI exclusively for Asahi Industries. In EVERSTEEL, the developer side acquires the same acceptance inspection skills as the on-site inspection staff before development, in order to realize the unique grade assessment that is not limited to simple thickness and size, which is performed by domestic skilled workers, with AI inspection. My motto is to do things. In particular, in the field survey, we quantified the variation in grade assessment for each inspector. At factories, inspection staff visually check steel scraps and grade them, requiring experience and a high level of skill. I’m here. Through this on-site survey, we analyzed the assessment results of all acceptance inspectors and the acceptance inspection work of unloaded scrap, and quantified the variation in assessments by acceptance inspectors.
2. Development of this AI system
For grading, this AI system will perform a grading assessment for the scrap loaded on one truck. In the assessment of heavy scrap, EVERSTEEL’s assessment accuracy is about 88% compared to the inspection accuracy of the inspection staff, and this AI system can already perform acceptance inspections with accuracy close to the inspection skills of the inspection staff during the on-site verification period. I have confirmed. We also analyzed the scrap data taken at night in the yard, and verified that even the late-night data can be categorized without problems.
In normal grading AI development, an enormous amount of work is required for acceptance inspectors to enter grading assessment data for each image in order to create training data. However, with EVERSTEEL, it is possible to use only the acceptance inspection results for each truck of the acceptance inspector, which is normally performed in daily operations, as training data. In this joint effort with Asahi Industries, we were able to confirm a high accuracy of approximately 88% with this AI. In addition to the analysis of trucks composed of a single grade in Fig. 2, Fig. 3 shows that AI analysis is possible for trucks of multiple grades with an accuracy close to that of an inspection inspector. I can confirm.
Figure 2. Example of comparison of results for single grade classification scrap (blue: AI, orange: inspector)
Figure 3. Example of comparison of mixed grade scrap (blue: AI, orange: inspector)
For foreign matter detection, this AI system detects “foreign matter” such as impurities that can be mixed into the steel material and closed objects that cause rapid reactions in the furnace. During the on-site verification period, we confirmed that this AI system can detect motors, which are representative of Trump elements, as shown in the example in Fig. 4.
Fig.4. Example of foreign object detection result (motor)
Prospects for full operation
In November, we will start AI development aiming at the same level as on-site acceptance inspectors, in preparation for the full-scale operation of this AI system. During this period, assuming that it will be introduced to actual operations, in addition to the heavy scrap that has been analyzed so far, we will expand the analysis to include all types of receiving scrap such as new cuts, turning powder, and shredders. We will develop a full-featured AI that can assess In addition, by introducing cameras to all unloading ports in the yard and accelerating data collection speed, we aim to improve development efficiency at once and complete development in a short period of time. [Image 5
Figure 5. Examples of types of iron scrap distributed in Japan (left), examples of foreign matter (right)
Comments from representatives of both companies
Asahi Industry Co., Ltd.: Mr. Suguru Sekine, Managing Director, General Manager of Administration Headquarters
“In Japan, where the declining birthrate and aging population are progressing and the working population is declining, the introduction of AI technology is essential for the steel industry, which is a heavy industry, to improve operational efficiency. We have started developing an AI acceptance inspection with EVERSTEEL.Also, until now, it has been difficult for electric furnace manufacturers to identify all impurities in iron scrap only at the final stage of steelmaking, making it difficult to identify the cause when a problem occurs. In order to manufacture high-quality steel, it is necessary to implement stable acceptance inspection by AI at the scrap stage, which is the raw material.By advancing the development of AI inspection with EVERSTEEL, it has been difficult until now. We will work to resolve the issue.”
EVERSTEEL Co., Ltd.: Keijiro Tajima, President and CEO
“In recent years, there have been strong calls to shift to a decarbonized society, but it is not easy to shift to new operations, especially in heavy industries such as the steel industry. The efforts with Asahi Kogyo, which is carrying out business reforms, are very valuable for our company.Asahi Kogyo, which introduced cameras to all four unloading ports and achieved results at once in full-scale development. Based on our intention, we will complete the development of an AI that can perform acceptance inspections at the same level as acceptance inspectors at an early stage.”
The interview of the site verification is published in note
The content of interviews with Asahi Kogyo Co., Ltd. is published in note. https://note.com/eversteel/n/n7b6c7010a9bd
About Asahi Industry Co., Ltd.
Company name: Asahi Industry Co., Ltd.
Head office location: Daiwa Higashi-Ikebukuro Building, 3-23-5 Higashi-Ikebukuro, Toshima-ku, Tokyo
Representative Director and President: Noriyuki Nakamura
Established: October 29, 1991
About EVERSTEEL Co., Ltd.
Company name: EVERSTEEL Co., Ltd.
Head office location: 7-3-1 Hongo, Bunkyo-ku, Tokyo University of Tokyo South Research Building Entrepreneur Lab 202
Representative Director and President: Keijiro Tajima
Established: March 16, 2021
[Inquiries regarding this matter]
EVERSTEEL Co., Ltd. (Person in charge: Public Relations Komiya) Email: firstname.lastname@example.org 7-3-1 Hongo, Bunkyo-ku, Tokyo University of Tokyo South Research Building Entrepreneur Lab 202 Details about this release: