Recommender system for defects reduction in metallurgy

Recommender system for defects reduction


Now, when raw materials are becoming more expensive and society demands to reduce CO2 emissions, the issues of energy efficiency of production are becoming more and more urgent. Defective products are not only wasted time, but also one of the sources of excess energy consumption. By reducing waste - your business simultaneously reduces costs, CO2 emissions, and increases productivity.


At all metallurgical enterprises where I was, there was quality control of finished products. Defective products were sent for melting and the client received only high-quality products. But it turns out that the resources spent on the production of a defective product are wasted.

Can these unnecessary costs be reduced with a help of machine learning algorithms?


Our experience shows that yes.


AI analyzing of steel slabs


We applied machine learning methods to predict sheet metal rejects. Our algorithm analyzed the production parameters of the slab, from which the sheet was rolled. We identified patterns indicating quite accurately that a defective final product would be obtained from this slab.


The production was digitized rather well, which made it possible to get a large amount of data (several hundred parameters). This was a challenge for client's specialists to analyze, but it was good for our data scientists (the more data the more precise results). However, only 25 parameters turned out to be significant (significantly influencing the probability of defects), which could be later monitored by technologists.



Before


After



Due to the early identification of defective slabs, the company did not spend resources on transportation, heating, rolling the slab into a sheet, sorting the sheet, sending it back for remelting. That is, artificial intelligence not only increased productivity, but also reduced the energy consumption of the enterprise.



Result


As a result of using the algorithm, we were able to reduce waste by 34% with a false rejection of 8.7%. For example, if before the introduction of the machine learning algorithm, there were 180 slabs, from which defective sheets were made. The algorithm showed 70 slabs, of which 60 were defective. Thus, the processing of 50 slabs was saved, that is, by more than a quarter to reduce the unproductive consumption of resources.


In addition, technologists received valuable information to optimize the production process in order to reduce the overall number of defective products.





Thank you for leaving a request

We will contact you shortly

Contact us