Jozef Filipovský works as a chief process engineer at finishing. In addition to the support of colleagues, process engineers, it is especially dedicated to galvanizing lines No. 2 and No. 3. He has been with the ironworks since 1990 and has tried all the jobs on the line, except for the work of a crane operator, and therefore knows it perfectly. Until 2020, he worked as a head of operation for a galvanizing plant, and in 2020 he became a process engineer. In May this year, after a medium repair, a problem arose. The TP1 pyrometer began to show unrealistically high values, despite the fact that the furnace and line settings remained the same. Another shutdown was out of the question, but the operators needed to find a way to set the temperatures in the non-oxidation furnace at least approximately correctly. Therefore, they asked Jozef Filipovský for help.
What does a pyrometer measure and what can cause incorrect temperature readings?
After the May intermediate repair, the pyrometer showed a sheet temperature higher by up to 70-80 degrees than our many years of experience. If the pyrometer does not measure accurately, it may happen that you do not anneal properly and the sheet remains hard, brittle, or breaks. It can achieve the required parameters after repeated annealing, but the cost per ton produced is much higher because the energies are expensive. We could have shut down the line again for repairs for a few days, but this would have meant a production outage, or we could have looked for another solution .
Today we know that you have found another solution, literally in a matter of days...
At the beginning, my colleagues from the operation wanted me to help them find out from historical data what settings worked for the most common assortments. At that time, no one had any idea that this task would start the development of an advanced computing tool. It all started with the manual processing of historical data, which was then analyzed and logically linked to the process. Help also came from an unexpected place – from ChatGPT. Although it is a generative artificial intelligence designed primarily for text processing, it has proven that it can also be an excellent helper in the creation of technical computational models with the correct input and logical thinking. This resulted in three models. Model F, which is physical, based on line parameters such as belt dimensions, speed, furnace temperature, and so on. It is accurate under steady conditions, but with large variations in change. The second model S also includes natural gas consumption in the calculation. It is more accurate in stable production, but less reliable in transitions between ranges. And so the Adaptive Combined Model was created, which combines the outputs of the F and S models and achieves the highest accuracy thanks to mass flow interpolation. It works with the interval in which the expected temperature value is located. This makes the result more stable and resistant to fluctuations in operation. After the success of the TP1 calculations, the same methodology was applied to the TP3 pyrometer.
How does this model help the operation?
The model was visualized in Power BI reports and placed on a server to make it available to operations and operators. The report also includes a natural gas THN per tonne produced, providing a broader view of efficiency. The results of the model were so credible that the operation adjusted the output from the TP1 pyrometer through compensation – first by 40 °C, later by another 10 °C. Since then, no problems have been reported.
Will this project have a sequel?
After the successful deployment of the TP1 and TP3 AI computing models, we have an even more ambitious goal – to create an automatic control model for the cooling section of the reduction furnace on the galvanizing line number 2. This section is crucial for stabilizing the temperature of the steel strip before immersion in the zinc bath, where the quality of the galvanizing is decided. I presented this project at the last MetriX meeting.
What will be its main benefits?
Will you also "cooperate" with ChatGPT in the creation of this project?
ChatGPT is not leaving the scene, on the contrary, it will play a key role. Thanks to the ability to combine technical knowledge, physical context and logic of regulation, it is still an invaluable helper in the innovation process. This project has shown that advanced technologies and data analytics don't just have to be created in a lab or university – they can also be born in the field. For me, ChatGPT is also an invaluable helper because I am a metallurgist and self-taught in the field of IT.
Photo: Vladimír Adamčo
