OPTIMIZATION OF TECHNOLOGICAL PARAMETERS OF CASTING USING PHYSICO-INFORMED GENERATIVE MODELS TO MINIMIZE DEFECTS IN STEEL CASTINGS

Physicologically informed generative modeling (FIGM) is presented as an approach to system optimization of technological parameters of casting steel castings in order to minimize complex defects. The methodology combines machine learning methods (deep autoencoders for dimensionality reduction and clustering) with fundamental physical and technological limitations formalized in the form of analytical expressions. The latter describe the causal relationships between the input parameters of the process and the occurrence of specific types of defects, which ensures the generation of physically feasible and technologically sound solutions. The approach was tested on a historical dataset for the production of castings of the Frame type made of 20L grade steel, containing 15 input and 8 output parameters estimating the average defect level (SDA). As a result of the application of FIGM, two stable technological zones with different sets of parameters were identified and formalized, each of which provides a consistently acceptable level of quality. For quantitative assessment, an objective function has been developed that takes into account the weighted criticality of various types of defects. The key practical result is the creation of a visual analytics tool based on contour maps of the interrelationships of parameters and defects. It allows you to visually analyze technological compromises, identify areas of stable modes, and reasonably select optimal casting parameters. It is shown that the proposed methodology overcomes the limitations of traditional methods based on the sequential elimination of individual defects and ensures the transition to systematic quality management by simultaneously taking into account the mutual influence of key types of defects.

1. Automation of customer needs forecasting. World Applied Sciences Journal. 2013;24(24):126‒132. https://doi.org/10.5829/idosi.wasj.2013.24.itmies.80024; EDN SLAPBT

2. Odarchenko I.B., Zharanov V.A., Prusenko I.N. Extended assessment of the technological complexity of castings based on the application of neural network methods. In: Modern problems of me-chanical engineering: Proceedings of the XIII International Scientific and Technical Conference (scientific readings dedicated to the 125th anniversary of P. O. Sukhoi’s birth). Gomel: izd. Gomel State Technical University named after P. O. Sukhoi, 2020:102‒105. (In Russ.). EDN SMXSYJ.

3. Schmidhuber J. Deep Learning in neural net-works: An overview. Neural Networks. 2015;61:85‒117. https://doi.org/10.1016/j.neunet.2014.09.003; EDN YEWUEF.

4. Andreev S.M., Kolokol'tsev M.V., Savinov D.A. at al. The use of artificial neural networks to predict the strength properties of sand-clay mixtures of molds. Vestnik Magnitogorskogo gosudarstvennogo tekhnicheskogo universiteta im. G.I. Nosova. 2023;21(4):23‒33. (In Russ.).

5. Smykov A.F., Kuzovov S.S. Fundamentals of the development of an automated system for detecting casting defects in shaped castings based on neural networks and fuzzy logic. Zagotovitel'nye proizvodstva v mashi-nostroenii. 2025;23(7):295‒301. (In Russ.).

6. Schmidhuber J. Deep Learning in Neural Networks: An Overview. Neural Networks. 2015;61. EDN: YEWUEF.

7. Mikolov T., Karafiat M., Burget L., Cernocky J., Khudanpur S. Recurrent neural network based language model. In: 11th Annual Conference of the International Speech Communication Association. Japan. 2010: 38‒45.

8. Kuzovov S.S., Smykov A.F. The Application of Convolutional Neural Networks and Fuzzy Logic in Control and Automation Systems for Recognizing and Assessing the Danger Level of Hot Cracks in the Production of Cast Products. In: International Russian Smart Industry Conference (SmartIndustryCon). IEEE. 2024:984‒988.

9. Angelini L., Carlo F., Marangi C., Pellicoro M., Nardullia M., Stramaglia S. Clustering Data by Inhomogeneous Chaotic Map Lattices. Physical Review Letters. 2000;85:199‒329. EDN: LTNGNH.

10. Knyazev S.V., Kozyrev N.A., Usol'tsev A.A., Mikhno A.R. Algorithms for controlling the preparation of molding mixtures. Ferrous Metallurgy. Bulletin of Scientific, Technical and Economic Information. 2021;77(10):1076‒1080. https://doi.org/10.32339/0135-5910-2021-10-1076-1081. EDN: AMYKQZ.

11. Oborskiy G.A., Bundyuk A.N., Tarakhtiy O.S. Control system of cogeneration power plant at partial electrical loads. Journal of Automation and Information Sciences. 2018;50(7):70–78. https://doi.org/10.1615/jautomatinfscien.v50.i7.60; EDN: FOFKWW.

12. Odarchenko I.B., Zharanov V.A., Prusenko I.N. Optimization of hydrodynamics of gate-feeding systems using neural network methods for classifying technological parameters. Lit'e i metallurgiya. 2017;4(89):84‒88. (In Russ.).

13. Denisov M.S., Petreshin D.I., Chebotarev P.A., Ryauzov S.A. Development of a neural network model for predicting the mechanical properties of aluminum alloys during injection molding. Avtomatizatsiya. Sovremennye tekhnologii. 2023;77(10):439‒446. (In Russ.).

14. Yusupov L.R., Klochkova K.V., Simonova L.A. Methodology for modeling the chemical composition of cast iron with vermicular graphite based on a genetic algorithm. In: Innovative engineering technologies, equipment and materials. 2016:180‒185. (In Russ.).

15. Odarchenko I.B., Zharanov V.A., Prusenko I.N. Integration of neural network models into the processes of technological preparation of castings. Lit'e i metallurgiya. 2018;4(93):45‒49. (In Russ.).

16. Shaikhutdinova I.I. Automated quality control systems in the production of aluminum alloy castings. Liteinoe proizvodstvo. 2025;9:25‒33. (In Russ.).

17. Chernyaev M.A., Tereshina V.V. Hybrid models at the junction of physics and artificial intelligence: methods and applications. Vektor nauchnoi mysli. 2025;7(24):159‒164. (In Russ.).

18. Dli M.I., Vlasova E.A., Sokolov A.M., Morgunova E.V. Creation of a chemical-technological system digital twin using the Python language. Journal of Applied Informatics. 2021;16(1(91)):22‒31. https://doi.org/10.37791/2687-0649-2021-16-1-22-31; EDN AGPYPO.

19. Holmes J., Ahmed I., Brindescu C. et al. Using Relative Lines of Code to Guide Automated Test Generation for Python. ACM Transactions on Software Engineering and Methodology. 2020;29(4):1‒38. https://doi.org/10.1145/3408896; EDN: IZTCZY.

20. Kharisma Raharjana I., Faris Arifin M., Iklil Nur A., Mubarak N. Conversion of User Story Scenarios to Python-Based Selenium Source Code for Automated Testing. TEM Journal. 2023;12(1):309‒315. https://doi.org/10.18421/tem121-39; EDN: TDIVAF.

21. Knyazev S.V., Kutsenko A.I., Usol'tsev A.A. at al. Prospects and directions of digital transformation in foundry production. Izvestiya vuzov. Chernaya metallurgiya. 2023;66(2):140‒147. (In Russ.). https://doi.org/10.17073/0368-0797-2023-2-140-147. EDN: JNMFIQ.