ENVIRONMENT AND TECHNOLOGIES

Development of adaptive models for industrial product quality management using machine learning algorithms and production data

Authors

  • Aleksandr V. Zakazchikov Sergo Ordzhonikidze Russian State University for Geological Prospecting, 117997, Moscow, GSP-7, Miklukho-Maklaya street, 23

How to cite

GOST Zakazchikov A. V. Development of adaptive models for industrial product quality management using machine learning algorithms and production data // Environmental Management Issues. 2026. Vol. 5. No. 4. P. 34-44. DOI: 10.25726/y5105-0118-8555-x
APA Zakazchikov, A. V. (2026). Development of adaptive models for industrial product quality management using machine learning algorithms and production data. Environmental Management Issues, 5(4), 34-44. https://doi.org/10.25726/y5105-0118-8555-x

Abstract

A defect prediction model on a production line is applied after its training period, when process conditions have already changed, so inspection decisions depend on its adaptation. The study develops and compares adaptive quality management models in which the model update rule and the inspection threshold rule are varied independently. The SECOM open dataset covers 1,567 semiconductor production units with 590 process signals, a test label and a timestamp (July – October 2008). On a chronological stream of 1,067 units, eight prequentially evaluated strategies combined four update rules for logistic regression (static, expanding window, sliding window, retraining on an input drift signal) with two threshold rules (cross-validated on the training window, or a quantile of earlier stream scores). Fifty permutations of the unit order were the stationary control. On the chronological stream the static model reached an AUC of 0.559, below the 0.601-0.717 range of shuffled streams. Retraining brought no confirmed gain: the expanding window added 0.039 AUC (interval −0.118 to 0.144). Against a 10% target, the training-window threshold sent 3,7-17,3% of units to inspection and the stream-score threshold 7,7-11,7%. The expanding window with the stream-score threshold detected 11 of 45 defective units at a 7,7% inspection rate (balanced error rate 41,3%) versus 4 of 45 for the static model. On this stream with rare defects, the decisive adaptive component is recalculating the threshold from current scores, while retraining should use accumulated data.

Keywords

product quality management machine learning adaptive model data drift defect prediction sampling inspection production data

Funding

The authors did not declare any external funding for this research.

References

Vasin S.A., Malikov A.A., Nikolsky S.M. Justification of the volume of inspection operations based on statistical methods of product quality management // Bulletin of Nosov Magnitogorsk State Technical University. 2025. vol. 23. no. 1. pp. 103-109.

Grusho A.A., Grusho N.A., Zabezhaylo M.I., Smirnov D.V., Timonina E.E., Shorgin S.Ya. Detection of distribution drift // Systems and Means of Informatics. 2022. vol. 32. no. 4. pp. 14-20.

Zhukov A.V., Sidorov D.N. Modification of the random forest algorithm for classification of non-stationary streaming data // Bulletin of South Ural State University. Series: Mathematical Modelling and Programming. 2016. vol. 9. no. 4. pp. 86-95.

Konev K.A. Machine learning for decision support in the field of quality at an industrial enterprise // Economics. Informatics. 2023. vol. 50. no. 3. pp. 689-703.

Kuvayskova Yu.E. Application of ensemble machine learning methods for predicting the technical condition of an object // Proceedings of the Samara Scientific Center of the Russian Academy of Sciences. 2021. vol. 23. no. 1. pp. 111-114.

Lebedev I.S. Adaptive application of machine learning models to individual sample segments in regression and classification tasks // Information and Control Systems. 2022. no. 3. pp. 20-30.

Leontyev A.S., Rybenko I.A. Experience in implementing machine learning for calculating the quality and production of sinter // Izvestiya of Higher Educational Institutions. Ferrous Metallurgy. 2024. vol. 67. no. 1. pp. 76-82.

Timofeev G.A., Barbashov N.N., Terentyeva A.D. Statistical methods for managing technological processes // Izvestiya of Higher Educational Institutions. Mechanical Engineering. 2016. no. 12(681). pp. 58-65.

Timofeeva A.Yu., Avdeenko T.V. Feature selection under class imbalance: an empirical comparison of correlation and mutual information // Informatics and Automation. 2026. vol. 25. no. 5. pp. 1412-1443.

Shper V.L., Sheremeteva S.A., Smelov V.Yu., Khunuzidi E.I. Shewhart control charts: a simple but not easy-to-use data analysis tool // Izvestiya of Higher Educational Institutions. Ferrous Metallurgy. 2024. vol. 67. no. 1. pp. 121-131.

Bilen A., Grommes A., Stamer F., Lanza G. Coping with concept drift in a virtual metrology application to predict part quality in micro gear manufacturing // Measurement: Sensors. 2025. Vol. 38. Art. 101797.

Park H.J., Koo Y.S., Yang H.Y., Han Y.S., Nam C.S. Study on data preprocessing for machine learning based on semiconductor manufacturing processes // Sensors. 2024. Vol. 24. no. 17. Art. 5461.

Salem M., Taheri S., Yuan J.S. An experimental evaluation of fault diagnosis from imbalanced and incomplete data for smart semiconductor manufacturing // Big Data and Cognitive Computing. 2018. Vol. 2. no. 4. Art. 30.

Tercan H., Meisen T. Machine learning and deep learning based predictive quality in manufacturing: a systematic review // Journal of Intelligent Manufacturing. 2022. Vol. 33. no. 7. pp. 1879-1905.

Varfolomeev I.A., Ershov E.V., Vinogradova L.N. Statistical control of defects in a continuously cast billet based on machine learning and data analysis methods // Automation and Remote Control. 2018. Vol. 79. no. 8. pp. 1450-1457.

Issue

Section

ENVIRONMENT AND TECHNOLOGIES

Metrics

2 views
0 downloads
Want to publish with us?
Submit an article

Similar Articles

1 2 3 4 5 6 7 8 9 10 > >> 

You may also start an advanced similarity search for this article.