EXPLICABLE AI FOR INTERPRETING PREDICTIONS OF PREDICTIVE MAINTENANCE NEURAL NETWORK MODELS

The article discusses the application of Explicable Artificial Intelligence (XAI) methods for interpreting forecasts of neural network models in predictive maintenance systems for industrial equipment. The aim of the study is to increase transparency, accuracy and trust in the results of machine learning in the diagnosis and prediction of technical malfunctions. The paper uses SHAP, LIME, and Grad-CAM methods to visualize the contribution of features to model predictions. Using the example of the enterprises of the Kostanay automobile cluster, it is shown that the integration of Explicable AI can increase diagnostic reliability by 12-15%, reduce unplanned equipment downtime by 13% and reduce the response time of engineers by 18%. The study demonstrates that interpreted machine learning models form the basis for transparent solutions within the framework of digitalization and the introduction of the Industrial Internet of Things (IIoT) in Kazakhstan. The proposed approach corresponds to the strategic directions of the Digital Kazakhstan and the Concept of Artificial Intelligence Development until 2029 programs, contributing to improving the efficiency and reliability of domestic maintenance systems.
The experimental results were obtained using anonymized data from an industrial automotive engineering enterprise in the Republic of Kazakhstan, which confirms the applicability of the proposed approach in real production conditions.

1. Moosavi S. Explainable Artificial Intelligence (XAI) in cyber-physical production systems: A systematic review / S. Moosavi, A. Kor, M. Fathi // Journal of Manufacturing Systems. – 2024. – № 72. – Р. 130-152.

2. Li X. Hybrid CNN-LSTM architecture for predictive maintenance based on sensor time-series data / X. Li, Y. Li // Mechanical Systems and Signal Processing. – 2025. – № 208. – Р. 110987.

3. Cummins S. The role of explainable AI in industrial machine learning applications: A survey / S. Cummins, S. Anwar, M. Darwish // Engineering Applications of Artificial Intelligence. – 2023. – № 125. – Р. 106620.

4. Nor N.F.M. Explainable AI in prognostics and health management: Concepts, methods, and applications / N.F.M. Nor, M.A. Hussain, H.A. Rahman // IEEE Access. – 2021. – № 9. – Р. 152830-152845.

5. Lundberg S. A unified approach to interpreting model predictions (SHAP) / S. Lundberg, S.-I. Lee // Advances in Neural Information Processing Systems. – 2017. – № 30.

6. Ribeiro M.T. Why Should I Trust You? Explaining the predictions of any classifier (LIME) / M.T. Ribeiro, S. Singh, C. Guestrin // Proceedings of the 22nd ACM SIGKDD. – 2016. – Р. 1135-1144.

7. Hochreiter S. J. Long Short-Term Memory / S. Hochreiter, Schmidhuber, // Neural Computation. – 1997. – № 9(8). – Р. 1735-1780.

8. Zhang W. Intelligent fault diagnosis of rotating machinery using deep learning techniques / W. Zhang, D. Yang, H. Wang // Mechanical Systems and Signal Processing. – 2022. – № 170. – Р. 108965.

9. Ministry of Digital Development of Kazakhstan. Digital Kazakhstan State Program (2018-2025). Министерство цифрового развития РК. (официальный документ).

10. Government of the Republic of Kazakhstan. Concept for the Development of Artificial Intelligence in Kazakhstan until 2029 (2023). (утверждённая госпрограмма).