Научно-учебная лаборатория методов анализа больших данных LAMBDA

Публикации

• Книга

  Buzaev F., Mullakhmetov R., Bogachev R. et al.

  Proceedings of the 4th Workshop on NLP for Music and Audio (NLP4MusA 2026)

  Association for Computational Linguistics, 2026.

• Статья

  Butorova A., Bobakov V., Sergeev A. et al.

  Artificial intelligence and digital twins for failure prediction in data center cooling systems: a comprehensive literature review (2018–2026)

  European Physical Journal: Special Topics. 2026. P. 1-19.

• Глава в книге

  Khizhik A., Popov A., Ali S. et al.

  Parameter Search for MCSA-Guided Synthetic Fault Injection in Induction Motor Diagnostics

  Signature-Guided Data Augmentation for Induction Motor Diagnostics (SGDA) synthesizes physically plausible faulty spectra by injecting Motor Current Signature Analysis (MCSA) harmonics into healthy current recordings in the frequency domain. A key limitation in industrial deployments is that the motor parameters required to compute MCSA targets (e.g., supply frequency and slip or rotor frequency quantities) are often unavailable, unreliable, or drift over time. Parameter mismatch shifts the expected fault lines and can cause SGDA to inject peaks at incorrect frequencies, reducing detector robustness. We propose SGDA Parameter Search (SGDA-PS), a practical extension that treats unknown parameters as a finite-grid modelselection problem. SGDA-PS trains a family of SGDA-based binary classifiers over a discrete grid of candidate parameter settings and selects the configuration that maximizes an unlabeled file-level scoring criterion on observed suspected-fault recordings, without requiring labels for those recordings. The resulting score surface is visualized as a heatmap, offering an interpretable view of parameter sensitivity, identifiability, and ambiguity. Experiments in controlled settings show that SGDA-PS recovers parameter regions consistent with known configurations and yields structured search landscapes that remain informative even when the optimum is not unique. The approach requires no simulations, preserves SGDA’s frequency-domain interpretability, and provides a practical offline commissioning procedure when motor specifications cannot be trusted.

  In bk.: Conference Proceedings: 2026 IEEE Ural-Siberian Conference on Biomedical Engineering, Radioelectronics and Information Technology (USBEREIT), 14-15 May 2026. IEEE, 2026. P. 1-4.

• Препринт

  Shipilov F., Barnyakov A., Ivanov A. et al.

  ML-based Fast Simulation of FARICH Responses

  arxiv.org. Physics. Cornell University, 2026