Gemelo digital híbrido adaptativo para predecir la vida útil remanente de rodamientos

Byron Hernán Gualancañay Guachamin; Darwin Leonardo Quispe Ordoñez

doi:10.31243/id.v22.2026.3332

Autores/as

Byron Hernán Gualancañay Guachamin Universidad Politécnica de Valencia https://orcid.org/0009-0005-2795-8272
Darwin Leonardo Quispe Ordoñez Universidad Politécnica de Valencia https://orcid.org/0009-0008-4674-8352

DOI:

https://doi.org/10.31243/id.v22.2026.3332

Palabras clave:

vida útil remanente; rodamientos; gemelo digital; aprendizaje híbrido; cambio de dominio; incertidumbre

Resumen

La vida útil remanente (RUL) de rodamientos debe estimarse de manera causal y robusta ante cambios de régimen. Este estudio evalúa un gemelo digital híbrido que combina características vibratorias, un estado de salud actualizado mediante filtro de Kalman y un regresor Gradient Boosting. Se analizaron 24 889 adquisiciones de 17 rodamientos PRONOSTIA. La partición se realizó por unidad: 12 rodamientos de las condiciones 1 y 2 para entrenamiento, dos para validación y tres de la condición 3, no observada durante el ajuste, para prueba. Se compararon un prior físico, Random Forest, Gradient Boosting y el híbrido. En prueba, Gradient Boosting obtuvo el mejor resultado (MAE=0,1244; RMSE=0,1514; R²=0,7256), mientras que el híbrido alcanzó MAE=0,1399; RMSE=0,1683; R²=0,6608. El híbrido mejoró levemente la validación, pero aumentó el MAE 12,45 % bajo cambio de dominio. Sus intervalos al 95 % cubrieron el 81,28 % de las observaciones. Los resultados muestran que incorporar un estado físico no garantiza generalización: la calibración y la adaptación entre condiciones siguen siendo necesarias para un uso industrial confiable.

Referencias

Cao, Y., Ding, Y., Jia, M., & Tian, R. (2021). A novel temporal convolutional network with residual self-attention mechanism for remaining useful life prediction of rolling bearings. Reliability Engineering & System Safety, 215, 107813. https://doi.org/10.1016/j.ress.2021.107813

Gong, X., Zhang, X., Sun, Q., & Sun, L. (2022). Bearing remaining useful life prediction based on multi-scale temporal convolutional network. Measurement, 199, 111488.

Inturi, V., Sabareesh, S., & Supradeepan, V. (2024). Digital twin-based predictive maintenance for industrial machines: A systematic review. Sensors, 24(15), 5002. https://doi.org/10.3390/s24155002

Li, N., Xu, P., Lei, Y., Cai, X., & Kong, D. (2022). A self-data-driven method for remaining useful life prediction of wind turbines considering continuously varying speeds. Mechanical Systems and Signal Processing, 165, 108315. https://doi.org/10.1016/j.ymssp.2021.108315

Li, X., Zhang, W., Ma, H., Luo, Z., & Li, X. (2024). Data-driven remaining useful life prediction of rolling element bearings: A review. Mechanical Systems and Signal Processing, 209, 111120. https://doi.org/10.1016/j.ymssp.2024.111120

Liao, M., Wang, G., Yang, Y., & Li, H. (2023). Multisensor data fusion for remaining useful life prediction of rolling bearings under variable working conditions. Advanced Engineering Informatics, 58, 102195. https://doi.org/10.1016/j.aei.2023.102195

Liu, J., & Fan, Z. (2022). A hybrid deep learning approach with adaptive feature fusion for remaining useful life prediction. Reliability Engineering & System Safety, 218, 108182. https://doi.org/10.1016/j.ress.2021.108182

Lu, Y., Li, S., & Huang, H. (2025). Digital-twin-driven remaining useful life prediction with dynamic information interaction. IEEE Access, 13, 111614–111627. https://doi.org/10.1109/ACCESS.2025.3583313

Shi, Y., Liu, M., Jiang, S., & Chen, H. (2023). Digital twin-enabled prognostics and health management for complex systems: A review. Mechanical Systems and Signal Processing, 190, 110152.

Xiao, Y., Wang, J., Li, H., & Liu, Z. (2025). Domain generalization for remaining useful life prediction under unseen operating conditions. Reliability Engineering & System Safety, 253, 110534. https://doi.org/10.1016/j.ress.2024.110534

Xu, H., Li, X., Lei, Y., & Li, N. (2024). Probabilistic remaining useful life prediction with calibrated uncertainty. Reliability Engineering & System Safety, 250, 110257.

Yang, S., Li, X., Liu, J., & Lei, Y. (2024). Physics-informed deep learning for remaining useful life prediction under varying operating conditions. Reliability Engineering & System Safety, 242, 109716.

Yin, C., Li, Y., Wang, Y., & Dong, Y. (2025). Physics-guided domain adaptation for remaining useful life prediction under variable operating conditions. Mechanical Systems and Signal Processing, 224, 112192. https://doi.org/10.1016/j.ymssp.2024.112192

Zhang, B., Zheng, W., Li, M., & Han, J. Y. (2024). Physics-informed machine learning for bearing remaining useful life prediction. Quality and Reliability Engineering International, 40, 3018–3036. https://doi.org/10.1002/qre.3563

Zhang, Q., Chen, J., & Yan, R. (2024). Uncertainty quantification for remaining useful life prediction using deep ensembles. Mechanical Systems and Signal Processing, 208, 110953.

Zhao, Z., Chen, X., Ma, J., & Zhang, J. (2023). Digital twin-driven remaining useful life prediction for rotating machinery. IEEE Access, 11, 51280–51293. https://doi.org/10.1109/ACCESS.2023.3277587

Zhou, Y., Huang, Y., Pang, J., & Wang, K. (2023). Remaining useful life prediction for supercapacitor based on long short-term memory neural network. Mechanical Systems and Signal Processing, 182, 109610. https://doi.org/10.1016/j.ymssp.2022.109610

Zio, E. (2022). Prognostics and health management (PHM): Where are we and where do we (need to) go in theory and practice. Reliability Engineering & System Safety, 218, 108119.