Otimização de classificadores de redes neurais por neuroevolução aplicada ao diagnóstico de defeitos em máquinas rotativas

Hygor Santiago; Milton Dias Júnior

doi:10.20397/2177-6652/2026.v26i3.3474

Authors

Hygor Santiago Universidade Estadual de Campinas http://orcid.org/0000-0002-4835-5498
Milton Dias Júnior UNICAMP https://orcid.org/0009-0006-2362-4942

DOI:

https://doi.org/10.20397/2177-6652/2026.v26i3.3474

Keywords:

Neuroevolução; Algoritmos Genéticos; Diagnóstico de falhas; Máquinas rotativas; Redes neurais.

Abstract

Objective: To develop and analyze a neuroevolution-based methodology for automatically optimizing the architecture of an Artificial Neural Network classifier applied to fault diagnosis in rotating machines using vibration signals.

Methodology: A Genetic Algorithm was employed to explore different neural architectures and their corresponding hyperparameters. Each individual represents a complete classifier configuration, including a sequence of convolutional, recurrent and dense blocks, as well as training parameters. The evolutionary process uses tournament selection, elitism, structural crossover, architectural mutation and parametric mutation. The fitness function considers predictive performance, model complexity and computational cost. Network training is performed with validation and early stopping.

Originality: The study proposes the joint optimization of the architecture and hyperparameters of a classifier designed for vibration-signal analysis. The approach reduces dependence on manual design decisions and enables the investigation of variable-depth architectures combining convolutional blocks, LSTM recurrent blocks and dense layers.

Main results: Neuroevolution identified hybrid architectures capable of combining local signal-pattern extraction with temporal-dependency modeling. Among the evaluated configurations, the CNN-LSTM architecture achieved the best overall performance. The results also indicated that data representativeness has a decisive influence on classifier generalization, meaning that architectural optimization alone is insufficient.

Theoretical/methodological contributions: The study presents a genetic representation for neural architectures, a multi-criteria fitness function and evolutionary operators for the automatic optimization of classifiers applied to fault diagnosis in rotating machines.

Author Biographies

Hygor Santiago, Universidade Estadual de Campinas

Engenheiro mecânico formado pela UFSJ onde participou dos grupos Cyros, Milhas Gerais e Gep_LASID onde exerceu diversas atividades em pesquisa e liderança. Mestre em Engenharia Mecânica na Unicamp onde trabalha com Inteligência Artificial em sistemas de classificação baseados em séries temporais, aplicado a eletrocardiologia. Cientista de dados e Business Intelligence na Thinkseg e Bidu.

Milton Dias Júnior, UNICAMP

Possui graduação em Engenharia Mecânica pela Universidade Estadual de Campinas (1984), mestrado em Engenharia Mecânica pela Universidade Estadual de Campinas (1987) e doutorado em Engenharia Mecânica pela Universidade Estadual de Campinas (1994). Realizou pós-doutorado no Structural Dynamics Research Laboratory, da Universidade de Cincinnati, EUA. Atualmente é Professor Associado I da Universidade Estadual de Campinas. Tem experiência na área de Engenharia Mecânica, com ênfase em Dinâmica dos Corpos Rígidos, Elásticos e Plásticos, atuando principalmente nos seguintes temas: Vibrações, Análise Modal, Processamento de Sinais, Dinâmica de Máquinas Rotativas e de Trens de Potência.

References

Al-Hussain, K. (2003). Dynamic stability of two rigid rotors connected by a flexible coupling with angular misalignment. Journal of Sound and Vibration, 266(2), 217–234. https://doi.org/10.1016/S0022-460X(02)01627-9 DOI: https://doi.org/10.1016/S0022-460X(02)01627-9

Al-Hussain, K., & Redmond, I. (2002). Dynamic response of two rotors connected by rigid mechanical coupling with parallel misalignment. Journal of Sound and Vibration, 249(3), 483–498. https://doi.org/10.1006/jsvi.2001.3866 DOI: https://doi.org/10.1006/jsvi.2001.3866

Bengio, Y., Simard, P., & Frasconi, P. (1994). Learning long-term dependencies with gradient descent is difficult. IEEE Transactions on Neural Networks, 5(2), 157–166. https://doi.org/10.1109/72.279181 DOI: https://doi.org/10.1109/72.279181

Deb, K. (2001). Multi-objective optimization using evolutionary algorithms. Wiley.

Dong, K., & Lotfipoor, A. (2023). Intelligent bearing fault diagnosis based on feature fusion of one-dimensional dilated CNN and multi-domain signal processing. Sensors, 23(12), Article 5607. https://doi.org/10.3390/s23125607 DOI: https://doi.org/10.3390/s23125607

Eiben, A. E., & Smith, J. E. (2003). Introduction to evolutionary computing. Springer. https://doi.org/10.1007/978-3-662-05094-1 DOI: https://doi.org/10.1007/978-3-662-05094-1_1

Goldberg, D. E. (1989). Genetic algorithms in search, optimization, and machine learning. Addison-Wesley.

Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep learning. MIT Press.

Hochreiter, S., & Schmidhuber, J. (1997). Long short-term memory. Neural Computation, 9(8), 1735–1780. https://doi.org/10.1162/neco.1997.9.8.1735 DOI: https://doi.org/10.1162/neco.1997.9.8.1735

Holland, J. H. (1975). Adaptation in natural and artificial systems. University of Michigan Press.

Inman, D. J. (2013). Engineering vibration (4th ed.). Prentice Hall.

Jia, F., Lei, Y., Lin, J., Zhou, X., & Lu, N. (2016). Deep neural networks: A promising tool for fault characteristic mining and intelligent diagnosis of rotating machinery with massive data. Mechanical Systems and Signal Processing, 72–73, 303–315. https://doi.org/10.1016/j.ymssp.2015.10.025 DOI: https://doi.org/10.1016/j.ymssp.2015.10.025

McFadden, P., & Smith, J. (1984). Model for the vibration produced by a single point defect in a rolling element bearing. Journal of Sound and Vibration, 96(1), 69–82. https://doi.org/10.1016/0022-460X(84)90595-9 DOI: https://doi.org/10.1016/0022-460X(84)90595-9

Michalewicz, Z. (1994). Genetic algorithms + data structures = Evolution programs (2nd ed.). Springer-Verlag. DOI: https://doi.org/10.1007/978-3-662-07418-3

Mitchell, M. (1996). An introduction to genetic algorithms. MIT Press. https://doi.org/10.7551/mitpress/3927.001.0001 DOI: https://doi.org/10.7551/mitpress/3927.001.0001

Muszyńska, A. (2005). Rotordynamics. CRC Press. DOI: https://doi.org/10.1201/9781420027792

Rao, J. S. (1983). Rotor dynamics. John Wiley & Sons.

Randall, R. B. (2011). Vibration-based condition monitoring: Industrial, aerospace and automotive applications. John Wiley & Sons. DOI: https://doi.org/10.1002/9780470977668

Randall, R. B., & Antoni, J. (2011). Rolling element bearing diagnostics—A tutorial. Mechanical Systems and Signal Processing, 25(2), 485–520. https://doi.org/10.1016/j.ymssp.2010.07.017 DOI: https://doi.org/10.1016/j.ymssp.2010.07.017

Scheffer, C., & Girdhar, P. (2004). Practical machinery vibration analysis and predictive maintenance. Newnes. DOI: https://doi.org/10.1016/B978-075066275-8/50001-1

Srivastava, N., Hinton, G. E., Krizhevsky, A., Sutskever, I., & Salakhutdinov, R. (2014). Dropout: A simple way to prevent neural networks from overfitting. Journal of Machine Learning Research, 15(1), 1929–1958

Zhao, R., Yan, R., Chen, Z., Mao, K., Wang, P., & Gao, R. X. (2019). Deep learning and its applications to machine health monitoring. Mechanical Systems and Signal Processing, 115, 213–237. https://doi.org/10.1016/j.ymssp.2018.05.050 DOI: https://doi.org/10.1016/j.ymssp.2018.05.050