TinyML for Fault Detection in Rotating Machinery: A Post-Training Quantization Assessment on the MAFAULDA Dataset

• Cândido Alfredo Carvalho de Lucena Filho Universidade de São Paulo (USP)
• Felipe Valencia de Almeida Universidade de São Paulo (USP) https://orcid.org/0000-0003-2031-6443

DOI: https://doi.org/10.5753/bresci.2026.249387

Resumo

Unplanned downtime in rotating machinery is a major cost driver in industrial operations, and early fault detection through vibration analysis can prevent it. Running these classifiers directly on microcontrollers (TinyML) reduces inference latency and enables faster on-device response, but it is still unclear how different model architectures and quantization schemes behave under strict memory and flash constraints. This paper evaluates the feasibility of embedding rotating-machinery fault machine learning classifiers on low-cost microcontrollers. We compare MLP, 1D-CNN, LSTM and Random Forest on the MAFAULDA dataset, using 132 features per window and stratified group k-fold cross-validation to prevent leakage across windows from the same file. Three post-training quantization schemes (FP32, FP16, INT8) are evaluated against the requirements of the ESP32-S3, STM32F4, and ESP32-WROOM. The INT8 MLP was the most efficient model conforming to all three platforms (37 KB, F1=0.98), reducing size by 67% over its FP32 counterpart with negligible accuracy loss. The 1D-CNN lost accuracy under INT8 (F1 from 0.94 to 0.75), and the unrolled LSTM grew 3× rather than shrinking, results that highlight the need to jointly evaluate the model and quantization, not in isolation.

Palavras-chave: Vibration Analysis, Embedded Machine Learning, Microcontrollers

Referências

Breiman, L. (2001). Random forests. Machine Learning, 45(1):5–32.

David, R., Duke, J., Jain, A., Janapa Reddi, V., Jeffries, N., Li, J., Kreeger, N., Nappier, I., Natraj, M., Wang, T., Warden, P., and Rhodes, R. (2021). TensorFlow Lite Micro: Embedded machine learning for TinyML systems. In Proceedings of Machine Learning and Systems (MLSys), volume 3, pages 800–811.

Gholami, A., Kim, S., Dong, Z., Yao, Z., Mahoney, M. W., and Keutzer, K. (2021). A survey of quantization methods for efficient neural network inference. arXiv preprint arXiv:2103.13630.

Jacob, B., Kligys, S., Chen, B., Zhu, M., Tang, M., Howard, A., Adam, H., and Kalenichenko, D. (2018). Quantization and training of neural networks for efficient integer-arithmetic-only inference. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 2704–2713.

Kapoor, S. and Narayanan, A. (2023). Leakage and the reproducibility crisis in machine-learning-based science. Patterns, 4(9):100804.

Khan, A., Hwang, H., and Kim, H. S. (2021). Synthetic data augmentation and deep learning for the fault diagnosis of rotating machines. Mathematics, 9(18):2336.

Lei, Y., Yang, B., Jiang, X., Jia, F., Li, N., and Nandi, A. K. (2020). Applications of machine learning to machine fault diagnosis: A review and roadmap. Mechanical Systems and Signal Processing, 138:106587.

Marins, M. A., Ribeiro, F. M. L., Netto, S. L., and da Silva, E. A. B. (2018). Improved similarity-based modeling for the classification of rotating-machine failures. Journal of the Franklin Institute, 355(4):1913–1930.

Martinez-Rau, L. S., Zhang, Y., Oelmann, B., and Bader, S. (2024). TinyML anomaly detection for industrial machines with periodic duty cycles. In Proceedings of the IEEE Sensors Applications Symposium (SAS).

Randall, R. B. and Antoni, J. (2011). Rolling element bearing diagnostics—a tutorial. Mechanical Systems and Signal Processing, 25(2):485–520.

Souza, R. M., Nascimento, E. G. S., Miranda, U. A., Silva, W. J. D., and Lepikson, H. A. (2021). Deep learning for diagnosis and classification of faults in industrial rotating machinery. Computers & Industrial Engineering, 153:107060.

Thota, Y. R., Afshar, M., Boden, S., Dunlap, B., Akin, B., and Nikoubin, T. (2025). TinyML enabled real-time bearing fault classification in motors using vibration signals. In Proceedings of the Great Lakes Symposium on VLSI (GLSVLSI).

Warden, P. and Situnayake, D. (2019). TinyML: Machine Learning with TensorFlow Lite on Arduino and Ultra-Low-Power Microcontrollers. O’Reilly Media.