Probabilistic deep learning models provide essential Uncertainty Quantification (UQ) for Predictive Maintenance (PdM) of industrial assets. However, their substantial computational demands require cloud deployment, which exposes the monitoring pipeline to industrial IoT network failures. Packet loss may result in timeouts and monitoring blind spots. To mitigate these risks, an edge-cloud offloading framework is introduced that physically separates the prognostic workload. The architecture comprises a lightweight edge node for deterministic, real-time baseline predictions and a high-capacity cloud node for probabilistic Remaining Useful Life (RUL) estimation. By continuously monitoring network transmission status, the system dynamically reverts to the edge node during communication interruptions. Preliminary results on the NASA C-MAPSS dataset (turbofan engines) indicate that the proposed offloading function constrains prediction errors even under 95% packet loss, thereby maintaining operational safety without compromising peak cloud accuracy when the network is stable.

Adaptive Continuum Computing for Resilient Predictive Maintenance / Del Prete, A., Mashaallah, Z., Amato, F.. - (2026), pp. 800-801. (22nd IEEE International Conference on eScience Naples, Italy 28 September 2026 - 02 October 2026) [10.1109/escience70237.2026.00127].

Adaptive Continuum Computing for Resilient Predictive Maintenance

Del Prete, Alessandro
Primo
;
Mashaallah, Zahida;Amato, Flora
2026

Abstract

Probabilistic deep learning models provide essential Uncertainty Quantification (UQ) for Predictive Maintenance (PdM) of industrial assets. However, their substantial computational demands require cloud deployment, which exposes the monitoring pipeline to industrial IoT network failures. Packet loss may result in timeouts and monitoring blind spots. To mitigate these risks, an edge-cloud offloading framework is introduced that physically separates the prognostic workload. The architecture comprises a lightweight edge node for deterministic, real-time baseline predictions and a high-capacity cloud node for probabilistic Remaining Useful Life (RUL) estimation. By continuously monitoring network transmission status, the system dynamically reverts to the edge node during communication interruptions. Preliminary results on the NASA C-MAPSS dataset (turbofan engines) indicate that the proposed offloading function constrains prediction errors even under 95% packet loss, thereby maintaining operational safety without compromising peak cloud accuracy when the network is stable.
2026
979-8-3195-3257-2
Adaptive Continuum Computing for Resilient Predictive Maintenance / Del Prete, A., Mashaallah, Z., Amato, F.. - (2026), pp. 800-801. (22nd IEEE International Conference on eScience Naples, Italy 28 September 2026 - 02 October 2026) [10.1109/escience70237.2026.00127].
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11588/1076301
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact