Machine Learning-based Predictive Maintenance for Optical Networks

Optical networks provide the backbone of modern telecommunications by connecting the world faster than ever before. However, such networks are susceptible to several failures (e.g., optical fiber cuts, malfunctioning optical devices), which might result in degradation in the network operation, massive data loss,  and network disruption. It is challenging to accurately and quickly detect and localize such failures due to the complexity of such networks, the time required to identify the fault and pinpoint it using conventional approaches, and the lack of proactive efficient fault management mechanisms. Therefore, it is highly beneficial to perform fault management in optical communication systems in order to reduce the mean time to repair, to meet service level agreements  more easily, and to enhance the network reliability. 

In this thesis,  the aforementioned challenges and needs are tackled by investigating the use of machine learning (ML) techniques for implementing efficient proactive fault detection, diagnosis, and localization schemes for optical communication systems. In particular, the adoption of ML methods for solving the following problems is explored:

- Degradation prediction of semiconductor lasers,

- Lifetime (mean time to failure) prediction of semiconductor lasers,

- Remaining useful life (the length of time a machine is likely to operate before it requires repair or replacement) prediction of semiconductor lasers,

- Optical fiber fault detection, localization, characterization, and identification for different optical network architectures, 

- Anomaly detection in optical fiber monitoring.

Such ML approaches outperform the conventionally employed methods for all the investigated use cases by achieving better prediction accuracy and earlier prediction or detection capability.


Use and reproduction:


Please note that individual components of the publication may be subject to other licensing or copyright conditions.


Citation style:
Could not load citation form.