Expertise-Enhanced Machine Learning for Failure Detection on Field-Deployed Optical Modules

The health state of optical modules is crucial for ensuring the stable and reliable operation of optical transport networks (OTNs). Recently, data-driven techniques have shown immense potential in failure detection within optical networks. However, in practical OTNs, failures in field-deployed optical modules are rare events, leading to a scarcity of fault samples while normal samples are relatively abundant, which significantly limits the accuracy of data-driven models. Moreover, although the domain of field-deployed optical module failures is rich in expertise, this knowledge is often highly abstract and specialized, making it difficult to convert into a form understandable by computers, and manual data labeling by experts is costly. To address these challenges, this study innovatively proposes an expertise-enhanced failure detection scheme. This scheme ingeniously embeds expertise into the data preprocessing stage of the data-driven modeling, effectively combining the advantages of expertise and data-driven methods, significantly enhancing the performance of failure detection. Specifically, we employ decision tree algorithms to successfully solidify expertise into computationally understandable decision tree models, enabling the effective extraction and intuitive representation of expertise, expanding the application scope of expertise in failure detection. The experiments are validated using field-deployed optical module dataset managed by network operators. The results demonstrate the effectiveness of the expertise-enhanced method in improving detection performance with limited fault samples. Additionally, by testing the impact of varying fault sample quantities on model detection results, we explore the potential limitations of the expertise-enhanced method in field-deployed optical module failure detection, providing valuable insights for its practical application.