FIBER FAULTS DETECTION IN OPTICAL FIBER MONITORING USING DEEP LEARNING MODEL

Abstract

Backhaul and access communication networks are heavily reliant on effective optical network management to prevent service interruptions and ensure Quality of Service (QoS) compliance. New types of failures in optical networks present challenges that could expose the network to various risks. Traditional detection techniques are increasingly inadequate in addressing these issues. In contrast, Deep Learning (DL) has emerged as a promising approach for fault identification and prevention. This study introduces several key contributions that distinguish it from conventional fault management systems. The first contribution is the development of a Long Short-Term Memory (LSTM) model, integrated with the Mutual Information (MI) technique, to assess their combined effectiveness in detecting normal optical fibers and seven distinct fault types, including fiber cutting, fiber eavesdropping (fiber tapping), dirty connectors, bad splices, bending, reflectors, and PC connectors, achieving an accuracy rate of up to 96%. Finally, the proposed model is benchmarked against other deep learning models, such as BiLSTM, CNN, DNN, and RNN, to evaluate critical performance metrics of the AI model.