An Effective Deep Learning Model Designed for Detecting Fiber Faults in the OTDR Dataset

Fiber optic cables enable global connectivity by efficiently transporting large amounts of data across the internet, mobile, and core network platforms. Signal degrading can harm broad areas. Hence, it is essential to address any problems quickly and effectively. Optical fibers are prone to errors that can be detected using OTDR (Optical Time Domain Reflectometer) for anomaly detection. An OTDR utilizes the Rayleigh backscattering phenomenon to quantify and examine reflections of optical signals, hence identifying any issues within the fiber. These issues can be addressed through human effort or automation using a machine-learning model. This paper proposes a deep learning model to classify common fiber optic network issues using publicly accessible OTDR statistics. Our model outperforms previous studies on accuracy detection when combined with principal component analysis of PCA with a deep neuron network. Our validation results show that the accuracy is up to 99.96%.