Passive optical networks are a suitable and promising solution for the service delivery network, particularly for data, video, and voice communications. An optical distribution consists of passive elements (e.g., splitters, fibers), providing a cost-effective high bandwidth tree-like topology. This paper describes a new analysis system designed for an inspection of passive optical networks. The system uses an FPGA-accelerated network interface card to capture the communication from the optical network and machine learning models in order to evaluate the possible anomalies, errors, or attacks.