Within industrial systems, dependability is critical to keep the system reliable and available. Faults leading to failures of the whole system must be mitigated by a laborious design, testing, and verification process. This thesis aims to build a Network Management Agent (NMA), capable of fault detection, localization, and recovery in an Ethernet environment. The chosen NMA environment was a distributed industrial system using a redundant controller pair, where the controllers must determine when the roles should switch in case of primary failure. In the industrial context, this role-switching must be bounded and robust enough to withstand mixed traffic classes competing for network resources. An NMA was built to monitor the network with the help of Media Redundancy Protocol (MRP), heartbeats, and industrial switch queue status queries. The NMA could distinguish between node and link failure by localizing the fault while adjusting the network's Quality of Service (QoS). The controllers could safely switch roles after an average difference of 29.7065 ms from the moment the primary failed, and the secondary took over. Link failure was detected in three possible locations within 31.297 ms and the location was found within 373.419 ms. To the authors' best knowledge, other solutions mainly target L3 networks or require specialized supporting technology, whereas MRP was found in the majority of the investigated industrial switches. The proposed solution requires any heartbeat-like function sent from the switch, which MRP offers, and can be generalized to any environment where distinguishing a link from a node failure is important. QoS anomaly detection however requires capable switches and configuration of rules to prioritize the traffic accordingly.