Fiber optics has become the core of todays telecommunications and data networking infrastructures. Thanks to its advantageous properties like low attenuation, huge bandwidth and immunity against electromagnetic interference fiber cables are replacing existing copper cables. They have been widely deployed to realize high-speed links that may carry either a single wavelength channel or multiple wavelength channels by means of wavelength division multiplexing (WDM). The rapid evolution of technology, coupled with the insatiable demand for bandwidth resulted with the existence of two generations of fiber optic networks. Existing recovery mechanisms make optical networks fault-tolerant. Fault tolerance refers to the ability of the network to reconfigure and reestablish communication upon a failure. Usual recovery mechanisms that deal with network failures can be divided into protection and restoration. The protection mechanism activates in advance backup resource that will be used in case of failure, while the restoration mechanism takes over backup resource upon a failure; that is why protection mechanisms can recover quickly but are more demanding in terms of resource. Restoration mechanisms are less demanding when it comes to resource and therefore may be less costly than protection mechanisms in term of initial investments, but they generate longer service disruption. The goal of the work conducted during this thesis was to find a solution for technical implementation of the proactive fault detection and fast recovery in the core optical network. This dissertation outlines the different solutions and their evaluation from the point of view of the network telecommunications operator. First chapter of this thesis is devoted to reviewing the existing recovery methods in the optical networks. In order to understand this thesis it is necessary to establish the vocabulary and basic elements of telecommunications and in particular fiber-optic communications. Chapter II addresses the existing methods of prediction. This part covers the area of machine learning. Once the reader has been introduced to the typical methods for prediction, an enhanced system design is proposed. Chapter III consolidates all the previous review work to an estimation of the most promising technique for proactive fault detection. An experimental test bed using fiber as a sensor was set up and used to quantify exactly the performance of the selected solution. This requires building an advanced representation of data that was used as input data for machine learning software. The relative importance of some effects is discussed. Once a deeper understanding of the potential of this technique has been achieved with an experimental test bed, theoretical investigations through the development of analytical models are conducted to validate and generalize them. Chapter IV describes the advantages and drawbacks of this kind of detection. Simulations were made that prove the knowledge obtained using the test-bed. Later on, a case study concerning the phase of recovery in the transport network is presented.
