International audienceThe dynamic and distributed nature of telecommunication networks makes complex the design of model-based approaches for network fault diagnosis. Most model-based approaches assume the prior existence of the model which is reduced to a static image of the network. Such models become rapidly obsolete when the network changes. We propose in this paper a 3-layered self-reconfigurable generic model of fault diagnosis in telecommunication networks. The Layer 1 of the model is an undirected graph which models the network topology. Network behavior, also called fault propagation, is modeled in Layer 2 using a set of directed acyclic graphs interconnected via the Layer 1. We handle uncertainties of fault propagation by quantifying strengths of dependencies between Layer 2 nodes with conditional probability distributions estimated from network generated data. Layer 3 is the junction tree representation of the loopy obtained Layer 2 Bayesian networks. The junction tree is the diagnosis computational layer since exact inference algorithms fail on loopy bayesian networks. This generic model embeds intelligent self-reconfiguration capabilities in order to track some changes in network topology and network behavior. These self-reconfiguration capabilities are highlighted through some example scenarios that we describe. We apply this 3-layered generic model to carry out active self-diagnosis of the GPON-FTTH access network.We present and analyze some experimental diagnosis results carried out by running a Python implementation of the generic model

Carrying out self-diagnosis of telecommunication networks requires an understanding of the phenomenon of fault propagation on these networks. This understanding makes it possible to acquire relevant knowledge in order to automatically solve the problem of reverse fault propagation. Two main types of methods can be used to understand fault propagation in order to guess or approximate as much as possible the root causes of observed alarms. Expert systems formulate laws or rules that best describe the phenomenon. Artificial intelligence methods consider that a phenomenon is understood if it can be reproduced by modeling. We propose in this paper, a generic probabilistic modeling method which facilitates fault propagation modeling on large-scale telecommunication networks. A Bayesian network (BN) model of fault propagation on GPON-FTTH (Gigabit-capable Passive Optical Network-Fiber To The Home) access network is designed according to the generic model. GPON-FTTH network skills are used to build structure and approximatively determine parameters of the BN model so-called expert BN model of the GPON-FTTH network. This BN model is confronted with reality by carrying out self-diagnosis of real malfunctions encountered on a commercial GPON-FTTH network. Obtained self-diagnosis results are very satisfying and we show how and why these results of the probabilistic model are more consistent with 1 the behaviour of the GPON-FTTH network, and more reasonable on a representative sample of diagnosis cases, than a rule-based expert system. With the main goal to improve diagnostic performances of the BN model, we study and apply EM (Expectation Maximization) algorithm in order to automatically fine-tune parameters of the BN model from real data generated by a commercial GPON-FTTH network. We show that the new BN model with optimized parameters reasonably improves self-diagnosis previously carried out by the expert Bayesian network model of the GPON-FTTH access network.

No abstract

Abstract-Network behavior modelling is a central issue for model-based approaches of self-diagnosis of telecommunication networks. There are two methods to build such models. The model can be built from expert knowledge acquired from network standards and/or the model can be learnt from data generated by network components by data mining algorithms. In a recent work, we proposed a model of architecture and fault propagation for the GPON-FTTH (Gigabit Passive Optical Network-Fiber To The Home) access network. This model is based on a Bayesian network which encodes expert knowledge. This includes dependencies that encode fault propagation and conditional probability distributions that encode the strength of those dependencies. In this paper we consider the problem of automatically tuning the above mentioned probability distributions. This is a parameter estimation problem under missing data conditions that we solve with the Expectation Maximization (EM) algorithm. Conditional probability distributions are learnt from the tremendous amount of alarms generated by an operating GPON-FTTH network during two months in 2015. Self-diagnosis is carried out to analyze the root cause of alarms. The performance of the diagnosis is evaluated with respect to an expert system based on deterministic decision rules currently used by the Internet Access Provider to diagnose network problems.