A probabilistic approach to fault diagnosis in linear lightwave networks

R.H. Deng, A.A. Lazar, W. Wang
1993 IEEE Journal on Selected Areas in Communications  
Citation DENG, Robert H.; Lazar, A. A.; and WANG, W.. A probabilistic approach to fault diagnosis in linear lightwave networks. (1993). Abstract-The application of probabilistic reasoning to fault diagnosis in Linear Lightwave Networks (LLN's) is investigated. The LLN inference model is represented by a Bayesian network (or causal network). An inference algorithm is proposed that is capable of conducting fault diagnosis (inference) with incomplete evidence and on an interactive basis. Two
more » ... ve basis. Two belief updating algorithms are presented which are used by the inference algorithm for performing fault diagnosis. The first belief updating algorithm is a simplified version of the one proposed by Pearl for singly connected inference models. The second belief updating algorithm applies to multiply connected inference models and is more general than the first. We also introduce a t-fault diagnosis system and an adaptive diagnosis system to further reduce the computational complexity of the fault diagnosis process. I. INTRODUCTION HE complexity of communication networks and the vol-T ume of information provided by these networks have caused an increase in demand for network management systems and personnel. In particular, the area of network fault management requires a great deal of network expertise (design, operation, management, etc.) which has proved to be difficult to acquire and maintain. The application of expert, or knowledge-based, systems to attack the inherent complexity of network fault management (e.g., NDS [23] , YES/MVS [8], ACE [14] , Troubleshooter [12], and ISM [7]) is a growing effort. However, most of the network fault management systems were built on an ad-hoc and unstructured basis. The research on network fault management is still in its infancy. There is a pressing need, therefore, for establishing a theoretical foundation of network fault management and for bridging the gap between research and working systems. Fig. 1 presents a generic network fault diagnosis system architecture. The fault diagnosis system is data-driven and operates in real-time. It consists of four parts: the alarm acquisition system, the event manager, the interference engine, and the knowledge base. The alarm acquisition system gathers the network status information (alarm messages) from online monitors, and passes them to the event manager. The event manager filters the alarm messages according to certain criteria. The filtered messages, called evidence, are used as the input to the inference engine. The inference engine conducts fault diagnosis based on the available evidence and the Manuscript
doi:10.1109/49.257935 fatcat:mfad4s6jf5a7rl6uwfao7cmlja