Intelligent Adapter for Simplified Central Monitoring of Network Alarms

Wide spread network of telecom operators includes multiple device types from multiple vendors. For central network fault monitoring, these devices feed alarms to Network Operating Center (NOC) in multiple formats over SNMP (Simple Network Management Protocol). Standard representation of network alarms is the key for efficient monitoring at NOC by the engineers. Adapters are developed to translate the varied network alarm types and objects into meaningful standard format. The adapter development requires a skilled domain expert, who analyze and configures thousands of alarm types and objects. This is a time consuming, complex and skill dependent task that takes 6-12 weeks for adapter readiness and its integration.Our machine learning based Unified Alarm Mapper (UAM) model (Patent Filed) offers automated approach of alarm analysis, object classification and translation to standardized alarms formats. The model is trained on learnings from realworld data and experience of network integration implementations for few of world's leading telecom operators. The model trained on keywords and definitions of 120,000 different alarm object from 60 different vendors. It classifies alarm objects of new alarm types and auto configures them in standard format. Synonym Vector approach used in the model gives unsupervised learning capability for classification of any future unseen alarm types and their attributes. The self-learning intelligence in the adapter makes it vendor agnostic and avoids need for development of separate adapters and enables near real-time readiness for monitoring at NOC. Overall accuracy of 95% was observed when compared the results with real-world mapped data by domain experts.