logo_header
  • Topics
  • Research & Analysis
  • Features & Opinion
  • Webinars & Podcasts
  • Videos
  • Event videos

Catalyst taps AI-driven service assurance to solve problems before they happen

In a short sprint of a few months, a TM Forum Catalyst team has developed a “commercially ready” AI and machine learning-driven service assurance.

Sarah Wray
08 Jul 2019
Catalyst taps AI-driven service assurance to solve problems before they happen

Catalyst taps AI-driven service assurance to solve problems before they happen

In a short sprint of a few months, a TM Forum Catalyst team has developed a “commercially ready” framework for artificial intelligence (AI) and machine learning (ML)-driven service assurance. With cost pressures and growing expectations from customers, service providers need to continuously improve the quality of service (QoS) on their networks. The situation is becoming even more complicated as networks operate with a mixture of legacy and virtualized (SDN/NFV/5G) infrastructure, which cannot be effectively managed manually. New service assurance solutions are urgently required to manage this complexity. This was the focus of a TM Forum Catalyst project, which presented its findings at the recent Digital Transformation World event in Nice. The Proactive Service Assurance via Closed Loop Predictive AI/ML team demonstrated a solution which uses AI and ML to predict and prevent service-affecting network issues. The system predicts events on the network as well as the related impacts on infrastructure and associated services. It then automatically responds with preventive and/or corrective actions to prevent the problem reaching the customer.

Real data, real problems

The Champions of the Catalyst are Du and Sri Lanka Telecom. TM Forum Catalyst Champions bring a specific business challenge and in this project, they also contributed real network data. Champions work alongside Participants – Enghouse Networks, NetYCE and Florida Institute of Technology in this case — to rapidly develop a solution.

“Today’s transforming networks are extremely complex, consisting of legacy and SDN/NFV infrastructure, and require new service assurance software solutions to ensure automated, reliable and swift resolutions.” said Valentin Vaduva, CTO at Enghouse Networks. “Together with our partners, we are demonstrating a pragmatic solution for proactive service assurance, presenting the case for closed-loop automation via artificial intelligence.”

Enghouse Networks provides AI-powered Service Assurance; NetYCE is focused on automated provisioning in response to network events; and the Florida Institute of Technology provides data science and analysis. The data from the Champion carriers covered 90 days and included 2.5 million events and 8,500 tickets. Decision tree algorithms were trained on a selection of fault types to create predictions. One of the identified use cases (based on being a common issue) was that of a port exceeding bandwidth utilization. This was then mapped to network actions, and these actions were executed on a simulated network, resolving the problem.

The team was able to predict with a 99.98% degrees of accuracy when the threshold-crossing use case issue would occur. The whole process can be fully automated, depending on the issue. If the problem isn’t suitable for full automation or the prediction confidence is below a certain level, the system can present a range of solution options for an engineer in the Network Operations Center (NOC) to make a final decision.

Start small

In the historical data, the biggest issue was port utilization (13% of total faults) but 80% of all problems were linked with 15 types of events.

Steve Lowndes, Principal Solution Engineer, Enghouse Networks, explained: “You don’t have to try and find 50 different problems you can automatically predict and resolve. Even a small number will have a significant impact.”

One thing the team noted is that better ticketing information datasets are required to support machine-learning-driven resolution automation, and this will be fed back to TM Forum’s collaboration team.

Pasan Nishantha, DGM/IT Solutions & DevOps, Sri Lanka Telecom, commented: “This project is an eye-opener to identify the missing components and tools to support proactive Service Assurance. It will also be a path to alarm and fault correlation, and improve the assurance process.”

Business benefits

The Catalyst team says that their solution could have a positive impact on various business metrics, and ultimately, the bottom line. Reduced service issues and network interruptions should cut the customer churn rate; required truck rolls; mean time to repair (MTTR); network queries into the contact centre; and mean cost to repair (MCTR).

Nishantha says that through what they have learned in the Catalyst, “Sri Lanka Telecom hopes to improve NOC operations and reduce fault-handling time from hours to minutes, or, ultimately automate them altogether”.

Ready to go

Aravind Chennuru, Software Architect, Enghouse Networks, said: “We built a solution architecture to showcase how the AI/ML fits in the service assurance platform. Our designed framework focused on predicting the most frequently occurring issues and enabling closed-loop proactive issue resolution through automated system provisioning.”

The team says that the solution they came up with in just one six-month sprint of the Catalyst is already commercially viable. They put this down to the system being set up on Azure Cloud, providing a ready-made AI environment, as well as the use of TM Forum standards and best practices.

As part of this work, the Catalyst team used and will contribute back to TM Forum’s Digital Maturity Model; the AI Maturity Model; AI Data Model; Business Metrics Scorecard; and Customer Experience Management Lifecycle Metrics. In the next phase, the team plans to pick a more business-critical use case and focus on closing the loop to solve problems. Further, the team will look to bring in further sources of potentially network-affecting information. As well as traditional OSS data – weather information, notifications of sporting events or even real-time road accident data etc. could be pulled in via APIs.