Predictive Maintenance

Telco use cases

• Automated baseline generation and anomaly detection (ABGAD)

  Monitor key performance indicators, quality performance indices and customer satisfaction indices for the detection of violations. This functionality uses predictive maintenance based on the clustering of violations into anomalies and a root-cause analysis of these anomalies.

  Problem

  Baseline generation and its violations require analysis of multiple sources, and comparison of the given parameter values with the typical ones. It requires the designation of typical parameters (baselines). Then, violations are monitored and correlated for investigation of their root cause. There are hundreds of thousands of parameters which should be monitored in typical CSP’s network.

   

  Solution

  Machine learning-based anomaly detection has the potential to identify parameters and events that do not conform to an expected pattern in a data set, and to improve the breadth of detection by uncovering new patterns and patterns consisting of violations of many baselines. Once an anomaly is detected, it can be prioritized. Cases with the highest priority can be analyzed first, therefore speeding up the entire process.

   

  Value

  Automating this type of activity will increase the number of variables and make them more flexible and dynamic. Additional benefits are the ability to create links between many other variables, and to determine the symptoms leading up to the events associated with the degradation of the quality of service even before customer experience is hit.

• Automated situation detection (ASD)

  Often, automated procedures detect all events without identifying the correlation or cause among these alarms. Automated situation detection allows for root-cause analysis for alarms by clustering alarms into situations and allowing users to understand each situation’s classification – whether it is a root-cause alarm or simply symptoms and noise of the real cause of the alarm.

  Problem

  Situation detection requires analysis of multiple sources. In a typical CSP’s network, automated procedures designed to detect threats must monitor hundreds of thousands of events at the same time. Machine learning is used to analyze and remove the reasons reported for multiple events by all media (network elements, network management systems, performance monitoring parts and systems monitoring the quality of services, as well as reports from customers).

   

  Solution
  Machine learning-based situation detection has the potential to identify events that do not conform to the expected, and to uncover correlations between them. This is much more effective than similar work that is currently carried out manually. When the situation is detected, it can be prioritized based on its identification. Cases with the highest priority can be analyzed first, therefore speeding up the entire process and ensuring that the system can operate at the highest capacity possible. The time required to complete this step is also an important parameter, as this is the stage at which services are most likely to be available for CSPs’ customers.

   

  Value
  Boost your productivity and asset utilization by ensuring that situations detected in your system are identified and classified, thereby reducing the overall number of situations in favor of recognizing the root-cause of these situations. The main benefit of ASD is to further reduce the number of events leading to the definition of a single failure. Each event typically generates work in many different teams, which leads to inefficiency within your organization. The work required by your teams will be categorized by root cause of the situations, and the symptoms or noise which are typically reported as well will disappear once the source element has been repaired. It is expected that machines will lead this process continuously, much faster than their experienced operators. In some cases, the machines may outpace the experts.

• Automated problem detection (APD)

  Like ASD, Automated problem detection (APD) performs root-cause analysis on trouble tickets that have been clustered into problems and problem classification, which helps identify root-causes of trouble tickets versus symptoms.

  Problem
  Analyzing events and incidents in telecommunications networks is a common task for telco experts. It relies on the analysis of past events, and on establishing links and correlations between these situations. This leads to the detection of ongoing irregularities that trigger events and emergency situations. This analysis requires the study of factors from a number of sources, and can be a very tedious process.
   
  Solution
  High-efficiency machine learning techniques can be employed to automate this process. By automatically analyzing which item is the actual source of a string of events, the process by which these events are detected and analyzed can be both minimized and improved drastically.
   
  Value
  With this tool, the system using ML functions will be able to identify more problems than the current experts. This is a proactive solution searching for potential new events and situations in the network, where any unsolved problem could lead to several emergency situations, and even more in the long term. This approach reduces workload for many teams and will not disrupt service provision.

• Knowledge accumulation (KA)

  This Machine Learning capability allows your operations to run more smoothly than ever before and reduce costs associated with training and hiring new employees. Knowledge accumulation allows for automated recommendations for operators regarding solutions for situations, problems and anomalies, thereby streamlining the process and ensuring a consistent quality of service.

  Problem
  After an anomaly, situation or problem is detected, it is essential to implement the most effective remedy that neutralizes the irregularity. Today, much of the knowledge used to find the correct remedy is collected in the form of lists of steps to be taken for any operational procedures. These tools are static and outdated, and it is tedious and time-consuming to keep them up to date.
   
  Solution
  Analysis of the decisions taken by experts allows for the creation of a Machine Learning-powered system. In its first phase, this builds on the set of the most effective solutions based on human decisions, and then analyzes this data to find persistent patterns of action and reaction. This data, along with the knowledge derived from experts, will be used to create a system which can automatically make decisions based on the anomaly at hand.
   
  Value
  This solution enables a decision-making process independent of the knowledge of a particular expert, thereby minimizing the discontinuity associated with a lack of experts. Introducing the machine-based process shortens the time required by operators to obtain skills and reduces the time and entry cost for new employees.