Applying machine intelligence to network automation
Advances in computing power, cloud architectures, digitalization and big data analytics are opening new opportunities for artificial intelligence (AI). AI is making the leap from use cases that mimic human behaviors to large complex systems that leverage human capabilities. Within the field of AI, there has been rapid progress in machine intelligence, a discipline which augments the structuring and modeling of machine learning with reasoning and planning techniques.
- Within the field of artificial intelligence (AI), there has been rapid progress in machine intelligence, which is the result of adding AI techniques such as reasoning and planning to machine learning.
- Machine intelligence will help service providers to handle the growing complexity in mobile networks.
- AI and machine learning can be used to leverage the skills of experienced engineers and technicians.
- Automated decision-making relies on knowledge collection and its subsequent organization into a graph of interlinked domains – known as the “telco knowledge graph”.
- Machine intelligence is allowing systems to expand in size and complexity while improving productivity.
Over the last six decades, AI has experienced recurring cycles of optimism followed by disappointment when it has not met inflated expectations. It is no secret that interest in the field is running high once again. This time, however, it is here to stay. Tools and techniques from AI are rapidly finding their way into all corners of the digital landscape. Clear examples of this development are emerging in mobile network operations and maintenance.
From enhanced mobile broadband to the Internet of Things (IoT), 5G will enable mobile networks to support diverse demands. Networks are rapidly growing in terms of the number of devices served and are increasingly complex. One way to handle this is to leverage the skills of experienced network engineers and technicians. AI and machine learning can be used to distribute their knowledge in autonomously managed network operations, as well as in the field, to assist installation and maintenance tasks.
Using structured data as the input, machine learning software builds models, rules and procedures. These results are then applied to new data as they are collected, enabling machine-based, automated decision-making. Adding further AI techniques such as reasoning and planning to machine learning enables the creation of applications we term “machine intelligence”.
A cornerstone in developing automated decision-making is knowledge collection and the subsequent organization of the knowledge into a graph of interlinked domains. Within the telecom industry, the knowledge that exists in the heads of experts or documented in natural language – such as product descriptions, trouble reports and customer support requests – must be transformed into structured, machine-readable data.
The structured data is referred to as the “telco knowledge graph” (see the figure above). The information is also correlated – for example, product instructions solving a specific issue in the network are linked to trouble reports, which document how the issue was solved in the field, and the empirical knowledge of experienced field technicians.
In this article, the use of the telco knowledge graph to automate aspects of network operations and maintenance is illustrated by two prototypes:
- An application for automating Network Operations Centers (NOCs), enabling network self-recovery for machine-resolvable issues
- A mobile digital assistant to guide field technicians through fault resolution for hardware-related errors
As the centralized monitoring and control center of a telecom network, the primary role of an NOC is to maintain network availability and operational efficiency through fault and performance management.
Today, typical NOC challenges are largely met through technicians handling incoming alarms (fault management), identifying root causes of the alarm conditions and implementing appropriate solutions. These processes require domain experts to code solutions, which are complex to implement and maintain as the network technologies and architectures evolve.
The prototype NOC software enables automatic fault management by applying machine intelligence techniques. This enables it to:
- Map composite conditions from historical information (performing intelligent grouping of cross-domain alarms for detection using pattern mining techniques)
- Form rules from the composite conditions using machine learning
- Detect incidents based on the rules
- Identify root causes, and derive appropriate actions by mapping root causes to resolution procedures from system or solution documentation
The prototype produces rules independent of technology, topology and the architecture of the infrastructure. As a result, it is a reusable component that generates data behavior patterns for further applications in incident detection and analysis.
Network management will become a largely autonomous operation, with insights, rules, policies and workflows being continuously developed and refined. Imminent fault conditions will be predicted and corrective actions performed.1
Intelligent digital assistants are a way in which machine intelligence can be applied to tasks on radio base station sites. Installation, configuration and maintenance are expensive and time consuming. The mobile device application assists technicians by performing diagnosis and troubleshooting. This reduces the time spent on site, while increasing quality assurance.
The prototype uses a combination of visual object detection technology and semantically annotated product documentation to guide a technician to complete a given task.2 For example, in the case of troubleshooting a faulty cable adaptor, visual object detection and an augmented reality (AR) application identify and indicate the faulty port, while the steps to resolve the fault are also displayed.
Another example is the use of the application to recognize and locate various components of a radio base station. By tapping on a component’s image on screen, the technician can retrieve more information about it from the documentation. The images below show the assistant in use.
Machine intelligence is used to prepare the telco knowledge graph for the process of troubleshooting. Two sets of input data are needed: a set of images for the object detector and product documentation. The documentation’s references to hardware components are linked to the detected objects and vice versa.
The visual object detector is based on Convolutional Neural Network (CNN) architecture. In one step, such a system performs:
- Feature extraction from the input image pixels
- Prediction of the type of visual object
- Prediction of the location of the visual object in the scene
The object detector can be used in a client-server configuration (running on a powerful Graphics Processing Unit machine), or as a stand-alone application (running on a technician’s smartphone or tablet).
The product documentation typically exists in HTML or PDF files and follows loosely defined structural guidelines.
For the content of these documents to be correctly interpreted by the application and presented to technicians, it needs to be converted to a machine-readable format through a process known as “knowledge extraction”. This involves software extracting the information and transcribing it into a structured information model such as a graph.
Machine intelligence is making its way out of the lab and will increasingly be applied to mobile network operations and maintenance. Leveraging the expertise of network engineers and technicians will allow systems to expand in size and complexity while improving productivity.