Predictive Maintenance using AI – Case Study for MBNL

Data is at the heart of many artificial intelligent (AI) applications, and the source of considerable value to those who own and control it. But many data-rich organisations do not understand how to extract all that value from it.

MBNLThis was the challenge facing Mobile Broadband Network Ltd (MBNL), a company jointly owned by mobile telecom carriers, EE and Three (H3G). MBNL provides and manages all of the telecom infrastructure for the two carriers, including over 20,000 masts, each of which reports its performance on a very regular basis (in some cases every 15 minutes). And any problems reported by the network are recorded in an Incident Management system, with notes added by the engineers following a site visit to fix the issue.

MBNLPredictive Maintenance AI therefore had plenty of structured data around the equipment performance but far too much to be able to crunch using traditional statistical methods. The incident notes, on the other hand, were often written in free-form text making any structured analysis very difficult indeed.

MBNL were originally introduced to Andrew Burgess, a strategic adviser on AI, when he gave a talk to the company’s staff at one of their annual off-site meetings. MBNL subsequently engaged Andrew to carry out a study to identify opportunities for AI across the whole business, working closely with senior staff from MBNL, led by James Barber, Technical Contract Manager for Transmission. The output from that study included a heat map of where each of those opportunities were, the benefits that could be gained, and a roadmap of how they could be implemented.

Following discussions with Pat Coxen, MBNL’s Managing Director, a proof-of-concept was selected which would try and predict fan and aircon failures in the mast cabinets. Failures in these pieces of equipment can result in over-heating and therefore cause downtime for the mast, severely affecting network performance in that cell.

Andrew brought in Kortical, an AI company that has developed a unique data science platform that vastly simplifies the process of managing data and algorithms. Kortical worked with MBNL’s data team to identify the data that was available and, most importantly, what it all meant.  According to Gartner, typical AI / ML projects take between 1 and 2 years just to get to the pilot stage, but using the Kortical platform meant that MBNL were able to do this in 6 weeks, iterating on a wide variety of data and models.

Predictive Maintenance AI

Initially the team focused on the ticket data, which provided records of equipment failure and, through the engineers’ notes, the cause of failure. Kortical used the Natural Language Processing (NLP) capabilities of their solution to extract the different failure reasons so that they could be structured and analysed. The first results were fairly good, allowing MBNL to predict fan or aircon failures for the upcoming quarter in about 21% of cases.

The real breakthrough, though, came when smart meter data was added to the mix. Although not all sites have smart meters installed, thosethat do will report on a wide range of indicators every 15 minutes or so. Feeding this data into the Kortical platform along with the ticket data allowed much higher levels of accuracy to be achieved. They found the system is able to predict 1 in every 2 fan or aircon failures for the month ahead. This is, of course, of huge benefit to MBNL, allowing them to proactively maintain the pieces of equipment in question and prevent downtime of the network.

Predictive Maintenance AI

From an AI point of view, having the Kortical platform meant that over 50,000 machine learning model iterations could be assessed, allowing the team to focus on getting the most value out of the data. The smart meter data provided the ability to significantly improve the prediction accuracy levels. One interesting aspect is that the system identified predictive failure signals that humans would have missed completely, in this case minute fluctuations in the power usage of the equipment.

The proof-of-concept has now moved into a live trial that will run for six months. If the results of the PoC are borne out, then MBNL will be able to significantly improve the performance of the network for the carriers and, of course, the mobile phone customers. But that is only the beginning of the journey: there are other pieces of equipment that can fail besides fans and aircon units, and there are other aspects of running a telecom infrastructure that will benefit from predictive analytics. MBNL will be working closely with Andrew Burgess and the Kortical team to make sure that they are able to realise the full value that is possible from exploiting AI in their business.

To learn more about how AI can benefit your business, please contact Andrew Burgess.

Share on:

Leave a comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Follow on Feedly