Automated Machine Diagnostics for Tritium's DC Fast Chargers
Aside one of our early-adopters, Tritium is an Australian manufacturer of DC Fast Chargers for electric vehicles. With more than five thousand chargers operating worldwide, they are one of the market leaders. And in the past couple of years, Tritium has expanded massively, landing huge contracts with Chargefox, Ionity and the U.S. army to name a few.
Alongside a lot of new chargers, the Service Engineering team is seeing an increasing amount of new issues coming in every day. As a result, the team expanded just as quickly while seeing an increasing Mean Time To Diagnose. Due to the time consuming chain between end-user and service engineer (especially Charge Point Operators and distributed field technicians slow this process down), downtime of the chargers became an issue. This resulted in two negative aspects: A sub optimal customer experience as they cannot fully rely on their EV charger and an unreasonably high work load for the Service Engineering team.
Luckily, the team used a customer service ticket system which recorded and labelled all issues. You could literally filter these tickets for a particular issue and get all past occurrences. This helped significantly to organize training data, as logs of machine sensor data were stored on Tritium's servers. One by one we collected machine logs of the more frequently occurring issues and uploaded them on the Amplo AI Management Portal. Our Automated Machine Learning pipeline cleaned and enriched the data and made classification models which compare each issue with a set of healthy machine logs. Based on machine data of past issues, these models are able to automatically diagnose whatever subsystem of a broken charger needs replacement. Once the accuracy (and precision!) was exceeding Tritium's requirements, we reported this back to Tritium and made the trained machine learning models available over our AI Management Portal.
Our machine learning models were tested extensively by Tritium, with at first only a few service engineers using the AI Portal to diagnose new service tickets. After a trail period of 3 months and many models further, Tritium decided to directly use our API endpoints. This allows them to integrate the machine learning models into their own Salesforce as if it is part of their own software! Now whenever a new service ticket is created, the models are called instantly and diagnose the issue within seconds. This reduced the Mean Time To Diagnose significantly, shortening the communication loop from end-user to field technician. Broken chargers are fixed faster, the service engineering team's backlog decreased and allowed them to focus on other important Quality Control tasks.