Big Data analytics for improved maintenance and flawless operation of the in-flight internet

First, we helped Gogo migrate their on-premise data solutions to the cloud in order to collect and process a considerable amount of data from more than 20 sources. Not only did the migration to AWS cloud expand the data processing capacity needed for further data analytics solutions, but it also saved Gogo costs spent on licenses and the on-premise infrastructure. Next, we built a cloud-based unified data platform that collects and aggregates both structured (i.e. systems uptime, latency) and unstructured data (i.e. the number of concurrent Wi-Fi sessions and video views during a flight). For this purpose, our engineers made an end-to-end delivery pipeline: from the moment when the logs come from the equipment to the moment when they are entirely analyzed, processed, and stored in the data lake on AWS.

Further, in order to predict the antenna equipment failure and reduce the no-fault-found rate (NFF), we applied the data science models, namely Gaussian Mixture Model and Regression Analysis. We set up a data retrieving process from seriously degraded antennas and correlated it with weather conditions and the antenna construction.

We also provided Gogo C-level decision-makers with comprehensive reports on financial data (data consumption, purchases, data usage, etc.) and operational health (modem performance, WAP, Service Level Agreements, etc.). We are developing a reporting tool that allows identifying the users’ pain-points during the first fifteen seconds of the Internet connection. With the help of this data, we identified the point where a lot of passengers had difficulties while connecting to Gogo Wi-Fi.

Together with our client, we completely migrated Gogo data solutions to the AWS cloud and shut down its costly on-premises infrastructure. Our team built a unified AWS-based data platform that collects and aggregates data from 20 different sources and can process up to 3 TB of streaming data per day.

With Data Science and Machine Learning algorithms, we developed models for predicting satellite antenna failures and antenna health monitoring. The solution provides interpretable reasoning behind each recommendation. For example, we identified that when antifreeze liquid is applied, and the plane does a couple of flights, the fluid gets inside the antenna and affects its performance. Therefore, the recommendation was to add an additional layer of protection from the fluid.

• Migration to the cloud allowed our partner to reduce costs on licenses (Cloudera/Microsoft) and on-premises servers and enabled them to control and optimize resources in the cloud based on their processing needs.
• The solution allows Gogo to significantly save operational expenses on penalties to airlines for the ill-performance of Wi-Fi services.
• The no-fault-found rate was reduced by 75%, thus saving costs on unnecessary removal of equipment for servicing.
• Predictive analytics allows predicting the failure of antennas (>90 %, 20-30 days in advance) and ensures servicing of the in-flight equipment at the most suitable time, for example, when there are no flights scheduled for a plane.
• The reasons behind the ill-performance of antennas have been identified (i.e. antennas often failed after the use of antifreeze fluid). That allowed Gogo to prevent some of the typical failure causes (i.e. add an additional protection layer to the antennas).