MACHINE LEARNING

AI in Logistics Hackathon with BMW Group and Splunk

At the end of 2018, BMW Group teamed up with Splunk to organize a Hackathon event hosted at Antonine University in Beirut, Lebanon. Around 60 students from seven different Lebanese universities competed to solve an “AI in Logistics” use case. Machine data was provided by BMW Group and each team had access to Splunk Enterprise and the Splunk Machine Learning Toolkit to build their solution. Shortlisted teams then had the opportunity to pitch their idea to a panel of experts during the Smart Beirut Summit, later that week.

How AI supports Logistics Planning

BMW Group is currently facing a fast and sustainable process of change. Due to the rising number of car models and configuration possibilities, the complexity in the logistics supply chain increases. Consequently, logistics planners need to be supported in an optimal way with the help of intelligent systems to reduce complexity, increase efficiency, but to also make the daily work of hundreds of employees easier. During the hackathon, 20 teams each with three members collaborated on building a solution that assists logistics planners in their workflow.

Hacking with User Centric Design

If we follow the principles of Design Thinking and User Centered Design methodologies, we need to put ourselves in the shoes of a logistics planner. This is what all participants intuitively also started to consider, and designed their solution not only around the best set of machine learning algorithms but also strongly with the user in mind. Using an Angular based frontend they implemented their UI ideas to capture the given workflow in the best way possible, while maintaining the look and feel of a serious enterprise application designed for daily use at BMW Group.

Multi-Stage Predictive Analytics with Human Feedback

Now imagine you need to plan complex logistics processes on a daily basis. At the first step you have a set of items that you plan to dispatch in a global distribution network. Each item needs to be packaged depending e.g. on sizes, weights and many other constraints. This was the first multi-class prediction model that attendees built and tuned to give an immediate answer to the planner’s question “What packaging should be used for my item”. If the prediction was incorrect, the system can collect feedback from the logistics expert to get smarter over time. In the next step we can use the packaging information to predict the material flow process indicating what and how many packages can be put together, for example on a container, or in special cases for fragile or dangerous goods. In the same way logistics experts can give feedback into the system, retraining models accordingly to incorporate new knowledge. Such a system also aligns very well with BMW Group’s corporate future strategy “The next 100 years” to keep intelligent systems future proof and incorporate “human centered machine learning” - as we learned at .conf18. Needless to say that Splunk’s Machine Learning Toolkit not only provided the participants with speed to accelerate their model creation, validation and operationalization, but also gave the flexibility to construct a multi-stage predictive analytics system without getting lost in writing lots of complicated code to achieve this goal.

And the Winner is … Team Spirit!

Finally, after only 24 hours of hard work each team really achieved some remarkable results. The top 6 teams pitched their work to a jury of experts including Splunk and BMW representatives as well as experts from government and academia. I was personally impressed by how all participants worked together - even helping out other teams despite competing for valuable prizes. The three winning teams ranked as follows:

With all these positive impressions and valuable time spent together, I want to thank all contributing organizers and participants and look forward to the next developments and new ideas that come from their data journeys.

Happy Splunking!

Philipp Drieger
Posted by

Philipp Drieger

Philipp works as Senior Sales Engineer at Splunk. His background is in data visualization and analytics with experience in automotive, transportation and software industries. Philipp is especially interested to leverage Splunk as a data platform for analytics and visualization. As an SME for Business Analytics, IoT and Machine Learning, Philipp enjoys working with Splunk customers and partners across DACH and EE.

Join the Discussion