Here is the story of Tristan, Rick, Sarthak and Muhammad and how they secured the 2nd place prize in our first Oerlikon Digital Hub Hackathon. Find out more about them and their Data Science Solution below.
Meet "Additive Manufacturing 5.0", the winners of the 2nd Prize at the Oerlikon Digital Hub Hackathon
Who are the people behind “Additive Manufacturing 5.0”?
Tristan is a Junior Software Developer that comes from a little town close to Munich, currently working as a Genius Admin. He is a big fan of AlphaGo and discovered through playing GO how cool deep learning can be and how productive one can be in the Python ecosystem.
Rick is a Software developer, coming from North of Hamburg and working in the Oerlikon Man Made Fiber unit. Most of his free time is dedicated to his wife and two kids, but he also manages to squeeze in some board games or visits to Larp conventions, like the Drachenfest in Diemelstadt.
Sarthak Gupta is a Graduate Data Scientist at Munich RE, originally from India and currently based in Munich. In his free time, he plays and codes computer games, activities that help him continue to be excited about coding outside working hours and space.
Muhammad is a Senior Data Scientist and a co-founder of a consultancy firm in Munich, that delivers AI and Machine Learning solutions.
The hack. The team
A number of things sparked their interested and brought them to our event: curiosity to experience writing code in a team and working with real industry data, hope to learn new things from other participants, solving problems that could create a real impact in a niche industry like the ones we are present in.
The solution
The team chose this Predictive Maintenance challenge as they thought the benefit would be the biggest if they create a running solution. In this way, they could help the technicians to avoid the upcoming failure, which otherwise can slow down the production capacity, and to increase efficiency and save costs.
They received two weeks of sensor data from a fabric spinning production line and were challenged to find a model that could predict failure states five minutes before a negative impact on production takes place.
In order to do this, the team had to find an efficient way to consolidate and transform the big data set, deal with the inherently skewed distribution of rare anomalies surrounded by tons of regular data, play around with the size of the time window to hide from the model and preferably choose a model that offers some explainability. They settled on using Python for data science pipelines, Azure for hosting models and web app, and XGBoost, after flattening the time series with manually engineered features and deployed it in the cloud with a nice analytics interface.
What did they enjoy the most at the hackathon?
Tristan: It was fun to nerd out with so many smart people and learn about their diverse backgrounds.
Rick: The Mood, the Foodtrucks and the big help from doubleSlash. Such great guys.
Sarthak: The challenge was unique and really creates an impact. The organization of the hackathon was flawless and I enjoyed most the pitch session.
Return to the Hackathon blog post to read the other stories.
Feedback or questions? Get in touch with us!
