Predictive Maintenance in Membrane pumps
 During my graduation internship I made a pilot for a Predictive maintenance system using production data implemented with Python. In this project I got the chance to combine my technical knowledge with data science and machine learning. My system was able to accurately predict failures of equipment before they happened, saving a predicted 150 hours production loss yearly. The data I worked with was time-series factory measurement data where everything was logged in company systems. I built a workflow to extract all data automatically and format it into a MS Lakehouse. From there three approaches were tested. An Auto-encoder for anamolous behaviour, an aggregation classifier and a bare Neural Network. In the end, extracting features from the timeseries using TSfresh and then learning from those features ended up being the most reliable and interpretable approach. I was rewarded with an average grade of 9 for this graduation internship, and with that recieved my Bachelor of Science.