Machine learning for large scale manufacturing data with limited information

Abstract

Improving the efficiency of the production plants has always been focus of manufacturing industry. Recently the utilization of data analytics tool is dramatically increased since these methods bring new insights into already existing data. Nevertheless, manufacturing industry in general is still reluctant to make the data available to researcher due to privacy issues. One such example is the challenge sponsored by Bosch and run by kaggle.com, where anonymized data was made available to data scientist with very limited description. In this work, we present our solution to the Bosch assembly line performance challenge, specifically in respect to dealing with raw big data without detailed explanation. The data science methods applied were used to successfully predict internal failures along the assembly lines, although no details of the structure and line description was available.