Manufacturing Systems at Scale with Big Data Streaming and Online Machine Learning

Real time analysis of data collected from the shop floor opens the path towards efficient scheduling of batch execution for large scale distributed manufacturing systems. Prediction of the shop floor activities has a great potential to reduce manufacturing costs, by providing the information required for operational decisions like preventive maintenance, automatic remediation or scheduling optimization. Research has been focusing on how machine learning algorithms can be used to better understand and extract insights from historical data collected from manufacturing systems. However, in the current manufacturing environments, driven by mass customization and short time to market, these approaches fail to be agile enough to be useful. In this paper we propose a real-time machine learning approach for large scale manufacturing systems that can predict various scenarios before service degradation occurs, thus allowing for corrective actions. At the same time, outliner detection algorithms can be used to evaluate the system's health at a holistic level. Scalability requirements are achieved by modelling the architecture around data streams processed in real time by map-reduce operations. The concepts presented in this paper build on recent developments on flexible, distributed and cloud based manufacturing, where these real time actions can be efficiently implemented.