A Machine Learning approach to Predictive Control: study on a real industrial application

The aim of the thesis is to build a system of data collection and analysis, through the integration of a distributed storage and processing system and machine learning techniques, at a small scale. A dataset of industrial data is available for predictive control analysis. Specifically, the dataset contains historical time series of several parameters regarding a manufacturing process, annotated with breakdowns. Our goal is the develop classification strategies that identifies the conditions that cause a breakdown, exploiting Machine Learning tools. Hadoop and TensorFlow, or analogous softwares could be used.