Comparing machine learning methods on concept drift detection for Predictive Maintenance

In this work we present a comparison of various machine learning algorithms with the objective of detecting concept drifts in data streams characteristical for condition monitoring of industrial production plants. Although there is a fair number of contributions employing machine learning algorithms in related fields such as traditional time series forecasting or concept drift learning, data sets with sensor streams from a production plant are rarely covered. This work aims at shedding some light on the matter of how efficient the depicted algorithms perform on concept drift detection to pave the way for Predictive Maintenance (PdM) and which intermediate data processing steps therefore might be beneficial.