Industrial production processes generate huge amounts of streaming data, usually collected by the deployed machines. To allow the analysis of this data (e.g., for process stability monitoring or predictive maintenance), it is necessary that the data streams are of high quality and comparable between machines. A common problem in such scenarios is "semantic shift". For example, a sensor's weight unit might shift from tons to kilograms after a firmware update and still store the collected values to the same variable. In this paper, we discuss semantic shift theoretically and by means of an industrial case study from a production plant in Austria, where several hundred injection molding machines are employed. The data collected by these machines is used to monitor the stability of the production process with machine learning algorithms. In the following, we present and discuss the data preprocessing system we developed for the production plant to handle semantic shift for huge amounts of streaming data.