Anomaly Detection for Multivariate Time Series Data in Sintering Processes

Heat treatment technology is a fundamental technology in the production of components, and as such it is indispensable and must be considered in the sustainable transformation to a CO2-neutral economy and society in the coming decades. For this reason, the sintering process is analyzed in more detail in this paper as a representative example of heat treatment processes. An unsupervised anomaly detection model is proposed that identifies data anomalies based on the parameters of the sintering process. To provide a holistic view of the sintering process, over 100 parameters from the pre-heating zone to the cooling zone of the sintering oven are analyzed. When an anomaly is detected, this approach allows to determine in which sub-process the anomaly has occurred to intervene specifically in this sintering zone. By preemptively identifying anomalies and intervening accordingly, the potential production of substandard components is prevented, thereby enhancing the sustainability and reducing CO 2 emissions in the sintering process.