On Preprocessing Multi-channel Sensor Data for Online Process Monitoring

This paper discusses online monitoring production processes based on multi-channel sensor data. Particularly the problem of transient and anomaly detection is addressed for which a processing framework consisting of a preprocessing module and a reasoning engine is outlined. While there is much theory available in the literature for the reasoning engine this is not the case for the preprocessing which massively depends on the physical interpretation and semantics of the data. The paper addresses these problems and proposes new normalization concepts based on regularization especially for making transients of multi-channel data comparable and adequate for further processing by a reasoning engine. A proof of concept is demonstrated by means of real data from an injection moulding process.