Estimation of Stochastic Time Lags between Data Sources in Distributed Production Facilities Based on Cross-Correlated Signals

The increasing degree of automation in modern industrial production is accompanied by a growing number of data sources located in the shop floor at the machines and products. Methods of data mining allow to make statements and predictions about the machine processing hidden in heterogeneous data. In condition monitoring, however, large and often indefinite time delays exist between the collected process data and quality measurements at distributed work stations, which make these analyses difficult. In order to estimate the time lags, a probabilistic directed graph can be used. Parameters of such a graph are the sojourn times of the production goods in the stations. To identify these times, cross-correlation and transfer entropy can be applied to signals with known collinearity. To compare the two methods, both simulated and real-world data of an industrial return sand cooler are used. For evaluation, we propose a novel method for image-based time delay distribution measurement in bulk material processing by use of a fluorescent contrast medium. A dependence of the estimation result on the signal’s statistic is shown. Nevertheless, without knowledge of the signal model a statement about the magnitude and scattering of the stochastic time delay can be made.