doi: 10.53962/qpde-b1qe
Originally published on 2024-02-02 under a CC BY 4.0
I want to write a paper explaining how manufacturing data has to be handled to be properly used in data analytic models.
Often, research does not take into account the information on where in the process data was measured. The multisource data structure and the fact that much of the data are timeseries makes correlations between variables (nearly) meaningless. By incorporating expert knowledge of the manufacturing process into the data analysis, it becomes possible to predict data in subsequent machines at every point in time. But there are no models yet that can find relations between multivariate timeseries. As an addition, I will show that simply transposing the data enables classification of produced items based on the data observed during their production.
Considering the underlying structure of the data will increase the potential of data to improve manufacturing processes in realtime and get better insights into the relations between the various parts of the manufacturing.