Implementing any model based industrial application requires practitioner
to perform tests on the plants. Historical data can be mined to obtain
high quality data segments which can be used for model identification.
Best available segments in historical data are those with high quality
data and minimal disturbance effects with respect to the choice of model.
Historical data can be considered to be a collection of data segments with
different properties concatenated together temporally. Compared to SISO
systems, MIMO systems pose a variety of challenges like correlated input
moves, collinear inputs and the regions where only a few of the input
variables are changed. This work focuses on finding the best available
data segments from a given historical data set of a MIMO system.
Hierarchical multi-label segmentation approach is used to annotate both
the quality and presence of disturbance for each data segment.
Effectiveness of the proposed algorithm is demonstrated on simulated data
sets. All the high quality data segments are identified with minimal data
loss, with reasonable accuracy.