Multivariate Monitoring Batches Online

1
Multivariate Monitoring Batches Online

• Batches data form a 3 way table

2
HIERARCHICAL BATCH MODELLING

• Two steps, gives two models and monitoring spaces
• Step 1 Monitoring the batch evolution
  (Observation level)
• Corresponds to unfolding the 3 way table as
  displayed.
• Individual observations (time points) are the
  unit.
• Scores of PLS versus time are then used as a
  summary of the batch evolution

3
The Batch Level

• Step 2 Monitor the whole batch
• One observation per batch.
• The variables Scores derived from Level 1
  initial conditions and final batch quality (Y)
• Whole batches are monitored, classified, and
  interpreted.
• Quality measurements are predicted by initial
  conditions scores of level 1

4
Monitoring the Observation levelControl charts
of the scores (summary of all variables) in real
timeEvolving (bad) batch in the cook phase of a
digester is shown in black
5
Control Charts displaying a bad batch that is
much longer than average good ones (training set)
in the chip phase
6
Information Panel gives information about current
batch being monitored.
7
Displaying all the phases in a batch digester
with a bad batch clearly visible (black)
8
Many choices for displaying plots
9
Viewing the cook phase in Normalized units
10
Viewing all phases in normalized units
11
Selecting reference batches to be displayed with
current batch
12
Monitoring the batch level during evolution, and
at completion
13
Batch level Score plot as the batch is evolving3
sequential models (M1,M2,M3) at various percent
of completion
14
Current (evolving, shown as arrow) and historical
batches in the Batch level Score plot
15
Batch Level Distance to the model as the batch is
evolving (3 sequential models)
16
Detecting a problem with current batch Marked
time point outside limits in cook phase
17
Contribution plot on that time point indicates
the problem variable
18
Problem is visible in the control chart of the
culprit variable
19
Problem starts to be corrected at later time point
20
Contribution chart of that time point. Culprit
variable starts to be corrected
21
Correction visible in control chart of that
variable
22
Completion of batch 28, DModX is out limitA new
good batch is evolving
23
Contribution plot for DmodX Culprit variable
scores in cook phase between time 130-150
24
Double clicking on a culprit score displays the
culprit variable (here digester pressure)
25
Trend plot of digester pressure, show Batch 28
out of specs between times 130-155
26
Summary

• Simca Batch Online (SBOL)
• Gives opportunities to detect deviations from
  normal batch
• When batches are evolving
• When the batch is finished
• Shows time when problem occurs
• Batch Control charts
• Shows culprit variables
• Contribution plots
• Dynamic plots of culprit variables