Process Modeling and Optimization O. Rodionova Institute of Chemical Physics, Moscow

1
Process Modeling and Optimization O. Rodionova
Institute of Chemical Physics, Moscow
MIR Space Station, Star City, Feb.17 2005
2
Based on Paper

• Process control and optimization with simple
  interval calculation method
• Pomerantseva, O. Rodionovaa, and A. Höskuldssonb
• a Semenov Institute of Chemical Physics, Moscow,
  Russia
• b Technical University of Denmark, Lyngby,
  Denmark

in print
3
Outline

• Introduction
• Real-world example description
• Passive optimization
• Sic- in brief
• Active optimization
• Conclusions

4
PAT a gift for chemometrics(Process Analytical
Technology)
MAN WITH THE GIFT by Natar Ungalaq
Guidance for Industry PAT A Framework for
Innovative Pharmaceutical Development,
Manufacturing, and Quality Assurance Pharmaceutica
l CGMPs, September 2004
FDA U.S. Department of Health and Human
Services Food and Drug Administration
5
PAT Tools

• Multivariate tools for design, data acquisition
  and analysis
• Process analyzers
• Process control tools
• Continuous improvement and knowledge management
  tools

(Guidance )
6
Multivariate Statistical Process Control (MSPC)

• MSPC Objective
• To monitor the performance of the process

• MSPC Concept
• To study historical data representing good
  past process behavior

• MSPC Method
• Projection methods of Multivariate Data
  Analysis (PCA, PCR, PLS)

• MSPC Approach
• To plot multivariate score and control limits
  plots to monitor the process behavior

7
Multivariate Statistical Process Optimization
(MSPO)

• MSPO Objective
• To optimize the performance of the process
  (product quality)

• MSPO Concept
• To study historical data representing good
  past process behavior

• MSPO Method
• Projection methods and Simple Interval
  Calculation (SIC) method

• MSPO Approach
• To plot predicted quality at each process
  stage

8
Real-world Example (strong drink production)
9
Technological Scheme. Multistage Process
10
Data Set Description
Y preprocessing
X preprocessing
11
Quality Data (Standardized Y Set)
12
Overall PLS Model
13
Passive Optimization in Practice
Thinker by Rodin
14
Main Features

• Objective
• To predict future process output being in
  the middle of the process

• Concept
• To study historical data representing good
  past process behavior

• Method
• PLS and Simple Interval Prediction

• Approach
• Expanded Multivariate Process Modeling
  (E-MSPC)

15
Expanded Modeling. Example
16
Expanded PLS modeling
17
Simple Interval Calculations (SIC) in brief
Triple Mobius by F. Brown
18
SIC main steps
19
SIC-Residual and SIC-Leverage
They characterize interactions between prediction
and error intervals
20
Procedure Flow-Chart
Initial Data Set X,Y
PLS/PCR model Fixed number of PCs
SIC-modeling
RESULTS
yhat RMSEC RMSEP
21
SIC Prediction. All Test Samples
22
Expanded Modeling PLS SIC
23
Expanded SIC modeling
24
Samples 2 3
25
Samples 4 5
26
Passive Optimization. Stage V
PLS/SICprediction
27
The Necessity of Active Optimization

• F. Yacoub, J.F. MacGregor Product optimization
  and control in the latent variable space of
  nonlinear PLS models. Chemom. Intell. Lab. Syst
  7063-74, 2004
• B.-H. Mevik, E. M. Færgestad, M. R. Ellekjær, T.
  Næs Using raw material measurements in robust
  process optimization Chemom. Intell. Lab. Syst
  55135-145, 2001
• Höskuldsson Causal and path modelling. Chemom.
  Intell. Lab. Syst., 58 287-311, 2001

28
Active Optimization in Practice
Let Us Beat Our Swords into Ploughshares by
Vuchetich
29
Dubious Result of Optimization
Predicted Xopt variables are out of model!
30
Main Features

• Objective
• To find corrections for each process stage
  that improve the future process output (product
  quality)

• Concept
• Corrections are admissible if they are
  similar to ones that sometimes happened in the
  historical data in the similar situation

• Method
• PLS1, PLS2, SIC

• Approach
• Multivariate Statistical Process Optimization
  (MSPO)

31
Intermediate Stage
The Scheme of Three Data Block Modeling
32
Optimization Problem
Stage I XIW1, W2, W3 ,S1, S2, S3
33
Linear Optimization
Linear function always reaches extremum at the
border. So, the main problem of linear
optimization is not to find a solution, but to
restrict the area, where this solution should be
found.
34
Optimization restrictions
I. All process and quality variables should be
inside predefined control limits. xi?1 and
yj?1 for every i,j
II. Adjusted variables should not contradict
process model For new (x,z) maximize(xtbztc)
w.r.t. z, z ?Lz
35
How to define Lz ?
PLS1 XY X ? y Xtest
l0mh, l1mhsh, l2mh2sh,
l3mh3sh, r0md, r1mdsd, r2md2sd,
r3md3sd,
36
Three optimization strategies
37
Strategy G1
38
Sample 5 Normal Quality Insider (G3)
39
Sample 3 Normal Quality Abs. Outsider (G3)
40
Sample 4 Low Quality Outsider (G3)
41
Results of Optimization. Quality variable
42
Results of Optimization
43
Conclusions
The presented optimization methods are based on
the PLS block modeling as well as on the Simple
Interval Calculation
Application of the series of expanding PLS/SIC
models helps to predict the effect of planned
actions on the product quality, and thus enables
passive quality optimization.
For active optimization (1) No improvement in
quality obtained inside the model (2) To
yield a considerable improvement in y, the
optimized variable values should be located in
the boarder of the model (3) It is obligatory to
verify that optimized values do not contradict
the process history.