ProSensus, Inc.

The Multivariate Analysis of Batch Processes course is offered over three days and will cover topics on multivariate applications as presented by Professor John F. MacGregor and ProSensus staff. ProSensus, Inc. is a world leader in the development and use of multivariate data analysis methods for the analysis, monitoring and control of batch processes. Professor John F. MacGregor (Distinguished University Professor, McMaster University and President of ProSensus) is the developer of most of the modern multivariate data analysis methods for batch processes.

1. Review of Latent Variable (LV) Modeling

• Rationale & Concepts behind LV modeling
• LV models (more detail than in base course)
  • Principal Component Analysis (PCA)
  • Partial Least Squares (PLS)
• Diagnostic Plots
  • Score plots
  • Loading plots
  • Contribution Plots

2. Introduction to Batch Processes and Data Structures

• Nature of batch data
  • Initial condition, trajectory and quality data (Z, X, Y)
• Different problems/objectives
  • Analysis/troubleshooting using historical batch data
  • Identification of important sources of variability and their impact on quality
  • Monitoring of new batches (SPC) - (i) raw material monitoring; (ii) end of each batch monitoring; (ii) real-time monitoring
  • Prediction of end product quality or batch end point
  • Control of final product properties
  • Optimization of process operations to achieve desired product quality
  • Scale-up and transfer between plants
  • Merging of batch data at different scales (pilot plant, production plants) to improve understanding

3. Overview of Industrial Batch Data Sets

Data sets used throughout the course are described.

4. Modeling and Analysis of Historical Batch Data

• Purpose of analysis
• Different approaches to the modeling & analysis of the trajectory data
  • Brief discussion of approaches
• Capturing trajectory information by recording discrete landmarks
  • PCA/PLS analysis with historical landmark data
  • Workshop on industrial data using software
• PCA/PLS on the unfolded trajectory data (X array)
  • Alignment of trajectories
    • Purpose
    • Retaining the alignment information
    • Different approaches to alignment with examples
  • Different ways of unfolding, mean centering and scaling
    • Batch-wise unfolding
    • Scaling and centering
    • Nature of variance explained by PCA on batch-wise unfolding
    • Dynamic nature of the decomposition
    • Capturing non-linearities and changing covariance structure
    • Score and loading plots for the interpretation of batch behavior
    • Contribution plots for the diagnosis of specific events
    • Workshop on industrial data using software
  • Observation-wise unfolding
    • Scaling and centering
    • Nature of variance explained by PCA on observation-wise unfolding
    • Advantages and disadvantages between batch- and observation-wise unfolding
    • Three way decompositions
    • Limitations (unrealistic assumptions for batch processes)
• Conclusions

5. Prediction of End-point and Final Batch Quality

• End point detection
• Issues in prediction of final batch quality
• Missing data imputation in batch PLS - prediction of trajectories
• Prediction of final quality via different approaches
  • Landmark approach
  • Batch-wise unfolding approach
  • Observation-wise unfolding
  • Examples
• Workshop
  • Prediction of final quality using industrial process data
• Conclusions

6. Monitoring (MSPC) of Batch Processes

• Monitoring and SPC concepts
  • Approaches to batch monitoring
    • Assessment only at the end of each batch
    • Real-time monitoring
• Monitoring based on Landmark approach
  • Assessment only at end of batch
  • Real-time monitoring
  • Example
  • Workshop using industrial data
• Monitoring based on batch-wise unfolding
  • Assessment only at end of each batch
  • Real-time monitoring
    • Need for missing data imputation
    • Control charts and limits
    • Contribution plots
    • Example
    • Workshop using ProSensus MultiVariate
• Monitoring based on observation-wise unfolding
  • Real-time monitoring
    • Control charts and limits
    • Contribution plots
    • Example
• Discussion of different monitoring approaches
• Workshop
  • Setting up monitoring charts using industrial batch data

7. Overview of Control of Batch Processes by LV Methods

• Sources of variation
• Concept of active control over final quality
  • Feed-forward control for raw material variations
  • Need for trajectory and final quality predictions
  • Defining an end point target region
• Mid-course correction concepts
  • Control of final quality attributes
  • Objective functions and algorithms
• Industrial examples
• Discussion and Recommendations

8. Optimization of Batch Processes by LV Methods

• Important concepts for active use of LV models
• Optimization of batch recipes and trajectories to achieve a desired product
  • Nature of data needed
  • Specifying desired quality
  • Inversion of PLS model
  • Application to a batch polymerization process
• Product scale-up and transfer between plants
  • Data structures of different plants
  • PLS analysis and inversion
  • Application to a pulp digester process
• Combining data from multiple scales for modeling and optimization
• Workshop