Soft Sensor for Melt Flow Index in Polymerization Reactors
An inferential sensor estimates melt flow index (MFI) continuously from process measurements, so quality control does not have to wait for laboratory results that arrive hours later.
Soft Sensor Solution
Approach
A model — data-driven (for example PLS or neural networks) or based on reaction kinetics — maps continuously measured process conditions to the quality variable that the laboratory measures infrequently. Each new laboratory result is used to correct the model's bias, and the estimate is passed to the operators or to an advanced control layer.
Input Variables
- Reactor temperatures and pressure
- Monomer, comonomer and catalyst feed rates
- Hydrogen (chain-transfer agent) concentration or feed
Output Variables
- Melt flow index (MFI)
Model Type
- Data-driven regression (e.g. PLS, neural networks)
- First-principles kinetic model
Update Strategy
- Bias correction from each laboratory result
- Recursive or moving-window model adaptation
Results
-
Kadlec, Gabrys and Strandt review data-driven soft sensors in the process industry — the typical methods, their applications and the practical problems of building and maintaining them, including data preprocessing and model adaptation.
-
Dow Chemical reported nonlinear model predictive control of an industrial polymerization process, in continuous use since October 2012: polymer quality measures that the laboratory provides only infrequently are controlled in closed loop, with the NMPC setting the reactor temperature profile. The process model combines material balances with rate expressions and an energy balance.
Why It Matters
- MFI is a primary polymer quality specification. When it is known only from periodic laboratory analysis, quality deviations — especially during grade transitions — are detected late and produce off-spec material.
- A continuous estimate lets operators or an advanced control layer act on the current quality state instead of the last laboratory result.
Have a control challenge? Let's talk.
Sources
Kadlec, P., Gabrys, B., Strandt, S. — review.
Bindlish, R. (The Dow Chemical Company). Details from the published abstract.
Pattern Overview
In continuous polymerization, MFI is set by reactor conditions — temperatures, feeds and hydrogen as chain-transfer agent — but measured only by periodic laboratory analysis. An inferential sensor estimates it continuously from the process measurements.
When to Use This Pattern
- Process measurements are reliable and continuous, but quality is measured only off-line.
- Grade transitions produce off-spec material because corrections wait for laboratory feedback.
- A dedicated online analyser is not available or not justified.
- An advanced control layer (APC, NMPC) exists or is planned and needs a real-time quality signal.
Design choices
- Data-driven or first-principles: data-driven models are quick to build where historical data covers all grades; kinetic models extrapolate better to new grades and conditions.
- Keeping it accurate: every laboratory result is a chance to correct the model; the update scheme should be decided before commissioning, not after drift appears.
- Trust: operators accept the estimate faster when it is displayed next to the laboratory results it predicts.
Related Patterns
Contact
Send a message
Direct contact
Dr. Rafał Noga
Stay Updated
Get insights on Industrial AI, APC, and process optimization delivered to your inbox.