Data Mining for Polymer Production
Optimizing Production of Butadiene Rubber
Background:
Performance of the Butadiene rubber is determined by its molecular weight ML, which has a normal range of values: 45±2. In production, the ML value measured every 2 hours
The material will have less elasticity if its ML is too high and less strength if ML is too low. The manufacturing process is complicated to model since it involves heat transfer, mass transfer, fluid flow, and chemical reactions.
Objective:
Exact modeling is impossible, have to mine empirical data from history.
Method:
The ML values from history data were divided into 3 classes: class one for ML < 43, class two if ML falls within the interval of [43, 47], and class three if ML > 47.
Factors:
A total of 45 factors have been identified that will affect the molecular weight ML value of the material. They include temperature, flow rate, feed of catalyst, feed of solvent oil, and so on.
After data mining, we discovered five factors, called principle factors, {Z1, Z2, Z3, Z4, Z5} that have significant effect on the materials ML value, as described by the following table:
Factor
Property
Z1
feed of butadiene
Z2
feed of solvent oil
Z3
feed of catalyst
Z4
temperature of feed
Z5
temperature at lower part of first reactor
Results:
Using MasterMiner, we successfully built a reliable mathematical model for the ML value in the 5-dimenional hyper-space spanned by {Z1, Z2, Z43, Z4, Z5} as follows
ML = 2.111-0.661Z1-0.00636Z2+0.03737Z3+0.01255Z4-0.02397Z5The above model was used to control the material manufacturing with good results. The rate of good rubber was increased from 71% to 95.2%, and production yield rose from 89% to 93%. The net annual revenue was increased by US $0.25 million.
© Copyright, Zaptron Systems, Inc. Back to Home.