Rapid Custom Formulation of Silicone Antifoams


Rapid Custom Formulation of Silicone Antifoams2019-08-22T14:45:02-04:00

Project Description

In order for Dow to capitalize on the growing trend of low-volume, high-urgency orders of custom specialty chemicals, ProSensus has developed a novel product customization software tool which is backed by a data driven model.

Multivariate analysis (MVA) was applied to Dow’s historical product development database of silicone antifoams to develop a model capable of predicting key performance properties. Raw material properties, formulation ratios (recipes), and operating conditions employed during mixing were simultaneously considered as the inputs to the Partial Least Squares (PLS) model.

Constrained optimization was then applied to the model to determine if Dow is technically capable of producing the requested product, and furthermore, to accelerate the complex and dynamic business decision of whether or not it is feasible and strategic to fill the customer-requested product order.

The Challenge

The goal of this silicone antifoam RPD project was two-fold:

  • Develop a multivariate model that accurately predicts antifoam performance properties.
  • Use this model to formulate new antifoam products by back-calculating the required blending formula to reach specified performance property targets.

The ProSensus Approach

Following the ProSensus RPD framework, a multi-block multivariate PLS model was developed that predicts the performance properties of the final product mixture by simultaneously considering raw material properties, blending ratios, and operating conditions during mixing.

Model Building

After iteration with Dow process experts, the final PLS model explained 83% of the y-space with 6 principal components.

 

The Results

A predictive PLS model, subsequently used for optimization of new silicone antifoam formulations, was built by ProSensus on Dow’s database that included 22 products and approximately 100 unique raw ingredients (belonging to 11 ingredient classes). Note that each formulation typically contains 10-20 ingredients.

Two validation cases resulted in less than 15% error across each of the six performance properties. The observed versus predicted plots illustrate the model efficacy for one of these validation cases. In these plots, the validation data point is denoted with a red star.

Optimization

After model validation, constrained optimization was then applied to the models to facilitate rapid product development using model inversion. The optimization objectives were to:

  • Reach desired performance property targets
  • Minimize raw material costs
  • Minimize operating costs

In addition, multiple constraints were enforced to: meet regulatory requirements, ensure a physically feasible solution, restrict individual components to specified minimums and maximums, and to restrict certain raw material combinations.

Custom Software

Next, ProSensus developed a custom software tool for Dow to facilitate the rapid product development of new silicone antifoams, using the developed PLS model and constrained optimization problem.
This tool allows the user to provide several inputs including:

  • A formulation cost constraint
  • Targets, bounds, and priorities on performance properties/qualities
  • The availability and bounds on raw materials

The tool presents the results of the formulation optimization in various ways. Key features include:

  • A table summarizing the new formulation recipe
  • A table summarizing the predicted performance properties for the new recipe (not shown here)
  • A score plot showing the location of the new formulation in relation to the database
  • A plot showing the new formulation in relation to the database on the basis of each performance property (not shown here)

References

  1. 1. Brandon Corbett1, Marlene Cardin1, Kristin Wallace1, Alix Schmidt2, Haseeb Moten3, and Rebecca Beeson3, (1) ProSensus, Burlington, ON, Canada, (2) Continuous Improvement Center of Excellence, The Dow Chemical Company, Midland, MI, (3) Dow Consumer Solutions, Dow Chemical Company, Midlant, MI. “Accelerating Product Innovation at Dow Through Multivariate Modeling”. AIChE Spring Meeting, New Orleans, April 2019.