Formulating High-Performance Coatings

Coatings enhance the performance, durability and finish of painted surfaces and are commonly used in a variety of industries including automotive, marine, aviation, infrastructure and construction.
 
 
 
Recently, many owners of white cars have experienced first-hand the frustration of an underperforming coating.[1] Car paint that begins to peel, in the absence of obvious damage, such as a major chip or scratch, is typically the result of poor performance under UV exposure from the sun or simply a poor paint job.[2]

This example demonstrates that designing a coating system is a complex task; formulators must carefully select ingredients to reach and sustain performance under harsh conditions and remain robust to application conditions while adhering to environmental regulations.

Project Details

The Challenge

Our client wanted to reduce the number of unique ingredients required in order to produce a family of products (several high-performance coatings). Reducing the number of ingredients in each formulation has many benefits, including lower inventory costs and simplified production.

The Results

ProSensus reformulated several coatings products with a smaller set of ingredients. After successful field trials were completed by the client, the reformulated products were slated for commercialization.

Our Approach

ProSensus’ FormuSense accelerates product development by allowing formulators to simultaneously optimize the selection of ingredients, formulation ratios, and process conditions to reach targeted quality properties.

Where possible, formulation ratios and ingredient properties are combined to generate mixture properties. Mixture properties are a powerful aspect of our modeling approach, as they allow new ingredients to be considered in future formulations.

Model Building

The user-friendly no-code platform in FormuSense simplifies the tasks of evaluating and structuring available R&D data, as well as building predictive models on qualified data and interpreting the results.

A sample coatings dataset is used below to demonstrate the simplified workflow. Here, a predictive model was built on four quality variables (hardness, gel fraction, glass transition temperature, and viscosity) from:

Model Building

Mixture properties for 4 of the ingredient classes (catalyst, isocyanate, resin, and solvent) were included in the model. The available ingredient properties for these 4 classes sufficiently characterize the impact of an ingredient, removing the direct reliance on formulation ratios. This means that new ingredients from these classes can be considered in simulation and optimization of new formulations. Mixture properties for the additive class were not included in the model since there was insufficient variation.

Example: In this sample dataset, 17 unique resins have been used in past formulations, and each formulation combined 1-8 resins (in addition to ingredients from other classes). Since the glass transition temperature (Tg) for each resin is known, the combined resin Tg can be generated for each formulation using an appropriate mixing rule (linear, weighted average in this case). With this mixture property included as an input variable in the predictive model, the impact of using a new resin in a future formulation can then be predicted if the Tg of that new resin is known.

Interpreting the Model

FormuSense fit the dataset with 8 principal components in a PLS model. Several intuitive multivariate plots are shown below that summarize key correlations.

The score plot shows how formulations are related to one another, R2 displays model fit, y-weights show how quality (y-) variables are related, and VIY shows which x-variables have the strongest correlation to quality variables.

Hardness, glass transition temperature, and viscosity are fit with an R2 of over 70%, while gel fraction has a lower R2. As shown in the VIY plot, isocyanate mixture properties have the weakest correlation to the quality variables, while all other data blocks have multiple variables with stronger correlation.

With additional visualizations such as an observed vs predicted plot, coefficients plot, and x-variable line plot, individual variables with the strongest correlation to a specific quality variable can be investigated and confirmed.

As seen above, the variable with the largest coefficient for hardness is film thickness (a process condition), which was an expected result. To confirm this relationship, the formulations with the highest film thickness are highlighted in white on the x-variable line plot, and as evident, these same formulations are located on the bottom half of the obs vs pred plot. This observation provided domain experts confidence in the developed model.

Beyond confirming expected relationships, the PLS model identified more subtle relationships as well. For example, domain experts were already aware that the resin has a strong impact on glass transition temperature (evident by several large coefficients in the resin mixture property block), but the PLS model also found that evaporation time (a solvent mixture property) is positively correlated to glass transition temperature.

Formulating New Products (or Reformulating Existing Products)

One of the advantages of using PLS for quality prediction is that the model can be inverted to formulate new products using numerical optimization.

With a defined objective function and relevant bounds and constraints, ProSensus was able to reformulate 3 high-performance coatings using the desired smaller set of ingredients. New ingredients were also considered in the reformulation, thanks to the inclusion of mixture properties in the PLS model.

Get Started

Learn more about accelerating product development with FormuSense, and start your 30-day trial today.

Want more information? Request an introductory meeting with ProSensus.