Interview: “Biggest challenge is accuracy of data”
Where are the limitations on modelling at present?
Partha Majumdar: When experimental data is being used to generate predictive modelling tools, one of the biggest challenges facing coating formulators is the accuracy, consistency and reproducibility of the data. Additionally, incorporating historical data generated using non-identical testing protocols, conditions or apparatus can skew data ‘cleanliness’.
At Dow Coating Materials, we invest a significant amount of time in ‘cleaning’ the data to ensure the model is as representative as possible. In paint formulations, due to combinatorial explosion, constructing models for all available ingredients becomes very challenging. Also, the complex nature of and multiple interactions between the various liquids and solids present in a paint formulation require extensive, validated data generation before a model for predicting formulations and performance can be deployed.
What is the ideal use case?
Partha Majumdar: Subject-matter expertise is important for giving us a preliminary idea about the system we are dealing with. There are two main areas where predictive models can be highly valuable:
a) The development of predictive modelling tools is particularly valuable for facilitating major reformulations of product lines driven, for example, by a regulatory change or the introduction of new, more sustainable or efficient technologies into existing product ranges. Formulators value guidance as a way to limit the number of input variables that we need to evaluate.
b) Acceleration of new raw materials development, where a combination of automated high-throughput methodology and multi-variable design of experiments (DoEs) enables higher dimensional models to be generated and to be used to predict and optimise responses.
Do you see the speed of developments increasing or stagnating?
Partha Majumdar: The continued acceleration of innovation and new product development is in part being driven by the digital transformation occurring across various industries – and coatings are no exception. Focusing on digital infrastructure allows consistent, fast, and reliable data collection. In parallel, advances in modelling algorithms and in model development through commercial or in-house software packages are enabling more organisations to develop differentiating products and services for their customers.
What could help or hinder modelling from becoming established in the future?
Partha Majumdar: The expanded use of predictive modelling tools is dependent on the models being both reliable and relevant, which means that they are developed for the purpose of aiding rather than replacing the formulator. The reliability of the model will always be dependent upon the validity of the data infrastructure. For wider adoption of modelling tools in formulation design, the models must be accessible, reliable, and end-user friendly. If the barrier to access or the complexity of interpreting the output is too high, the tool will be of limited use.