Machine learning-led optimisation of home & personal care formulations

Personal care products, including shampoo, toothpaste etc., contain a complex mixture of ingredients that defines how well the product works, including getting the active ingredients to the correct place, how they feel in use, their fragrance and so-on. The blend of ingredients also defines how environmentally friendly and sustainable the final product is. Next generation personal care products must have excellent performance as well as being sustainable.

In this project, you will be working with Unilever, a world-leading company that manufactures a range of home, personal care and food products that are used by 3.4 billion consumers daily, and employs over 138,000 people. Unilever is one of the foremost global industrial research organisations with an equally strong reputation in chemistry and the wider sciences. In its publicly stated vision of 'Making Sustainable Living Commonplace', Unilever has prioritised the development of next generation products that are sustainable and environmentally friendly. You will be developing methods to discover new formulations of personal care products. You will do this using a data-driven, closed-loop approach, where machine learning (ML) designs new combinations of molecules, which are then made and tested. After analysis of the results, the ML system uses this new data as feedback to improve its predictions and hence completes one cycle of the loop.

This project is focused on developing the ML/artificial intelligence (AI) component of the closed-loop system. You will use Python to develop and optimise ML code that discovers novel mixtures of ingredients for use in personal care products, fulfilling both performance and sustainability goals. The trade-offs in these goals make finding the optimal design challenging, especially as there is an exponentially large combination of different possible ingredients. However, ML algorithms developed from high-quality data sets can act like a chemistry GPS system that guides us to the new personal care products we seek.

The project will be complemented by a second project to design and construct a fluidics system that will carry out the make and test parts of the closed-loop cycle. This means it will both generate data to be used for ML and verify the predictions made. You will use Python programming to add a degree of automation into this experimental part of the research and hence the wider closed-loop approach.

Before the experimental data has been produced, you will use computational models to generate hypothetical data on product performance and, for example, the greenhouse gas impact of the ingredients. This will allow you to test different ML approaches to determine which will work best. A key consideration at this stage will be to ensure that the approach developed will work within a closed-loop setting and hence be able to be continuously improved by feedback on its predictions.

During this PhD project the PhD candidate will develop, design and optimise a machine learning model to explore the large formulation space for specific personal care products, and perform closed-loop optimisation on mixtures of ingredients based on a range of performance and sustainability inputs. They will work as part of a team to implement a machine learning approach to deliver the next generation personal care products that allow us to live more sustainably.