- Alchemite™ is being applied to optimize the quality and yield of natural seaweed polymers to replace petroleum-based plastics
- A key application is to the soluble films used in laundry capsules, avoiding microplastics entering the environment
- Machine learning has cleansed data and identified key process parameters influencing formulation properties, helping to focus development work on productive pathways
Summary
At an Intellegens webinar, Dr Alex Newnes, CTO of sustainability innovator PlantSea discussed their use of the Alchemite™ software in developing sustainable alternatives to plastics. PlantSea’s natural seaweed polymers can replace petroleum-based plastics such as those used in laundry and dishwasher capsules, which are a major source of microplastics. These capsules use polymers such as PVOH in the soluble films that encase the detergent. US-based studies have estimated that 75% of this soluble plastic does not degrade in water treatment and is passed through into the natural environment. Emerging regulations challenge the use of such chemicals, but there are no suitable high-quality alternatives that meet performance and cost requirements.
PlantSea has developed a natural polymer system based on a seaweed biorefining process that can be a drop-in replacement for existing soluble film plastics. As it develops this system for scale-up and commercial delivery, PlantSea seeks to optimize product quality and yield. To help achieve these goals, PlantSea is now applying the Alchemite™ machine learning software, aiming to extract insights from R&D and process production data sets that will reduce dependency on trial-and-error experimentation and save time and raw materials. There are two focus areas for this work – the biorefining process and the formulation of soluble films. In both cases, PlantSea’s initial interest was to understand and enhance their data and to explore trends and relationships within that data to identify which process parameters were influencing key targets. With many different process parameters and metrics, this is a complex problem.
At the time of the presentation, the project was relatively new. PlantSea had been applying machine learning for just a couple of months. But Alex Newnes was able to show how the machine learning model was already filling gaps in their data and delivering valuable insights. He identified the graphical analytics available within Alchemite™ as particularly useful. These enable scientists to view and explore trends within the data and to drill into specific data points, all while offering clear information on the model uncertainty.
One trend plot, for example, enabled the PlantSea team to find errors in their data which “would have been very difficult to spot in conventional spreadsheets” but had potential to negatively impacted their quality standards. Another plot enabled them to understand the relative impact of process parameters such as temperature, reagant concentrations, or additive volumes in determining their key target properties. “It’s been really useful for seeing these multidimensional trends,” Alex Newnes explained, “It’s much quicker than anything we’ve been able to generate using other tools. [This is] influencing how we proceed with our R&D, pointing us towards where we need to add more data or increase our repetitions or variability.”
After just a short period applying Alchemite™, PlantSea is already gaining valuable insight, even from relatively straightforward exercises to understand and use trends in their data. The next steps planned in the project were to use the predictive functions of Alchemite™ to explore the likely outcomes of potential process or formulation changes and Design of Experiment in the design of new products. We look forward to seeing the results!
