Material development: getting there faster with co-pilot AI

We live in fast-paced times. This is reflected in consumer preferences. In addition, climate change is creating a rapidly changing regulatory landscape for chemical and materials companies. They have to adapt quickly. But the thorough research, development and commercialization of new materials can be lengthy. Artificial intelligence can speed up this process.

In an interview with K-MAG, Josh Tappan talks about the Citrine platform, how artificial intelligence (AI) can help companies develop materials and make the plastics industry more sustainable.

What is the purpose of your Citrine platform?

Josh Tappan: The Citrine Platform combines smart materials data infrastructure and AI to accelerate the development of next-generation materials and chemicals through an iterative, AI-guided workflow – which has demonstrated the ability to reduce R&D time by 90+ percent, optimize our customers' product portfolios, and codify research IP, which enables its reuse and prevents its loss.

Who is the IT solution aimed at? What does it look like to work with the platform?

Tappan: The Citrine Platform is aimed at product development and application engineering organizations in the materials and chemicals and discrete product manufacturing industries. Customers interact with our platform through a web-based user interface or a python-based API. A typical workflow for our customers is:

1. Import and structure historical data into a graphical format (Graphical Expression of Materials Data, or GEMD) that visually represents the context of a material from procurement to final product.
2. Define a "design space" on the platform, which is the set of possible candidates. Customers define this based on their scientific domain knowledge and manufacturing or supply chain constraints.
3. Train an AI model (with as few as 20 rows of data) to predict properties or performance targets of interest for new materials candidates.
4. The AI model identifies new materials and/or processes within the design space that are likely to hit the performance/property objectives.
5. Our customers synthesize and test the most promising candidates in the lab based on the AI-generated predictions and their knowledge of the problem at hand.
6. If the new candidates hit or exceed our customer's objectives, they proceed to scale-up and commercialization. If candidates do not hit some or all of the customer's objectives, they incorporate this data back into the model, improving the models predictions, and increasing the likelihood that new candidates will meet their objectives.

What role does AI play in material development? In what ways can it make work more efficient?

Tappan: We see it AI a co-pilot that enables product developers, scientists, and engineers to do their jobs better. Materials development problems are highly complex and multi-dimensional, so it's impossible for any human (or team of really smart humans) to evaluate all of the possible solutions and materials that might meet their goals in a reasonable amount of time. With our platform and an AI-guided product development workflow, our customers have accelerated the rate at which they develop new materials by 90+ percent compared to traditional methods.

On top of that, as scientists start to incorporate their expertise into the modeling process, new scientists and adjacent teams can begin to reuse these digital assets to get a jumpstart on new projects and product development initiatives. This both decreases the "ramp-up time" for new projects, and ensures that senior scientists are sharing their knowledge across the team and preventing knowledge loss as veterans begin to retire and new scientists enter the workforce. This domain knowledge can take a number of forms on our Platform from well-structured graphical datasets, "features" or "descriptors" of physical phenomena and known processing relationships that can be used as inputs to a machine learning model, design spaces that define the set of constraints and objectives for specific applications, and reusable AI models.

What further developments are conceivable in the future?

Tappan: In the short term, I think we'll see a lot more companies using AI, simulation, and experimental results in concert to rapidly design new materials, chemicals, and molecules. We've also invested in capabilities to unlock materials design as a new degree of freedom in part-level simulation for discrete product manufacturers. By leveraging materials informatics within part and system-level simulation tools, product developers can use part performance requirements to drive material development rather than designing around materials constraints or limiting themselves to materials available in supplier catalogues.

In the medium-term, we'll start to see more advances in polymer informatics. Right now, AI works well with formulations, small molecules, and inorganics, but we're seeing the field of polymer informatics rapidly evolve. We recently participated in a workshop hosted by MI on harnessing big data to develop new polymers and are actively working on advancements in the field of polymer informatics. 

In the longer-term, we will see "self-driving" labs adopted in industry. There have been recent advances in academia in combining AI, robotics, simulation, and laboratory automation software in a closed-loop guided workflow to autonomously discover and develop new materials. Many of these pieces already exist in isolation in commercial labs, so it's natrual for companies to invest in combining these capabilities to develop commercial "self-driving labs" in the future.

To what extent can digitization contribute to greater sustainability in the plastics industry?

Tappan: A more sustainable plastics industry is critical for the future of our planet and humanity, and digitalization plays a crucial role in this sustainable transformation. Recently, we've been excited by what we're seeing in the market, and believe that companies are making significant investments into making their sustainability goals a reality. Higher performing products alone will not meet our global environmental and societal needs. However, sustainable products that do not perform as well as conventional alternatives present business problems for producers (we commonly refer to this as the "paper straw problem").

Our customers have utilized AI-guided product development as a tool to optimize for both performance AND sustainability, whether it's adapting existing products to use more sustainable ingredients and feedstocks, optimizing manufacturing processes to decrease carbon footprint, energy or water usage, or designing high-performing plastics that are recyclable or biodegradeable.