Citrine Informatics’ customers are leveraging its artificial intelligence (AI) platform to optimize and streamline the development of coatings and other materials, enabling faster innovation and improving performance in a wide range of industrial applications, including corrosion control.
CEO Greg Mulholland shared his insights on the future of using AI for coatings with Materials Performance.
Materials Performance (MP): Can you explain how AI is being applied to the formulation of coatings specifically for corrosion control?
Greg Mulholland (GM), CEO, Citrine Informatics: Every problem in materials and chemistry is an exercise in balancing priorities. Lighter is easy. Stronger is easy. Lighter and stronger is hard. In coatings for corrosion control, those tradeoffs also exist.
Durability in many dimensions—adhesion, self-healing capabilities, ease of application, cost, environmental sustainability, and many other considerations—all trade off against one another, and there are no universal perfect materials. AI is now being used to optimally balance the property envelope of new coating formulations in a way that is very difficult for the human brain. AI excels at multi-objective optimization of this type, allowing scientists to rise above the basic optimization and think strategically about how to solve the problem.
More importantly, AI learns differently from people. The best AI can combine the known relationships that scientists already understand with data that will allow it to discover new relationships that might not be obvious to scientists today. This synergistic relationship yields incredible results.
A perfect example is when Citrine trained a metallic corrosion model on data alone. The Citrine Platform infrastructure was able to find some known and some surprising relationships. But excitingly, when an alloy scientist added an understanding of the pitting resistance equivalency number to the model, the AI was able to discover even deeper relationships and generate better corrosion-resisting metallic coatings.
Similar effects exist in formulated coatings as well: taking advantage of known science alongside data and not trying to relearn everything we have spent 500 years exploring in chemistry.
MP: What are some of the biggest challenges the corrosion control industry faces that AI can help address in the development of coatings?
GM: Corrosion is a process that happens over time, which makes the testing of corrosion protection a time-consuming and expensive endeavor. AI has repeatedly been used to understand leading indications of failure so that formulations can be screened and only sent for final testing when there is a higher chance of success, saving both time and money.
This can take the form of AI-accelerated life testing, AI-driven non-destructive testing, and the proactive digital prioritization of tests where the most promising or uncertain candidates are tested first to more quickly explore the chemistry space.
The other challenge that AI is best able to help with is the sheer complexity of the problem. Formulations can contain many different ingredients, processed in many different ways, and they are then applied in different ways to different types of surfaces. The formulations need to not just protect the asset they are applied to from corrosion, but also be cost-effective, sustainable, easy to apply, and sometimes beautiful!
With so many variables and so many properties to target, it is very hard to do multi-objective optimization without AI. A design-of-experiment (DoE) approach would take hundreds or thousands of experiments to carry out. AI cuts that down using smart math and allows scientists to be strategic conductors of science rather than requiring them to dig into the minutiae for hours and hours.
MP: How does an AI platform accelerate the discovery of new materials for corrosion-resistant coatings compared to traditional trial-and-error methods?
GM: We have consistently seen a reduction of >50% in the number of experiments needed to reach targets compared to trial and error in the first year of a team using AI. That rises to 80% to 99% in subsequent years as the model learns and the scientists learn how to use the model.
It does this by not only predicting the performance of a formulation, but then also calculating how uncertain it is in the prediction. This means that you can then make strategic decisions about which experiments to take to the lab—a safe bet or a more exploratory option.
There are at least three important differences between AI and more traditional methods:
1.) AI takes advantage of historical learnings and knowledge. A human scientist in a trial-and-error approach does this, too, but even a team of scientists can’t remember all of the historical learnings and use them in a synergistic way. Using DoE is actually worse at using historical knowledge: it simply doesn’t. Every DoE is naive to history, so while the math is helpful, each DoE is starting from square one.
2.) AI has a goal in mind. A scientist is always trying to hit a set of new targets, but a DoE doesn’t consider those targets. Scientists can push toward the targets as they learn from a DoE, but they don’t have a “digital partner” helping to push the frontier forward.
3.) AI can look in new places. Humans are often limited in their creativity by their own experience. AI operates differently. It learns and then explores into new spaces efficiently. This yields totally new research directions and strategies.
MP: Can you walk us through the data sources that feed into your AI models when formulating new coatings for corrosion protection? How do you ensure the data is comprehensive and accurate?
GM: Each customer has their own instance of the Citrine Platform. They upload their own data and make models themselves, after initial training, supported, of course, by our experts. This makes sure that their data and expertise are not shared with their competitors. The platform has tools to analyze the data and will, for example, highlight outlying data in case there is a typo or something needs fixing.
The data is usually not comprehensive when the first AI model is created. But our software has been specifically developed to work on small datasets. It is chemically aware and leverages the knowledge of product experts to guide it in the right direction.
Citrine’s external research team, who work with universities and government organizations on grant-funded projects, have worked with others on a database of electrochemical metrics for corrosion-resistant alloys. Your audience might be interested in this paper about it in Nature.
When we develop these databases, we are able to use them on the behalf of our customers and users, but they are only a support. If everyone works from the same data, they lose their competitive advantage. Being able to build on your private corpus and knowledge is how AI is able to create durable competitive advantage.
MP: What role do machine-learning algorithms play in predicting the long-term performance and durability of corrosion coatings under various environmental conditions?
GM: We have customers who have created machine-learning models to predict the stability of formulations in both the short and long term. They have found that the short-term tests together with other properties of the ingredients and formulation are an indicator of long-term results. While AI does not negate the importance of stringent long-term tests, it does guide the researcher such that they need to run fewer of the long-term tests before they have found the right solution.
MP: How does the AI system prioritize specific corrosion factors when designing coatings for different industries?
GM: The product experts using the AI platform choose what is important. They choose which factors to predict, and they choose what targets and constraints to set.
What is nice about the platform is that once you have an AI model set up that predicts the properties you are interested in, then when a new request for proposal (RFP) comes in from a customer with a different set of targets, you can go in and change your targets and get candidates to take to the lab in a few clicks and a few minutes.
Sometimes, but not often, this is optimizing a single property. More often, RFPs are asking for a balanced response across eight to 20 properties, and AI allows the rapid tailoring of recipes to this property profile. AI is ushering in the era of mass customization.
MP: How does AI help to reduce the time and cost of research and development (R&D) for corrosion-resistant materials, and what impact does this have on the broader coatings industry?
GM: As well as reducing the number of experiments needed as mentioned previously, the platform also makes R&D teams more productive by making knowledge and data more accessible across the team. Researchers can easily see what experiments have already been done.
New employees can review the data, AI models, and search spaces used by experienced staff and reuse them or learn from them and adapt them. This avoids the major unnecessary cost of redundant experimentation, which in some studies have been estimated to be 30% or more.
It is also very useful if you want to efficiently reformulate a product range. A search space is simply a statement of the constraints you are working within, what process settings are possible, which ingredients can you use, etc.
If you are working to reformulate a whole product range, you can reuse data, reuse AI models, and just tweak the search space for each product.
MP: In your opinion, how will the integration of AI-driven approaches in corrosion control coatings evolve over the next five to 10 years? What new possibilities do you foresee?
GM: I think that the acceleration of R&D will spur innovation and enable us to solve many of our biggest challenges. R&D functions will still need experienced, knowledgeable people, but those people will be able to get more done, and their role will shift to be one where good judgment between various options presented will be more prized than raw creativity.
A model I like to think about is the deployment of digital spreadsheets in finance. It is now before most of our times, but until the advent of digital spreadsheets like Excel, spreadsheets were paper. People would manually calculate and fill in cells. There was some resistance to computers early on, but now, if you ask any finance person if you can take away Excel, they will laugh at you.
In more technical fields, similar changes have taken place. No aircraft, car, phone, computer, or house is created without at least some digital design. Why? It allows the designer to experiment in a very low-cost way. A huge clay mold of a car is hard to modify (e.g., “what if I put the exhaust here?”, “what if I change the rake of the windshield?”) but a digital one is not. What’s more is that the designer can simulate effects on performance and fuel economy, not just styling.
This is the capability that is coming to materials. Rather than looking at a fixed list of materials, materials buyers are going to expect a dynamic experience where they can work with their suppliers to tailor a solution that fits their needs precisely. Those who fail to adapt will struggle to engage in this new market.
MP: For corrosion engineers who are new to AI, what key advantages can they expect when collaborating with AI-based platforms in their coatings development processes?
GM: Our customers have been surprised by what they have learned from using the platform. Because the platform is transparent and displays the factors that the AI model has found important in a prediction, product experts can check to see that it is identifying things that they know from their scientific intuition are important, but then sometimes something comes up that they weren’t expecting, but that they can see why it is happening.
It is these aha moments that help deepen the product expert’s own domain knowledge that I think our users enjoy most. Of course, the AI will also save you time, which means you might be able to try a more innovative experiment and have a big breakthrough.
Editor’s note: This article first appeared in the April 2025 print issue of Materials Performance (MP) Magazine. Reprinted with permission.
