Using Artificial Intelligence For The Design of Coatings

Artificial intelligence (AI) and machine learning (ML) are powerful methods to speed up the development of products, especially in technical areas such as coatings. Utilizing design-of experimentation techniques and advanced statistical analysis assists coating formulators in improving their products’ attributes while keeping up with the ever-growing environmental and regulatory demands.

Utilizing AI and ML techniques and strategies takes formulation science one step further and allows the integration of a greater variety of information to inform decisions.

Based on Garry Froese, CEO of ArmorThane, Artificial intelligence can be applied to any challenge in formulation or synthesis in various methods.

Machine learning models can aid coatings development to determine the performance of formulations or resins before any material is made physically in the lab. Additionally, algorithms for classification and clustering will look for patterns or trends in existing data and identify new heuristics or rules for guidelines that can be utilized in future tests.

Artificial intelligence and machine-learning workflows will enable users to create new formulations and resins by entering the final-use properties they want and using advanced algorithms that suggest the most appropriate formulations and recipes. This approach has already been tested for various formulation and resin types and is expected to become the norm for industry procedures for conducting coatings studies and research.

Modern machine learning and artificial intelligence algorithms can learn from data and search for patterns like humans. They can gain knowledge from the data delivered in large quantities and data inserted in a drop-by-drop manner, similar to what researchers experiment with in the lab. They observe and study the data and then adjust their approach according to the issue they seek to resolve.

Furthermore, AI and ML models are like humans: they grow as they gain time and use and are fed with new information. Due to these features, integrating AI-powered tools for discovery and design to existing processes of empirical research is an ideal fit when performed with care and concern for the scientist using these tools. Everyone on the team that formulates must feel that their work has been amplified or increased by the addition of their work by using AI tools.

A century of scientific knowledge and discovery doesn’t have to be thrown out the windows when creating the AI workflow. Instead, the new AI workflows could be constructed based on an institution’s domain-specific expertise. Machine learning models may utilize already existing chemical as well as physical concepts. Additionally, at the same time, using machine learning and artificial intelligence in In-silico (computerized simulation) formulating and synthesis to forecast the properties of resin and formulation properties before the trial-and-error process that occurs in the laboratory can also increase the sustainability of your business.

With the appropriate algorithm and data, the formulators and experimenters can “feel” a design area and make an informed decision on the future use of material resources as well as the time of people to make new materials become a possibility. The guess-and-check research process is taken from the lab to digital, interactive modeling systems.

The centuries of scientific research and discoveries do not have to be thrown out the window when creating the AI workflow. Instead, the new AI workflows could be developed around an organization’s specific domain and institution information.

Predictive capabilities are also enhanced as the systems grow and learn to make in-silico predictions and optimize more sophisticated optimization.

Testing the real-world properties of resins and formulating performance specifications will always have to be conducted in the physical world. But, a significant portion of the material design will be moved from the lab and into the computational realm as artificial intelligence-based tools develop and become a part of the routine R&D workflows. Ultimately, all formulation design spaces can be explored in a computer environment without the need to create one batch of formulas.

There are many hurdles to conquer before this level of acceptance can be reached. The main issue is information–how it’s structured, how old it is, who is responsible for it, and so on. These concerns exist regardless of the company’s material area of focus or size.

Businesses struggle with the issue of spending time finding the appropriate ones set up versus starting with existing data in its present form. Sapper recommends that businesses get input from several departments and teams within the organization to ensure that long-term database structures can be adapted to the latest information and materials, including new additive chemistry to monomers and reactions.

However, they must show quick results when applying AI to current problems to establish trust. In general, this means that data structures and implementations of databases are typically bespoke projects built around the clearly defined issue being tackled.

Selecting the appropriate data and who can access the data is crucial. Not all historical or existing data needs to be converted into digital format and put into some AI tool. The data collected in the past was not designed with the idea of digitization in mind, which means it is challenging (or even unattainable) to convert the data into new structures.

There must be an equilibrium between importing old data and generating new data through innovative experiments and formulating efforts. Data users who could benefit from it include technical directors, researchers, marketing teams, and many more. Artificial machine learning and intelligence tools must be developed specifically for these users.

Sapper cautions about the need to be aware of bias from institutions where only a select group of users are permitted to alter the prediction algorithms.

Computer models do only what they’re told to do. Human programmers may unintentionally create biases in models during the model development or training process if they’re unaware of biases or are not asking questions. The best method to avoid this problem is to make your design process visible to everyone in the company.

For businesses that wish to use machine learning and artificial intelligence to boost product development but feel overwhelmed by the issues that arise when starting a new program by, assembling a diverse team that includes technical management and executives, program managers and subject matter experts, younger scientists, as well as IT professionals and having it strive to create an initial product that can prove a quick success that could be used to build momentum within the company is the best strategy.

When the program’s support is in place and supported, it will be much simpler to handle the larger issues efficiently. The author also suggests that companies recognize that successfully incorporating AI into their R&D efforts is a continuous process, a complete change in R&D habits, and not a single initiative.

Additionally, artificial intelligence is an area that is continuously expanding and changing. Therefore, it is crucial to embrace a view of using AI in R&D workflows in the many years to come and be prepared to change in the process.

Dow Polyurethanes is one of the companies already on the path of digitalization. Their Predictive Intelligence capability to help develop products, developed in conjunction with Microsoft, makes up a wider digitalization program.

The company was recently awarded a FutureEdge 50 Award to acknowledge the importance of the application of this groundbreaking technology that combines Dow’s material knowledge with Microsoft’s AI and ML expertise speeding up the two-to-three-month process of developing products to polyurethane formulations by as much as 200,000x, reducing the discovery process to 30 seconds, as per Alan Robinson, North America commercial vice-president at Dow Polyurethanes.

We’re challenging our methods of working and disrupting our complete innovation process with customers in the forefront to increase market and customer innovation by using science and digital technology better to address the challenges of tomorrow. This is a significant milestone in Dow’s larger digitalization strategy, with additional initiatives expected to be revealed shortly.

Dow’s strategy of leveraging its coatings expertise by combining machine learning and artificial intelligence is the secret to achievement. He argues that businesses must think of AI as a tool for improving, speeding up, increasing, and enhancing human imagination and efforts, not as a substitute for it.

Intelligence amplifying (IA) was first coined during the beginning of research in cybernetics. We must embrace this view of working or in conjunction with, but not in opposition to, robots, computers, databases, and even automated science.

I know these ideas could sound like science fiction and may be close to being taken from Isaac Asimov’s story. However, the fact is that the future is here! AI-enabled workflows in R&D will boost our capacity to keep innovating regarding coating development for binders.

Leave your vote