by Thor Olavsrud

Predictive analytics gives Owens Corning an edge in turbine blade development

Jun 03, 20207 mins
CIO 100Data ScienceIT Leadership

With help from the manufacturer's analytics center of excellence, the Owens Corning innovation group has reduced the time it takes to test composite materials from weeks to hours.

data scientist woman at virtual monitor user interface tools for data science by metamorworks getty
Credit: Metamorworks / Getty Images

At manufacturer Owens Corning, data scientists have had the rare opportunity to work hand-in-hand with scientists to meld predictive analytics with scientific research to streamline the company’s innovation efforts.

“We don’t typically deal with very in-depth scientific, technical problems,” says Malavika Melkote, director of global information services and the analytics center of excellence at Owens Corning. “We work with a lot of problem sets that are in marketing, customer supply chain, manufacturing. This was the first time we were doing a project with pure science in the innovation group.”

That initiative resulted from the Fortune 500 company’s need to streamline its process for developing and testing materials used in the creation of wind turbine blades. Addressing that challenge required a close collaboration between IT, the analytics CoE, and the company’s innovation group in the form of a team led by Senior Scientific Advisor Eric Carlier.

To grow its wind energy business and meet customers’ requirements, Owens Corning must continuously improve composite material properties through optimized glass composition, higher modulus for stiffer, longer blades and superior fatigue performance for longevity, Carlier says.

“We have a special group within our innovation organization that is focused on developing new materials,” Carlier says. “When we talk about new materials, it’s at many levels.”

Getting predictive

The first of those levels is the creation of glass fabrics that are put into molds to create the blades. The second involves the composition of the glass fibers that give the blades the stiffness necessary to transfer energy efficiently.

“And the third area, which has been the focus for this work, is we are applying a chemical treatment on the glass fiber that bonds it to the resin,” Carlier says.

The process is complex, lengthy, and expensive. It can take months or years to develop a new binder, which makes glass fibers “stick” to resin or plastic to create glass-reinforced composite materials.

“Stiffness is very important and that is why we use glass fibers for the job,” Carlier says. “One of the other requirements is obviously to make sure that the material will last over a certain period of time. The blades are designed for 25 years or more. You need to guarantee the lifetime of the material.”

And therein was the meat of the challenge. While the minimum time to evaluate and predict fatigue performance was a few weeks, that testing had to determine whether the material would be durable over decades. Carlier’s team needed a repeatable, predictive and screen test of composite materials that could prove the relation between product composition and resulting product performance.

By bringing predictive analytics into the mix, the company has reduced testing time for any given new material from 10 days to about two hours, meaning faster product development and reduced cost of iterative performance testing.

The project has earned Owens Corning a CIO 100 Award in IT Excellence.

Blending science and data

Owens Corning Data Scientist Gideon van den Broek took point in working with Carlier’s team to supplement its empirical approach with a data-driven predictive model.

“He partnered with Eric and his team and tried to understand material science in the context of data and used data science to improve speed to insights, iterations, and the overall analysis,” Melkote says.

van den Broek and the data science team worked with a team that included polymer chemists like Carlier, material testing scientists, and composite applications engineers from across North America and Europe. They gathered data about the product properties they wanted to predict and built a model to understand how those properties related to the performance of the final material. They leveraged enterprise data and analytics platforms to speed up the collaboration of the global technical teams by enabling more effective collaboration across departments and around the globe. They also created a global enterprise solution for storage to enable analysis of results across all tests and all product developers’ work.

That’s not to say there weren’t hiccups along the way. The first predictive model van den Broek and his team built was a black-box model that wasn’t explicable.

“We were not confident with that, so we started to have discussions to come up with a model in which we were confident,” Carlier says. “For me, one of the key challenges was to convince my colleagues within the innovation organization that our test methods were the best. When you work with colleagues who are in front of the customer, testing the material according to industry standards, it’s fairly difficult to convince them that yes, we have something that can save time.”

Being able to explain how their model worked was essential to gaining that trust.

From the analytics side, the challenge was building enough domain expertise to really understand the problem and then develop that trust.

“It was just trying to get past that barrier of how deep our understanding of the domain is to be effective in applying data science,” Melkote says. “As soon as Gideon was able to establish credibility as a data scientist who could understand just enough to bring about results, the teams could actually build more sophistication into in and make it more trustworthy, more usable.”

That leap was groundbreaking, Melkote says. The analytics CoE and the innovation team are now working on several additional projects to bring about a deeper data-driven digital transformation in the innovation group.

“It was a really good affirmation that bringing pure data science into the picture to help you make sense of the data and having business partners who can help you through the interpretation, will get you good results pretty quickly,” Melkote says.

Melkote considers developing strong collaboration with subject matter experts, by establishing common language and communication, to be an important component of analytics success. She also recommends keeping teams small and focused with a clear scope, but with the ability to reach out to others as needed for help.

She also notes that analytics and technology leaders should not hesitate to involve themselves in the most complex problem sets.

“Don’t be afraid to get involved in those things because you bring your technical expertise and analytical expertise. You can easily make a connection to the problem set, and with a good partnership you can prove something out that you thought was not possible,” Melkote says.