Novelis Speeds Beverage Can Design Optimization

Novelis engineers streamlined the design process using AI Cloud instances with SigOpt’s Intelligent Experimentation platform.

At a glance:

  • Novelis Inc. is a leading sustainable aluminum solutions provider and the world leader in aluminum rolling and recycling.

  • The combination of SigOpt’s Intelligent Experimentation Platform and Intel® hardware-based cloud instances aids Novelis engineers in designing stronger, lighter aluminum products faster.

author-image

By

Executive Summary

Novelis Inc. is a leading sustainable aluminum solutions provider and the world leader in aluminum rolling and recycling. Novelis Inc. is driven by its purpose of shaping a sustainable world together. The company’s ambition is to be the leading provider of low-carbon, sustainable aluminum solutions and to achieve a fully circular economy by partnering with suppliers, as well as customers in the aerospace, automotive, beverage can and specialties industries throughout North America, Europe, Asia, and South America.

In the past, Novelis relied on CAD design workstations and trial-and-error methods to identify effective designs for prototyping and physical testing when supporting customers. Today, the combination of the SigOpt Intelligent Experimentation Platform and Intel® hardware-based cloud instances aids Novelis engineers in designing stronger, lighter aluminum products faster. Using SigOpt, Novelis engineers reduced the number of costly and time-consuming physical prototypes needed. The process also helps Novelis deliver solutions for their customers more rapidly than ever.
 

A beverage can is among the most highly engineered objects you will ever hold in your hand. With over 26 design parameters, the beverage can model requires over 67 million finite element computations, which would take a FE analyzer 150 years to analyze each permutation. SigOpt was used to find which combination of parameters best meets design needs while cutting back on required iterations significantly. 

Challenge

When working on behalf of its customers, Novelis constantly seeks to reduce product research and development time and costs. In the past, mechanical engineers and materials scientists at Novelis modeled car components, aircraft parts, and beverage can design using CAD software running on workstations. In a typical design cycle, about ten of the most promising designs were then prototyped and tested physically in the lab. The process required an expensive trial-and-error approach that often took months to complete.
 

“At Novelis, we have always trusted Intel-based components to speed critical simulations whether those workloads run on our workstations or our cloud-based instances for AI.”—Dr. Vishwanath Hegadekatte, Novelis

For the can, even tiny changes in product geometry can result in profound performance improvements. Most people do not realize that an aluminum can is among the most highly engineered products they will ever hold in their hands. Designing a can to be extremely light, but withstand the pressure of carbonated liquids, along with being resistant to damage in the supply chain, creates an enormous scientific and engineering challenge. Among other parameters, a can needs the robustness to withstand internal pressures of 90 pounds per square inch (PSI) yet remain as light as possible to reduce the amount of material needed. When designing a beverage can, finding an optimal design typically requires the evaluation of about 26 different design parameters and over 67 million finite element computations. Since each design simulation requires about five hours to complete on in-house workstations, it would require about 150 years and make the task impossible using traditional human-based methods.1

Automotive safety features are equally challenging for the Novelis team to engineer. Developing an axial crush component involves identifying an aluminum sheet geometry that will fold gradually during a vehicle collision to absorb energy without cracking. In a crash, the right geometry can save lives. However, performing finite element situations for something this complex can take years.

To bring solutions to market more efficiently, Novelis needed a faster and more intelligent way to cut down the number of product models worthy of prototyping—ideally, the best two or three.

“With the help of Intel and SigOpt software, we can explore new design approaches that are impossible for humans alone.”—Dr. Vishwanath Hegadekatte

Solution

The combination of SigOpt Intelligent Experimentation Platform and a cloud provider’s high performance computing (HPC) instances provided Novelis with a flexible and cost-effective solution to iterate product designs quickly and efficiently. The approach also dramatically reduced the number of simulations needed for development while obtaining a better result than legacy methods.

Cloud instances powered by Intel® Xeon® Platinum processors, which offer large memory capacity and improved I/O latencies, combined with SigOpt’s easy-to-use API helped Novelis engineers speed up the iterative design process and provide new insights quickly. Because engineers can work in a multidimensional format to identify ideal designs, the process reduces the number of prototypes required.

Optimization techniques helped Novelis perform tasks like optimizing automotive part geometry and other products without the tedious and expensive trial-and-error-based method used in the past. In the case of a beverage can, Bayesian optimization powered by SigOpt offered the potential to decrease the number of computations needed to predict new can designs for further testing.

“The main thing that surprised us was that the predicted performance bump we saw with artificial intelligence was significantly improved over that of our manual approaches.”—Dr. Vishwanath Hegadekatte

Results

Novelis’ goal was to modify the physical design of the bottom of the beverage can to withstand the most pressure before buckling among other challenges. Using a commercially available method and subject matter expertise, Novelis ran 15,000 simulations to predict an optimal design.

Results from a commercially available optimization method. 15,000 simulations were required to find the optimal solution.

Then using SigOpt, Novelis found a better predicted design while only running 99 simulations. Running fewer simulations reduces the demand for IT resources for this task, which can be used on other projects. Just a few lines of code to connect to the SigOpt API made this possible.1

The improved performance can now be leveraged to reduce the weight of the beverage can while meeting the performance requirements. 

Results from SigOpt Intelligent Experimentation Platform. SigOpt found a 15% stronger design while using only 99 simulations (a 99% reduction from the commercially available optimization).

Learnings

This AI-enhanced HPC workflow also enabled a variety of other benefits for the Novelis team:

Faster Design Iteration: By using a combination of AI and Bayesian optimization to reduce the time required to understand the possible permutations of can design, the Novelis team can more quickly iterate on these can designs to evolve their products. Compounding this benefit, the broader system they have set up also enables them to capture more granular data on each individual experiment in the process, which enhances their understanding of the problem space so they can constantly refine their designs in a much more iterative workflow.

“By cutting down the number of simulations needed for product development, Novelis requires much less time for physical prototyping. We aim to deliver design support outcomes to our clients much faster.”—Dr. Vishwanath Hegadekatte

Efficient Use of Resources: This approach enables Novelis to minimize both computing resources and team time required to achieve exceptional results from their experimentation. From a computing standpoint, SigOpt is designed to enable an AI workflow that vastly reduces the number of computationally intensive computing simulations run by the team. But even more importantly, it also enables them to hone in on just a few possible permutations of the can that need to be tested in real-world physical experiments, which are extremely time and resource intensive. This combination means they can significantly reduce the resources required to get to experimental results, increasing the likelihood that their experimentation pays off for the team.

Greater Impact: By having a workflow that enables faster iteration with more efficient use of resources, the Novelis team is also empowered to pursue more cutting-edge scientific discovery that would not otherwise be possible. This workflow therefore enables their team to transform how they structure and pursue research goals, broadening the variety and scope of experiments they run to effect even greater research impact on their business.

The Finite Element Analysis (FEA) workload Novelis uses benefits from several key features in Intel Xeon Platinum CPUs which provided improved manufacturing performance on the 3rd Gen Intel Xeon Scalable platform vs. prior gen by 1.52x2.

Solution Components

Spotlight on SigOpt

SigOpt, an Intel company, is an Intelligent Experimentation platform that combines complete experiment management with powerful optimization to empower AI and HPC teams to maximize the impact of their models. With a few lines of code, SigOpt enables modelers to track experiments, visualize metrics and parameters, and execute scalable hyperparameter optimization. As a result, these modelers realize better results in fewer experiments, boosting their time, compute, and resource efficiency in the process.

Spotlight on Novelis

Novelis is the world’s largest aluminum rolling and recycling company employing around 12,500 people across four continents and nine countries. The company specializes in developing unique and sustainable products used in the packaging, automotive, aerospace, and other industries.

Learn More

“How Novelis Optimizes the Design of Aluminum Cans Using SigOpt”

Download the PDF ›

Product and Performance Information

2Performance varies by use, configuration, and other factors. Learn more at http://www.Intel.com/PerformanceIndex.