NASA Develops Improved Friction Stir Welding Techniques

Combination of machine learning, simulations, and data analysis helps solve issues with weld strength and defects.

Mar. 16, 2026 at 1:52pm

Got story updates? Submit your updates here. ›

NASA's NESC team developed several innovative tools and techniques, including machine learning models, data analysis frameworks, and physics-based simulations, to improve the self-reacting friction stir welding (SRFSW) process used to join components of the Space Launch System rocket at the Michoud Vertical Assembly Center in New Orleans. The assessment helped identify root causes for poor tensile strength and low topography anomalies in welds, leading to process adjustments that eliminated these issues.

Why it matters

Friction stir welding is a critical technique for joining high-performance aerospace alloys like Aluminum 2219 that are traditionally difficult to weld using conventional methods. Improving the reliability and quality of these welds is essential for ensuring the structural integrity of launch vehicles like the SLS rocket.

The details

The NESC team used a combination of machine learning, statistical modeling, and physics-based simulations to tackle the issues with the SRFSW process. They developed a deep learning model to automatically detect and segment low topography anomalies (LTA) in weld fracture surface images, eliminating the need for manual identification. An integrated data-ingestion framework was created to compile the diverse data streams from the welding process. The team also implemented a space-filling design of experiments to efficiently explore the complex parameter space, and built a physics-based computational model of the SRFSW process to provide additional insights.

  • The NESC assessment was conducted in 2026.
  • The Michoud Vertical Assembly Center in New Orleans houses NASA's Friction Stir Welding lab, which is used to join components of the SLS rocket.

The players

NESC

NASA's Engineering and Safety Center, which led the assessment to improve the friction stir welding process.

Donald S. Parker

The contact person for more information on the friction stir welding techniques developed by the NESC team.

Got photos? Submit your photos here. ›

What they’re saying

“The NESC team developed a machine-learning model to detect and segment LTA in weld images. This step was crucial to linking process parameters with LTA occurrence in an objective, nonbiased way.”

— Meagan Chappell, Author (nasa.gov)

What’s next

The process models and tools developed by the NESC team have been shared with stakeholders for ongoing use in improving the friction stir welding process for the SLS rocket.

The takeaway

By leveraging a combination of advanced data analysis techniques, including machine learning and physics-based simulations, NASA was able to identify and address the root causes of quality issues in a critical welding process, demonstrating the power of data-driven innovation to solve complex manufacturing challenges.