Predicting failure produces success
Lynn Michelangelo
Nov 25, 2025
Justin Miner, a mechanical engineering doctoral student and his faculty advisor, Sneha Prabha Narra, an assistant professor in mechanical engineering, won first place in a failure location prediction challenge, held by AM Bench, a NIST-led organization that provides a continuing series of additive manufacturing (AM) benchmark measurements, challenge problems, and conferences.
The primary goal of the challenges, which are held every three years is to enable modelers to test their simulations against rigorous, highly controlled additive manufacturing benchmark measurement data. All AM Bench data are permanently archived for public use using comprehensive, custom data management systems.
The pair accepted the award at the International Mechanical Engineering Congress & Exposition® held by the American Society of Mechanical Engineers in Nashville, Tennessee from November 16-20.
Narra said, “This is yet another great example of students being inspired by advanced coursework at CMU and integrating it into the research we do to tackle real-world challenges.”
She explained that Justin’s approach sheds light on a critical aspect of uncertainty in the prediction depending on the image processing approaches one uses in the CT data analysis pipeline.
After participants were given detailed build parameters, material properties, and X-ray Computed Tomography (CT) scans of select titanium components in March, they had until the end of August to submit their results. They were to make predictions of median S-N curve (graphs that relate how much repeated stress a material can handle to how long it lasts before breaking), specimen-specific fatigue strength, and specimen-specific fatigue crack initiation locations.
Source: Justin Miner
X-ray CT of one of the NIST challenge specimens reveals internal porosity, with the predicted killer pore highlighted in red. This pore was identified on the fracture surface during post-mortem analysis, validating the prediction.
“Working without knowing the expected results meant I had to be very thorough, checking and re-checking that each part of my X-ray CT analysis was accurate and well-supported,” said Miner.
Narra appreciates the overall value of these exercises.
“We are grateful to our NIST colleagues for organizing these challenges that contain rich and curated datasets that have broader impacts for overall model development, calibration, and validation efforts beyond the benchmark competition.”