Creating digital twins of explosive tools provides confidence
In a metal high bay approximately 10 miles south of Sandia’s main New Mexico campus, Sandia engineers are using computerized tomography, or CT, scans and computing power to create digital twins of explosive tools.
Using HPC algorithms, Sandia engineers can detonate precise digital replicas of explosive tools over and over again, unlike real-life explosives which can only be used once, according to Theo Stangebye, Sandia software systems engineer involved in the project. This capability provides Sandia’s national security customers the confidence that these devices will always perform as expected.
Over the past 18 months, Sandia has built the HPC infrastructure and refurbished several HPC clusters to enable this important work at the far edge of Sandia’s information technology infrastructure, where the internet is slow and the engineers can only devote 12 kW of power for computing. These small clusters aren’t intended to replace Sandia’s corporate HPC resources, but rather provide an agile complement to them. They mimic, at a much smaller scale, the technologies that make HPC possible. That’s why it’s called mini-HPC at the edge.
While the engineers only have a few small clusters, they use the same scalable provisioning technology such as diskless booting and Simple Linux Universal Resource Manager queuing.
Forecasting explosive behavior for confidence
John Korbin and Sam Bowie, Sandia mechanical engineers, use CT scans, or computed tomography, and commercial image processing frameworks to turn the X-ray images of an explosive tool into a digital twin of the device. The process is somewhat similar to how CT scans in medicine can help locate a tumor. However, the team’s CT scanner is larger and uses more intense X-rays.
Then Korbin and Bowie use Sandia’s shock physics modeling code to create a high-reliability forecast of the behavior of the digital explosive. Having the CT machine allows them to create accurate digital representations of each individual explosive. This allows them to do inspections and take measurements to forecast how the explosive will perform, affording Sandia’s customers the confidence that the tool they hold in their hands will perform as expected.
By being able to automate the end-to-end construction of digital twins, the team can ensure the quality of each tool much quicker and examine more devices in a day. Some days the CT team produces up to a terabyte of data.
Lean, agile computing to aid the mission
Using this agile approach has allowed the team to find manufacturing defects in prototype tools, as well as smaller variances in devices due to the computer-aided design process. Using these methods and having a dedicated computing resource without an extensive queue has greatly increased the agility of the explosives group. Integrating the computing side with the hardware side has changed the culture of how the team thinks about engineering explosive tools.
Explosives experts create digital twins of every explosive tool before testing them. Sandia doesn’t use HPC forecasting for every device, but the digital twin is available for analysis if the device performs oddly or particularly well. And since the cutting-edge resources are conveniently located, the explosives experts can ask questions and run simulations they would never have thought of a decade ago.
Next steps for the project include incorporating more machine learning into the processes. The challenge with machine learning is voluminous training data sets, in this case X-ray images, need to be created and labeled. However, the team has developed new algorithms that can perform crude materials segmentations, which might be used as training data for machine learning.
The team hopes to share their paradigm of lean, agile, scalable HPC at the edge with other groups at Sandia which would benefit the Labs and the nation as a whole.