Predicting thermal runaway via noninvasive ultrasound transmission

Thermal runaway occurs in materials such as batteries, bulk foods, and high explosives (HE) when self-heating leads to uncontrollable rise in temperature, typically resulting in catastrophic failure and posing a significant safety risk. Currently, there is no method for predicting thermal runaway of HE in closed containers without puncturing the containers, which poses additional safety risks. We present a novel, noninvasive Ultrasound Thermal Runaway Early Detection (U-TRED) technique based on ultrasound transmission measurements and a transformer neural network. To increase real-world applicability of the method, we design the transformer to be agnostic to material properties, container geometry, and transducer placement. To demonstrate the technique, we numerically simulate ultrasound transmission measurements during thermal runaway in containers with randomized geometric and material properties. We demonstrate U-TRED on experimental measurements conducted in HE that was heated until ignition and find that it is able to alert operators approximately 15 min prior to thermal runaway, which provides critical time to evacuate prior to ignition. Additionally, due to the material—and geometry—agnostic nature of U-TRED, it could be expanded to other applications, providing life-saving advanced warning ahead of catastrophic events like battery fires, uncontrolled reactions during chemical synthesis, and grain silo combustion.