Differentiable Physics with Keep It Cool

Physics-based simulation is used to provide insight into many practical problems. At Keep it Cool, we use conservation of energy to model heat flow across the delivery cycle of temperature-sensitive packaging. However, uncertainties add considerable fog to the computational modeling process. Questions like “How exactly is this box packed?” or “Is this insulation really performing as well as advertised?” can cause issues. Luckily, a wealth of sensor data has been collected during transit of these temperature-sensitive packages, allowing us to connect the computational physics to real-world data. We can then use machine learning techniques to tune the physics model to match the sensor, and thus the real-world, behavior.