Inverse Problem Solving with Machine Learning for Predicting Water Droplet Impact Conditions from Erosion Rates Across Different Rigs in Aerospace Materials

Inverse problem-solving using machine learning offers a
promising approach to predicting water droplet impact conditions—specifically
droplet velocity and size—based on the observed maximum erosion rate of
aerospace metals like Ti-6Al-4V, Al7075, Al2024, PH17-4 SS, and 12% Cr SS.
These materials are tested across various experimental setups, such as rotating
disc rigs and arm-type rigs, which introduce challenges due to differing
mechanisms and conditions. For example, the same erosion rate value measured on
a rotating disc may correspond to entirely different impact conditions
(velocity and droplet size) compared to an arm-type rig, due to variations in
force distribution, contact time, and droplet behavior. Therefore, to ensure
consistent predictions, the erosion rate data must first be standardized across
rigs by normalizing it to a common reference frame. After standardization,
machine learning models can be employed to solve the inverse problem, where the
inputs (impact velocity and droplet size) are inferred from the erosion rate.
Techniques like neural networks or regression models are well-suited for this
task, enabling the system to account for the complex nonlinear relationships
between erosion rates and impact conditions, thereby providing reliable
predictions across diverse experimental setups.