Application of Quantum-Assisted PINN for Predicting Ice Accretion and Phase Change

Accurate solidification phase change prediction is critical in various engineering and environmental contexts, such as icing on aircrafts and wind turbines, additive manufacturing, and more. Recent advancements in phase-field modeling, like accurate conservative diffuse-interface (ACDI) method, relies on the Gibbs-Thompson (GT) coefficient, which is not trivially obtained because experiments are limited, and molecular dynamics are computationally expensive.

Determining the coefficient from real-world observed/numerical data through a Physics-Informed Neural Networks (PINN) formulated as an inverse problem with the ACDI as the governing equation is promising, but its efficacy is challenged by generalizability and training efficiency. A novel approach improving these two factors by utilizing a Quantum-Assisted PINN (QA-PINN), while maintaining accuracy, is proposed. By coupling the ACDI model with the QA-Inverse PINN, we incorporate available data to estimate the GT coefficient while enforcing the governing physics. The derived coefficient is introduced into the ACDI equation, which is solved using a QA-Forward PINN. This approach leverages the strengths of both data-driven quantum learning and physically consistent modeling, enabling accurate prediction of ice nucleation with reduced computational cost. Results obtained from QA-PINN, PINN, and numerical methods will be compared. The QA-PINN’s transfer learning potential to other ice nucleation scenarios will also be explored.