Material Data Identification in an Induction Hardening Test Rig with Physics-Informed Neural Networks.

Area of Science:

• Materials Science
• Computational Physics
• Machine Learning

Background:

• Physics-Informed Neural Networks (PINNs) integrate governing partial differential equations (PDEs) into the loss function for solving complex problems.
• PINNs have shown success in various forward and inverse problems, demonstrating their versatility.

Purpose of the Study:

• To assess the feasibility of employing PINNs for material property identification in an induction hardening process.
• To estimate thermo-physical properties like specific heat and thermal conductivity using temperature sensor data.

Main Methods:

• Utilized temperature sensor data and the heat equation with specified boundary conditions.
• Employed PINNs to estimate material properties, validating with finite element model (FEM) benchmark data.
• Investigated the impact of sensor placement and measurement noise on parameter uncertainty.

Main Results:

• PINNs accurately identified material data even with limited virtual temperature sensor inputs.
• The approach demonstrated robustness against measurement noise, though convergence time increased.
• Sensor position and noise levels were analyzed for their effect on parameter estimation uncertainty.

Conclusions:

• PINNs offer an accurate and robust method for offline material data estimation in induction hardening.
• The findings have significant implications for optimizing induction heat treatment processes.
• The study validates the use of PINNs with real-world measurement data for material characterization.