[P] Built a Interactive Web for PINN Solving the 2D Heat Equation

Hey everyone,

I’ve been working on the idea of taking Scientific AI out of research notebooks and making it accessible as a useful real-time tool. I just finished the first interactive demo, and I’d love some feedback.

I built and trained a 2D thermal simulation engine of two chips on a circuit board using Physics-Informed Neural Networks (PINNs), to solve the 2D heat equation.

Exporting the trained model as ONNX, I build up a simple interactive web app in the browser which allows users to interact with the PINN model by varying the parameters like chip power and ambient temperature to obtain the temperature heatmap and hotspot temperatures.

The Tech Stack:

• AI: Trained a custom PINN in Python using DeepXDE with PyTorch backend
• Deployment: Exported to ONNX for high-performance cross-platform execution.
• Web: Built with Blazor WebAssembly and hosted on Azure. The simulation runs entirely client-side.

Live Demo: https://www.quantyzelabs.com/thermal-inference

I'm currently working on improving the boundary condition flexibility and accuracy for more complex board layouts. I’d love to hear your feedback and where you think this approach has the most potential.

Cheers!