Physics-Informed Neural Networks (PINNs) — Mini Projects
This repository contains a collection of small, educational PINN implementations designed to demonstrate how neural networks can solve physics-governed systems by enforcing differential equations directly in the loss function. Each project uses TensorFlow, automatic differentiation, and carefully constructed PDE/IC/BC losses.
🔬 Included Projects
1D Motion (ODE) — Learning dynamics from Newton's laws using initial conditions.
Newton's Law of Cooling — Modeling temperature decay governed by a first-order ODE.
Poisson Equation (1D) — Solving a classical boundary value PDE using PINNs.
Projectile Motion — Predicting ball trajectory by embedding physics constraints.
⚙️ Core Techniques
Physics-Informed loss formulation (PDE + IC/BC)
Automatic differentiation for derivatives
Neural network approximation of continuous functions
Comparison with analytical solutions for validation
🎯 Purpose
These mini-projects serve as a compact introduction to Scientific Machine Learning, PINNs, and integrating physics constraints into deep learning models. Ideal for students and beginners exploring PINNs for the first time.