Mastering Dynamics PINNs: A Coding Guide

Dive into Physics-Informed Machine Learning this Weekend.

Most ML courses will waste six months of your evenings before you touch anything relevant to engineering. These nine notebooks get you there in a weekend.

You already know the physics. This is just giving you the code to translate that expertise into machine learning.

Physics-Informed Neural Networks let you solve differential equations with a neural network, fuse noisy sensor data with physical laws, and — by the final worksheet — automatically identify unknown equation terms from experimental measurements. That last capability is what structural dynamics teams and research labs spend serious money building. You will have built it yourself.

9 notebooks, one new idea each:

WS0 | Core Concepts: The essential onboarding for anyone new to Physics-Informed Machine Learning. Neural networks, automatic differentiation, and the physics residual explained from first principles.

WS1 | Pure Physics: Solving the nonlinear damped pendulum ODE with zero experimental data. Physics and initial conditions only.

WS2 | Parametric PINNs: One model for an entire family of pendulum lengths. One training run, infinite queries.

WS3 | Hybrid PINNs with Data: Fusing noisy experimental sensor measurements with the governing ODE. Normalisation, learning rate scheduling, and three-component loss functions.

WS4 | Coupled Dynamics: Modelling the 2-DOF chaotic double pendulum in a single two-output network. Two coupled ODEs, simultaneously satisfied.

WS5 | Inverse PINN: Automated discovery of unknown equation terms from sensor data. The cubic stiffness coefficient is identified automatically from 250 noisy displacement measurements using the same optimiser loop as the network weights. No grid search. Just backpropagation.

3 animation tutorials:

AWS1 | Hybrid PINN Animation: Watch your trained pendulum PINN learn in real time — prediction, loss, phase portrait, and a swinging pendulum arm, all updating frame by frame.

AWS2 | Double Pendulum Animation: Full asymmetric dashboard with both pendulum arms animated and a fading chaotic trail behind the lower bob.

AWS3 | Inverse Duffing Animation: Watch the identified parameter alpha converge toward its true value over training, alongside a sliding mass-spring-damper schematic.

What is included: Nine Jupyter notebooks, each in fill-in-the-blank worksheet and complete answers versions. A technical handbook PDF. A Google Colab quick-start guide. Everything runs in Google Colab — no installation, no environment setup, no package conflicts.

What you need: Basic Python. NumPy. Undergraduate dynamics. No ML background assumed.