Dyad SciML Tutorial - Model Discovery with Universal Differential Equations

Learn how to leverage Dyad's model discovery capabilities to automatically identify and complete missing physics in your engineering models using Scientific Machine Learning (SciML) and Universal Differential Equations (UDEs) in this comprehensive 21-minute tutorial. Discover how to create a simple thermal pot model with basic physics and incorporate neural network components into physical models to capture complex dynamics that are difficult to model from first principles alone. Master the process of training universal differential equations against experimental data, using symbolic regression to discover candidate physics equations, and validating trained models against real-world measurements. Follow along as the tutorial demonstrates the complete workflow from initial model creation to physics discovery, starting with setting up the component library and installing required packages, then progressing through creating and understanding model components including heat sources and thermal connections. Learn to run initial analysis and set up the Julia environment before comparing simulation results with experimental data to identify missing physics. Explore the creation of thermal neural network components, understanding neural network parameters, and defining UDE inputs and outputs as you build the neural pot model. Gain hands-on experience with setting up training analysis, training the UDE model, and analyzing training results and convergence plots. Evaluate calibrated simulation results and validate them against experimental data before diving into symbolic regression for physics discovery. Understand UDE analysis results, interpret candidate models, and learn to select physically meaningful models that best represent your system's behavior. Whether you're working on thermal systems, mechanical dynamics, or any domain where some physics may be unknown or too complex to model explicitly, master how UDEs can help you build more accurate, data-informed models that combine traditional physical modeling with machine learning approaches.