Presenters:
Van P. Carey
A. Richard Newton Chair in Engineering
Distinguished Professor, Mechanical Engineering Department
Mehdi Vahab, PhD
Academic Manager, Mechanical & Aerospace Engineering
MathWorks Inc.
Part 1: Fundamentals (Carey) (60 minutes)
- Basic principles of data science and overview of different ML tools: names and what they can do
- The intersection of data science and engineering: Physics based models and data science models of system behavior – can they be complementary perspectives?
- Artificial Neural Networks – how they work, computational tools to implement them—a fundamental ANN example, and how to set it up using python
- Genetic algorithms – how they work, advantages, an example
- Preprocessing data: reducing the order of a data set
- Convolution neural networks, encoder-decoders – image analysis
- PINNs – how they work, how they use custom loss functions
- Overview of other types: recurrent neural networks, reinforcement learning
- Use of machine learning tools for energy component/system design optimization
- Machine-learning-based energy system adaptive control
- Use of machine learning tools to make strategic choices in research experiments or device performance tests
Part 2: MathWorks Tools (Vahab) (60 minutes)
- Overview of MathWorks capabilities in AI/ML
- Tools & workflows with specific use cases in modeling thermofluidic systems
- Reduced-order modeling based on AI/ML methods
- Time-independent: Surface Fitting, ML Networks
- Dynamics: LSTM, Neural State Space, NLARX
- Example: Modeling turbine blade displacement under thermal load
- Nural PDE Solvers (with examples)
- Nural Operators
- Example: Battery Module Cooling Analysis with Fourier Neural Operator
- Graph-Based approaches
- Example: Solve Heat Equation Using Graph Neural Network
20 minute break
Part 3: Example Applications (60 minutes)
- Agentic AI for thermal system design (Vahab)
- Example: Developing a dynamic thermal system model with non-programming tools (Vahab)
- Strategies for framing the modeling and organizing data for use of machine learning in an application, with presentation of specific examples (Carey)
- Machine-learning-enhanced adaptive thermal storage for thermal control (Carey)
- Use of machine learning to enhance heat pipe design for electronics cooling (Carey)
- Machine-learning-enhanced analysis of image and digital data for boiling processes (Carey)
- PINNs models and strategies for loss function design in different applications (Carey)
- Machine learning enhanced computational modeling and simulation (Carey)
Questions, discussion (20 minutes)
Total workshop length: 3 hours, 40 minutes