Fundamentals of AI for Simulation Engineers

30 - 31 October 2024, Online

ATTENTION:
T​his course is fully booked! We have setup another course from 5 - 6 November.
Please book for the new dates here

h​ttps://www.nafems.org/ai24-3

Language: English

This intensive two-day training course is designed to equip simulation engineers with a good understanding and practical skills in applying Artificial Intelligence (AI) in their field. It is designed to be software-agnostic, prioritizing methodologies, and techniques that engineers can apply across various computational platforms. The course offers a balanced mix of theoretical knowledge and hands-on applications, ensuring participants gain a robust foundation in AI and its relevance to simulation engineering.

Detailed Course Program:

Day 1

• Introductory example
  - Problem description: Modeling compressor parameters
  -​ Traditional approach vs. Artificial Intelligence
  - Explanation: Why AI outperformed human intelligence

• Overview of AI modeling techniques
  -Explaining the difference between Artificial Intelligence, Machine Learning, and Deep Learning
  - The basic principles behind all AI systems
  - Categories of machine learning
  - Supervised learning
  - Unsupervised learning
  - Semi-supervised learning
  - Reinforcement learning
  - Types and examples of different supervised Machine Learning Algorithms
  - Linear & logistic regression
  - Decision Tree & Random Forest
  - Gradient Boost Algorithms
  - Support Vector Machines
  - Neural Networks & Deep Learning
  - The different input types of machine learning algorithms
  - Numerical data
  - Classes and vectorized data
  - Images
  - Text
  - Different applications of machine learning algorithms
  - Surrogate modeling
  - Classification
  - Generative AI
• Introduction to training deep neural networks
  - Neurons and activation functions
  - Forward pass and backpropagation
  - Layers and layer types
  - The influence of topology on neural networks
  - Hyperparameters and hyperparametertuning
  - Loss functions
  - Optimization algorithms to train model parameters
  - In-depth explanation: Gradient Descent
  - Comparison of common algorithms
  - Deciding which algorithm to use

• Physics-Informed Neural Networks (PINN)
  - How PINNs work
  - Advantages
  - Performance comparison
  - Pitfalls

• Example application of a surrogate model

Day 2

• Example project for a deep learning surrogate model for design optimization
• Creating machine learning models from scratch
  - A high-level overview of creating machine learning models
  - Reviewing available data and setting a goal
  - Data preparation
  - The importance of the training, test, validation split
  - Setting the model architecture
  - Choosing optimization algorithms and loss functions
  - Creating PINNs
  - Evaluating model performance
  - Overview of MLOps
  - An overview of tools to create machine learning models
  - Open-source libraries (TensorFlow vs. PyTorch)
  - Application software
• Project preparation
  - Reviewing available data
  - Using a preliminary exploratory data analysis to gauge the feasibility of the ML project
  - Defining a modeling target
  - Working in tandem with a simulation project

• Data preparation
  - Data transformation: File formats and making data trainable
  - Data cleaning
  - Handling classes and text with vectorization
  - Dimension reduction
  - Feature selection
  - Feature Engineering

• Sampling
  - Introduction to data sampling
  - Statistical sampling methods
  - Active sampling
  - Sampling errors

• Measuring model performance and validity
  - Performance measures for ML models

• Consuming the machine learning model
  - Predictive Tasks
  - Design optimization

• Limitations of machine learning models
  - Model biases
  - Limitations of interpolation and extrapolation
  - Model aging