Module overview
This course is designed for students and researchers in academia and industry who are focused on advanced topics in aerospace engineering, particularly in aerodynamic loads and aeroelastic analysis predictions. It also caters to technical decision-makers who seek to understand emerging machine learning and projection-based techniques for future development strategies. The content is tailored to equip students with both foundational knowledge and practical skills, ensuring they can apply modern machine learning and reduced-order model techniques to real-world aerospace challenges and make informed decisions in research and industrial settings.
Aims and Objectives
Learning Outcomes
Knowledge and Understanding
Having successfully completed this module, you will be able to demonstrate knowledge and understanding of:
- The classification and theoretical foundations of ROM techniques, including the distinction between data-driven and equations-derived approaches.
- The role of ROMs in reducing computational complexity for steady-state and transient flows without compromising physical accuracy.
- The operation and applicability of machine learning architectures, such as autoencoders and graph neural networks, for handling large aerospace datasets.
- The principles of projection-based methods, specifically nonlinear projections of the residuals in coupled systems.
- The fundamental aerospace engineering challenges related to aircraft performance and aerodynamic load prediction.
Transferable and Generic Skills
Having successfully completed this module you will be able to:
- Identify and critically evaluate relevant technical literature regarding cutting-edge ROM and ML developments.
- Demonstrate proficiency in managing and analyzing large-scale computational datasets using modern algorithmic tools.
- Produce well-structured technical reports that effectively communicate complex modeling results through clear visualization and data presentation.
- Assess and interpret the results of reduced-order simulations to make substantiated engineering conclusions.
Full CEng Programme Level Learning Outcomes
Having successfully completed this module you will be able to:
- Students will identify and analyse ethical concerns regarding the use of "black-box" machine learning in safety-critical aerospace systems, making reasoned choices informed by professional codes of conduct that prioritize transparency, explainability, and the trustworthiness of data-driven models as discussed by industry subject matter experts.
- Through the assessments, students will select and critically evaluate technical literature and other sources of information on machine learning techniques and their applications to solve complex problems in aerospace engineering.
- Students will formulate and analyse complex aerospace problems by evaluating high-fidelity data through first principles and machine learning algorithms, requiring the use of engineering judgment to reach substantiated conclusions while critically assessing the limitations and uncertainties of reduced-order approximations.
- Through taught classes, practical demonstrations and assessments, students will solve complex aerospace (load and aeroelastic) problems by applying advanced mathematics and machine learning, utilizing forefront reduced-order techniques while critically assessing the integration of emerging data-driven methods within the wider context of aerospace engineering.
- Students will design original AI/ML solutions for complex aerospace challenges, balancing computational efficiency with the ethical requirement for AI trustworthiness and safety-critical reliability as guided by EASA white papers and prevailing industry standards
- Students will select and apply adequate computational techniques to solve specific complex problem, discussing the limitations of the techniques by comparing the computational costs with the accuracy of the predictions.
Subject Specific Intellectual and Research Skills
Having successfully completed this module you will be able to:
- Analyze the performance of coupled systems by reducing their computational burden using nonlinear projection techniques.
- Quantify the efficiency gains and accuracy trade-offs associated with different ROM approaches in the design cycle.
- Implement data-driven and equations-derived models to predict steady and unsteady aerodynamic characteristics.
- Appreciate the strengths and limitations of ML versus NP methods when applied to complex aerospace computational problems.
- Explain the operating principles of both machine learning and projection-based ROMs in the context of aerospace load analysis.
- Demonstrate awareness of emerging trends in modern aerospace modeling, particularly the integration of ML with fluid mechanics fundamentals.
Syllabus
Part I
Introduction
•Providing background context (Theory)
•Aircraft loads analysis (Theory)
•Steady-state aerodynamic models(Theory)
•Unsteady aerodynamic models (Theory)
•Introduction to aeroelasticity (Theory)
•Introduction to reduced order models (Theory)
Part II
Data-driven reduced-order model: machine learning
•Introduction to machine learning(Theory)
•Regression with multiple input variables(Theory)
•Unsteady aerofoil problem(Practical)
•Introduction to Deep Learning: Neural Networks basics (Theory)
•Deep neural networks (Theory)
•Convolutional neural networks (Theory)
•Autoencoders (Theory)
•Recurrent neural networks(Theory)
•Graph neural networks (Theory)
•Aerofoil problem with graph neural networks (Practical)
•Realistic examples (Practical)
•EASA (European Union Aviation Safety Agency) Artificial Intelligence Roadmap (Theory)
Part III
Projection-based reduced-order modelling: equation derived
•Introduction to projection-based methods (Theory)
•Reduced basis for projection (Theory)
•Manipulating a linear dynamical system into a ROM using eigen-mode projection (Practical)
•Interpolating ROM output (Practical)
•Retaining dynamic nonlinearities in the reduced space (Theory)
•Computing nonlinear ROM terms (Practical)
•Large-scale case studies (Theory)
Learning and Teaching
Teaching and learning methods
Core concepts in aerodynamics, aeroelasticity, and Reduced-Order Modeling (ROM) are delivered through interactive lectures, which are immediately reinforced by practical demonstrations. Central to the learning experience are tutorial-based coding sessions, where students are provided with specialized scripts to run real-world aerospace examples. These sessions allow students to experiment with both data-driven machine learning (ML) and equation-derived nonlinear projections (NP), moving beyond passive learning to active implementation.
A distinctive feature of the module is the integration of industry subject-matter experts into the curriculum. These guest contributors provide perspectives on:
•Engineering Ethics: Discussing the moral and professional responsibilities of deploying "black-box" models in safety-critical sectors.
•Certification & Safety: The relevance of AI/ML within the context of EASA frameworks and the technical rigor required for aerospace certification.
•Cross-Sector Relevance: Exploring how these ROM and ML techniques translate to other high-performance fields, such as Motorsport engineering (e.g., F1 aerodynamics) and ship engineering.
| Type | Hours |
|---|---|
| Practical classes and workshops | 5 |
| Lecture | 25 |
| External visits | 2 |
| Assessment tasks | 78 |
| Preparation for scheduled sessions | 40 |
| Total study time | 150 |
Resources & Reading list
General Resources
Module resources. 1.Module presentation slides 2.Sample programming scripts to support practical demonstrations 3.Discussion board
Internet Resources
Textbooks
Geron, A. (2023).. Hands-On Machine Learning with Scikit-Learn, Keras & TensorFlow, 3rd ed. 2023. O’Reilly..
Assessment
Summative
This is how we’ll formally assess what you have learned in this module.
| Method | Percentage contribution |
|---|---|
| Portfolio | 50% |
| Quizzes | 50% |
Repeat Information
Repeat type: Internal & External