Machine Learning for aerospace engineers
Motivation, presentation and applications
This course aims to introduce the basics of machine learning, to put them into practice with the modern tools of data science (Python and R) and to show the benefit of these techniques for the aeronautical engineer.
The course lasts 2 days (12 hours) and includes 9 hours of theory and 3 hours of PC-based application, so learners can put into practice the methods taught.
Course in French, resources in English.
The deluge of data from the Web, the advent of generic computing on GPU and the incessant innovations presented by the big players of the Web led the majority of traditional industries to engage a so-called digital revolution. The aerospace sector is not excluded from this major trend. Nevertheless, to accompany this transformation towards data and machine learning, companies must train and recruit people trained in these areas so as to anticipate the benefits of these techniques but also to moderate the expectations that any industrial revolution generates. There are numerous new services that machine learning can create in the aeronautical sector: the predictive maintenance of aeronautical structures, the recommendation of contents for passengers, the automatic detection of manufacturing defects, the regression strategies of expensive digital simulation or the real-time detection of engine anomalies.
The purpose of this course is to present the basics of machine learning, its main techniques, its conventional and recent applications and potential ideas for applications in the aeronautics sector. For many years, ONERA has been offering innovative machine learning applications in numerous aeronautical fields (pre-project design, digital simulation, image recognition, anomaly detection, etc.).
Course level: Basic/Advanced
Engineering training or equivalent required; basic statistical knowledge is reviewed at the beginning of the course. Knowledge of Python is not essential but would be a plus.
Introduction and motivation
- What is machine learning?
- Why does machine learning work? Examples of conventional and recent applications
- Potential applications in the aeronautical world
Overview of machine learning
- Supervised vs. Unsupervised
- Overview of machine learning algorithms
- Practical construction (cost function, selection of hyperparameters, optimization algorithm, regularization)
- Data preparation and visualization
- Techniques for reducing variances
- Digital experience plans and sensitivity analysis
- Focus on the unsupervised
- Linear models
- Classic linear regression, levers
- Orthogonal linear regression
- Modern regulation techniques, lasso and ridge regression
- Logistic regression
- Widespread linear models
- Neural Networks and Deep Learning
- Origin and link with artificial intelligence
- Focus on multilayer perceptron: practical construction of an MLP and Deep Learning
- Kriging (Gaussian Process)
- Geostatistical origin, link with Gaussian processes
- Description of the different Krigings
- Multi-fidelity data
- Global optimization strategy
- Non-parametric methods
- Naive Bayes, Decision Tree, Random Forest
- Boosting, Adaboost
- Support vector machines
- General principles
- Examples of application in classification
- Put Python (or R) into practice
- Overview of the main Python modules (sklearn, keras, xgboost, pandas, seaborn)
- Put into practice on an aeronautical use-box
Scheduled in French:
PARIS: 4 to 5 February 2019
For the English realization, please, consult us.
€1,030 excluding tax (20% VAT)