Schedule (2025):
Wedensday: 11:00 - 14:00 Room A2 - DIAG, Via Ariosto 25
Thursday: 13:00 - 16:00 Room A7 - DIAG, Via Ariosto 25
The course is organized into two modules: the first, taught by me, deals with the design of the so-called intelligent control systems, whereas the second will cover hybrid systems and will be taught by prof. Luca Benvenuti.
Intelligent Control Systems are a class of controllers based on data-driven techniques such as deep learning and reinforcement learning.
The course will present both the role of machine learning in control systems, discussing its methodologies, potential and applications, and the design of control systems based on artificial intelligence techniques.
The course will also focus on how recent developments in deep learning can be exploited by standard control systems, discussing the basics of advanced data analysis using deep neural networks in the context of classical automation problems such as quality control, predictive maintenance and disturbance rejection.
In particular, great focus will be given to application examples from the domains of industrial automation, robotics and cyber security where intelligent control solutions may contribute positively to the robustness and the economic and environmental sustainability of the controlled system.
Selected topics include:
Introduction to Machine Learning problems and the role of AI in Control Systems: direct neural control schemes, indirect neural control schemes and neural enhanced control systems
Introduction to fundamental deep learning solutions (DNN, CNN, RNN) and their applications in the automation domain; Introduction to data science tools in data-driven control systems: SVD, PCA and their role for modelling and controlling dynamical systems
Data driven control: Dynamic Mode Decomposition (DMD), DMD-based model predictive control
Neural Lyapunov Control: learning stability certificates and direct neural control laws
Advanced Predictive Control through deep learning
Reference Book(s):
Brunton, S. L., & Kutz, J. N. (2022). Data-driven science and engineering: Machine learning, dynamical systems, and control. Cambridge University Press.
Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep learning. MIT press.
Additional material (papers of interest and complementary readings) will be provided during classes and on the course classroom.
Course Syllabus (will be updated during the classes):
The concepts behind data-driven engineering: information, data, knowledge, experience and training.
Single Value Decomposition and Principal Component Analysis as a tool for dimensionality reduction in dynamical system analysis.
Dynamic Mode Decomposition, Image-MPC and DMD-MPC
Fundamentals of Machine Learning: Regression and Classification tasks, generalization, overfit and underfit. Examples of ML sub-tasks and functionalities in control problems.
Deep Neural Networks for Control: Architectures and applications.
CNN for Chaotic system analysis through recurrence plots
Neural Lyapunov Control, Neural Stability Certificates
Applications of the proposed methods to the factory 4.0, satellite monitoring, e-health and robotics domains.
Lecture Slides and Lesson Diary (available on google classroom, code: b7mthlp) - to gain access to the classroom of the current year either use the code given during classes and present in this page or just send me an email at any time from your student account
Exam modalities:
Being a two-module class, the students have the ability to chose a single one of the two modules to do an individual project on a topic of interest for the student, whereas for the other module the student will be required to complete a simpler test (written or oral) accounting for about one fourth of the overall grade.
If you dedice to do the project on intelligent control systems, the following applies:
Projects will be first presented during the final lecture of the module, but the student can request to work on a project at any time of the year. On google classroom you will find a list of all available projects and the students will be asked to select 3-4 projects of interest, so that I will be able to balance project assignements for the entire class.
There is no pre-requisite regarding coding, so the projects may or may not involve the implementation of some software simulations.
To complete the exam, the students are required to send me an email to agree on a date on which they will be required to do a short presentation (about 20 minutes) on the project and answer some technical questions.
In order to have your mark registered please note that you are required to register for the exam on infostud for the next upcoming date. Please note that there is no other procedure to officially register the exam, so this is of crucial importance if you are subject to any strict deadline (graduation, scholarships, etc...).
Office hours: Please send me an email to book an appointment on google meet or at DIAG (room A225).
FAQ
I can't join the google classroom as there is no code on this page, what can I do?
To avoid confusion among the various accademic years, I remove from the website the invitation code that I show during classes after a while. To join the current google classroom send me an email at any time of the year (do not worry about which time of the year it is). Please do this from your student email account, as classroom has some restrictions.
Is attendance mandatory to have a project assigned?
Not for now. Students may require a project assignment at any time of the year (after the end of the module's lectures, typically around the 15-th of April) independently from their attendance, however, to be able to ensure a sufficient number of project proposals please try to attend one of the first 4-5 classes and sign the attendance list. Please note that this list will be used only to estimate the number of students enrolled in the course and dimension the project list accordingly. I you are unable to do that either send me an email or join the google classroom directly. I will commit to make sure that no student will be prevented from having a project assigned, so if for any reason you did not join in time just contact me directly.
I enrolled for the classes too late, how can I study for this module?
Albeit attending the classes would be the better way to understand the topics needed to complete the projects, the material and references provided will allow you to study individually. Join the google classroom to get access to all of the course material.
Is a project mandatory?
No, you can do the project for the hybrid control systems part of the exam, however, note that some of my projects will be more theory-oriented, so coding and implementations will not always be required.
I have an idea for a prossible project on a topic of my interest (e.g., due to my bachelor background), can I propose it?
I am always interested in new ideas, so if there is commintment on your end, we can discuss during my office hours how to build a project on a given application beyond the ones discussed during classes.
How long do I have to complete a project?
There is no hard deadline, but if for any reason you do not complete the exam after 12 month from the assignement I kindly ask you to re-check with me if the project is still of interest for the course objectives or if it needs some updates.
I chosed to do the project in the hybdrid control systems module, what is the test on intelligent systems going to be?
Either a short oral discussion on some key topics of the module, a multiple-choice written test or (if requested and available) a homework involving some coding a selected problem.
There is a problem with an exam date on infostud (e.g., it prevents me from graduating), what can I do?
Given enough notice, I will try to figure out a solution with the administration. Just contact me directly explaining the situation.
I did not register for a past exam session but I really need the exam to be registered, what can I do?
I cannot register students on infostud, the only option is to book an upcoming exam date.