Course detail

Data-Driven Modeling and Machine Learning for Industry

FSI-VAIAcad. year: 2026/2027

The course provides an overview of modern machine learning models for the analysis of sensor data, including image data, the detection of anomalies in such data, and the use of physics-informed neural networks for modeling industrial processes.

Language of instruction

Czech

Number of ECTS credits

5

Mode of study

Not applicable.

Guarantor

doc. Ing. Pavel Škrabánek, Ph.D.

Department

Institute of Automation and Computer Science (ÚAI)

Entry knowledge

A basic knowledge of machine learning, optimization, and programming is assumed.

Rules for evaluation and completion of the course

Knowledge and skills are verified by credit and examination. Credit requirements: elaboration of given tasks. Attendance at lectures is recommended, while attendance at practical sessions is mandatory. Practical sessions that a student is unable to attend in the regular term can be made up during a substitute term. The exam is oral and covers the entire course material.

Aims

The objective of the course is to familiarize students with advanced machine learning methods used in the processing of industrial data. Students will learn modern models for the analysis of time series, industrial signals, and image data, including recurrent and convolutional neural networks, autoencoders, transformers, and physics-informed neural networks. Emphasis is placed on understanding the principles of deep learning, reconstruction-based methods, anomaly detection, and modeling of complex processes. The course connects theory with practical applications for designing, training, and evaluating machine learning models in industrial practice.

Study aids

Not applicable.

Prerequisites and corequisites

Not applicable.

Basic literature

BATY, Hubert. A hands-on introduction to Physics-Informed Neural Networks for solving partial differential equations with benchmark tests taken from astrophysics and plasma physics. 2024. hal-04491808 (EN)
KUO, Chris. Modern Time Series Forecasting Techniques For Predictive Analytics and Anomaly Detection: From Classical Foundations to Cutting-Edge Applications. Innovation Press, 2024. ISBN 979-8990781009. (EN)
BOX, George E. P.; JENKINS, Gwilym M.; REINSEL, Gregory C. a LJUNG, Greta M. Time series analysis: forecasting and control. Fifth edition. Wiley series in probability and mathematical statistics. Hoboken: Wiley, [2016]. ISBN 978-1-118-67502-1. (EN)

Recommended reading

Not applicable.

Classification of course in study plans

  • Programme N-MAI-P Master's 1 year of study, winter semester, compulsory-optional
  • Programme N-AIŘ-P Master's 2 year of study, winter semester, compulsory

Type of course unit

 

Lecture

26 hod., optionally

Teacher / Lecturer

doc. Ing. Pavel Škrabánek, Ph.D.

Syllabus

  1. Applications of machine learning in industrial data processing. Classical statistical models for time series analysis and performance evaluation of time series models.
  2. Review of feedforward neural networks (MLP), the vanishing gradient problem, and deep neural networks.
  3. Recurrent neural networks (RNN), long short-term memory (LSTM), gated recurrent units (GRU), and their applications in industry.
  4. Convolutional neural networks (CNN), temporal convolutional networks (TCN), and their applications in industry.
  5. Deep convolutional neural networks for image processing and their applications in image classification and object detection.
  6. Basic idea of autoencoders (AE), convolutional autoencoders, denoising autoencoders, and their industrial applications.
  7. Variational autoencoders (VAE), autoencoders for time series, and their industrial applications.
  8. Image segmentation.
  9. Anomaly detection, performance evaluation of anomaly detection models, and one-class classification.
  10. Transformers for time series and images.
  11. Integration of physics and machine learning, physics-informed neural networks (PINNs): principles, loss functions, collocation points, and boundary conditions.
  12. Industrial applications of PINNs.
  13. Review

Computer-assisted exercise

26 hod., compulsory

Teacher / Lecturer

doc. Ing. Pavel Škrabánek, Ph.D.

Syllabus

  1. Working with time series, basic statistical models, evaluation of time series prediction performance.
  2. Demonstration of the vanishing gradient problem, training deep neural networks.
  3. Implementation of RNN, LSTM, and GRU for time series prediction.
  4. Implementation of 1D CNN, application of TCN to time series.
  5. Training a deep CNN for image classification and demonstration of object detection.
  6. Implementation of a basic autoencoder, convolutional autoencoder, and denoising autoencoder.
  7. Implementation of a variational autoencoder and a time series autoencoder.
  8. Implementation of image segmentation using U-Net.
  9. Anomaly detection using reconstruction, one-class classification.
  10. Basic implementation of Transformers for time series and images.
  11. Implementation of PINNs.
  12. Practical tasks with PINNs.
  13. Final assessment.