Course detail

Data-Driven Modeling and Machine Learning for Industry

FSI-VAI Acad. year: 2027/2028 Winter semester

The course focuses on advanced machine learning methods and their practical applications in industrial data processing, particularly time series and image data. Students will become familiar with methods for predicting the behavior of industrial processes, automatically identifying manufacturing defects, detecting anomalies, and analyzing image data. Particular attention is paid to deep neural networks, including recurrent and convolutional networks, autoencoders, and transformers, and their application to these tasks. The course also covers physics-informed neural networks, which enable the integration of machine learning with knowledge of physical laws. Emphasis is placed on the practical implementation, training, and evaluation of models using tasks inspired by industrial applications.

Language of instruction

Czech

Number of ECTS credits

5

Entry knowledge

Basic knowledge of machine learning principles, including supervised and unsupervised learning, neural networks, and model evaluation methods. Knowledge of the fundamentals of probability, statistics, and linear algebra. Ability to program in Python and use libraries for numerical computing and data processing.

Rules for evaluation and completion of the course

The course is completed by a course credit and an examination.

To obtain the course credit, students must attend compulsory lectures and laboratory classes and successfully complete all assigned tasks. Missed lectures may be made up through independent study of the relevant material. Missed laboratory classes may be made up by attending a replacement session or completing an alternative assignment according to the instructor's instructions.

The examination is oral and assesses students' knowledge and understanding of the topics covered in the lecture syllabus. The final grade is determined by the student's performance in the oral examination in accordance with the grading scale specified in the BUT Study and Examination Regulations.

Aims

The aim of the course is to develop students' ability to independently solve industrial data processing problems using advanced machine learning methods. Students will learn to analyze the nature of a given problem, design and select appropriate models, and implement, train, and optimize them. They will acquire skills in objectively evaluating model performance, interpreting results, and identifying the limitations of the methods used. Emphasis is placed on practical problem-solving, critical evaluation of results, and the ability to apply acquired knowledge to new challenges in industrial practice.

The study programmes with the given course

Programme N-AIŘ-P: Applied Computer Science and Control, Master's, compulsory

Programme N-MAI-P: Mathematical Engineering, Master's, compulsory-optional

Type of course unit

 

Lecture

26 hours, compulsory

Syllabus


  1. Applications of machine learning in industrial data processing. Classical statistical models for time series analysis, time series analysis, and performance evaluation of forecasting models.

  2. Review of feedforward neural networks (MLP), the vanishing gradient problem, deep neural networks, and techniques for stabilizing their training.

  3. Recurrent neural networks (RNN), long short-term memory networks (LSTM), gated recurrent units (GRU), and their applications in industry.

  4. Convolutional neural networks (CNN), temporal convolutional networks (TCN), and their applications in industrial data processing.

  5. Deep convolutional neural networks for image processing and their applications in image classification and object detection.

  6. Basic principles of autoencoders (AE), convolutional and denoising autoencoders, and their applications in industry.

  7. Variational and recurrent autoencoders, autoencoders for time series processing, and their applications in industry.

  8. Anomaly detection, reconstruction-based methods, one-class classification, and performance evaluation of anomaly detection models.

  9. Image segmentation using deep neural networks and its industrial applications.

  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 hours, compulsory

Syllabus


  1. Time-series forecasting. Analysis of time series, implementation of forecasting models, and evaluation of predictive performance.

  2. Multilayer perceptrons. Implementation and training of MLPs, and investigation of the effects of network architecture, activation functions, and weight initialization.

  3. Recurrent neural networks. Implementation and evaluation of RNN, LSTM, and GRU models for time-series prediction.

  4. Convolutional networks for time-series processing. Implementation and comparison of temporal convolutional networks for industrial signal processing.

  5. Image classification and object detection. Implementation, training, and evaluation of deep learning models for visual inspection.

  6. Implementation and comparison of reconstruction-based and denoising autoencoders for industrial data.

  7. Recurrent and variational autoencoders. Implementation and evaluation of autoencoder architectures for time-series processing.

  8. Anomaly detection. Implementation and comparison of anomaly detection methods and evaluation of their performance.

  9. Image segmentation. Implementation and evaluation of deep learning models for pixel-level image segmentation.

  10. Basic implementation of Transformer models for time-series and image processing.

  11. Physics-informed neural networks. Implementation of physics-informed neural networks (PINNs).

  12. Applications of PINNs. Practical tasks involving the application of physics-informed neural networks.

  13. Final assessment.