Vpeljava obstoječega znanja v podatkovno vodene pristope modeliranja dinamskih sistemov

Anko, Matej (author), Slavič, Janko (mentor)

Pri razumevanju in reševanju dinamskih sistemov si lahko pomagamo z različnimi meritvami in podatkovno vodenimi pristopi, kot so metode strojnega učenja. Pogosto imamo o obravnavanem problemu že nekaj predhodnega znanja, ki ga lahko vključimo v proces strojnega učenja na več različnih načinov. Na primeru balansiranja togih rotorjev, si bomo pogledali več možnih načinov vpeljave predhodnega znanja za napoved lokacije in mase, potrebne za zmanjšanje masne neuravnoteženosti. S pomočjo knjižnice Tensorflow bomo izdelali več modelov in jih ocenili glede na definirane kriterije.

• Language: Slovenian
• Keywords: strojno učenje, podatkovno vodeni pristopi, dinamski sistemi, fizikalno obogateno strojno učenje, balansiranje togih rotorjev, tensorflow
• Work type: Master's thesis/paper
• Typology: 2.09 - Master's Thesis
• Organization: FS - Faculty of Mechanical Engineering
• Year: 2024
• Number of pages: XXII, 84 str.
• PID: 20.500.12556/RUL-160196
• UDC: 531/534:004:004.85(043.2)
• COBISS.SI-ID: 218525443
• Publication date in RUL: 23.08.2024

ANKO, Matej, 2024, Vpeljava obstoječega znanja v podatkovno vodene pristope modeliranja dinamskih sistemov [online]. Master’s thesis. [Accessed 5 April 2025]. Retrieved from: https://repozitorij.uni-lj.si/IzpisGradiva.php?lang=eng&id=160196

Secondary language

• Language: English
• Title: Introduction of existing knowledge in data-driven approaches of modeling dynamical systems

Abstract

We can use different measurements and data-driven approaches, such as machine learning methods in understanding and solving dynamical systems. Often, we already have some prior knowledge about the problem at hand, which we can incorporate into the machine learning process in several different ways. Using the example of balancing of rigid rotors, we will look at several possible ways of introducing prior knowledge to predict the location and mass needed to reduce mass imbalance. With the help of the TensorFlow library, we will create multiple models and evaluate them based on defined criteria.

• Keywords: machine learning, data-driven approaches, dynamical systems, physics informed machine learning, balancing of rigid rotors, tensorflow