Study on the Performance of Machine Learning Algorithms as Surrogate Models for a Representative Model - Master’s thesis

Aug 5, 2026

Upload by Ezra

A master’s thesis in RAMS (Reliability, Availability, Maintainability and Safety) evaluates how machine learning algorithms can serve as surrogate models for a simple bouncing ball representative model. The work reviews modeling, surrogate modeling, and machine learning fundamentals, then designs experiments to train five algorithms: Support Vector Machines, Kernel Ridge, Decision Tree, K-Nearest Neighbors, and Gaussian Process. Hyperparameters and experimental setups are compared, and models are assessed by optimizing the launch angle to maximize the horizontal distance before the first bounce, showing strong accuracy for robust approaches, especially Gaussian Process.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages104

Reading Time4 h 22 min

AccessFree to download

Preface

Acknowledgment

Executive Summary

Modeling and surrogate models in engineering

Thesis objective and research scope

Methodology and experimental design

Algorithms and evaluation task

Results and comparative performance

Master’s thesis NT NU Norwegian University of Science and Technology Faculty of Engineering Department of Mechan ica l and Industrial Engineering José Nicolás Espinoza Guzmán Study on the Performance of Machine Learning Algorithms as Surrogate Models for a Representative Model Master’s thesis in RAMS Supervisor: Shen Yin Co-supervisor: Siegfried Eisinger June 2023 RAMS Reliability, Availability, Maintainability, and Safety José Nicolás Espinoza Guzmán Study on the Performance of Machine Learning Algorithms as Surrogate Models for a Representative Model Master’s thesis in RAMS Supervisor: Shen Yin Co-supervisor: Siegfried Eisinger June 2023 Norwegian University of Science and Technology Faculty of Engineering Department of Mechanical and Industrial Engineering i Preface This thesis is conducted as the final work to achieve the Master’s degree in Reliability, Availability, Maintenance and Safety (RAMS) given by the Department of Mechanical and Industrial Engineering of NTNU in Norway. It represents the culmination of interesting research, analysis, and reflection in pursuit of a deeper understanding of how machine learning can be used as surrogate models in production environments. This thesis journey started with a topic proposed by DNV Group AS with Siegfried Eisinger in contact with the supervisor Shen Yin from the RAMS department at NTNU. The initial step was a literature review and research related ended up in my specialization project called "Study on AI-based surrogate modeling technique and its efficiency in a production environment" delivered in December 2022 . It continued with a definition of the research followed by a review of more academic papers and relevant case studies, intellectual discourse with mentors, and the performance of different tests and experiments with programming. The objective was not to engage in an exhaustive review of every aspect of the subject but rather to identify key elements and present them concisely and coherently. This thesis work aims at readers who have background knowledge or understanding ofsystem modeling and simulations as well as basic knowledge and interest in statistics and machine learning and want to apply it to implement in a real productive environment or anything similar. José Nicolás Espinoza Guzmán Trondheim, June 2023 ii Acknowledgment I would like to take this opportunity to express my deepest gratitude to all those who have contributed to the completion of this master’s thesis. Their support, guidance, and encouragement have been instrumental in my academic journey, and I am profoundly indebted to their contributions. First, I am immensely grateful to my thesis supervisor and co-supervisor, Shen Yin and Siegfried Eisinger, for their constant guidance, expertise, and availability throughout this long journey. Their mentorship has influenced my thinking, challenged my ideas, and pushed me to reach new heights. I am truly fortunate to have had the opportunity to work with such a dedicated and inspiring team. Lastly, I would like to thank my family, friends, and colleagues that have also helped with ideas, support, shared experience, and motivation when needed. These key contributors have, to a certain extent, made the realization of this project possible and as a consequence, contributed to the future development of my career. J.N.E. iii Executive Summary Nowadays, in the engineering areas, modeling and simulation are widely used for productive and research purposes. They represent a very important discipline, especially in the scientific and design fields. The complexity of the models and their simulation has surpassed the physical capacity of computers to perform such requirements in an acceptable time. A possible solution for this problem is the development of simpler models for the already complex models. This answer is called surrogate models or metamodels and represents another abstract layer over reali

FAQs

What problem does the thesis address in engineering modeling and simulation?

The thesis addresses the computational limits of complex models and simulations by using simpler, data-driven surrogate models that approximate model outputs from inputs.

Which machine learning algorithms are used to build surrogate models?

Support Vector Machines, Kernel Ridge, Decision Tree, K-Nearest Neighbors, and Gaussian Process are trained as surrogate models for the representative bouncing ball model.

How are the surrogate models evaluated for performance?

Models are evaluated by solving a task: finding the launch angle that maximizes the horizontal distance traveled by the ball before the first bounce, using comparisons across hyperparameters and setups.

Aug 5, 2026

Upload by Ezra