Iskanje po katalogu digitalne knjižnice

Opis: Machine learning (ML) has demonstrated remarkable ability to uncover hidden patterns in data. However, the presence of biases and discrimination originating from the data itself and, consequently, emerging in the ML outcomes, remains a pressing concern. With the exponential growth of unstructured data, such as images, fairness has become increasingly critical, as neural network (NN) models may inadvertently learn and perpetuate societal and historical biases. To address this challenge, we propose a fairness-aware loss function that iteratively prioritizes the worst-performing sensitive group during NN training. This approach aims to balance treatment quality across sensitive groups, achieving fairer image classification outcomes while incurring only a slight compromise in overall performance. Our method, evaluated on the FairFace dataset, demonstrates significant improvements in fairness metrics while maintaining comparable overall quality. These trade-offs highlight that the minor decrease in overall quality is justified by the improvement in fairness of the models.
Ključne besede: fairness, search-basimage classification, machine learning, supervised learnign, neural networks
Objavljeno v DKUM: 23.04.2025; Ogledov: 0; Prenosov: 0
Celotno besedilo (2,01 MB)

Opis: This study develops and evaluates advanced predictive models for the trajectory planning of autonomous vehicles (AVs) in roundabouts, with the aim of significantly contributing to sustainable urban mobility. Starting from the “MRoundabout” speed model, several Artificial Intelligence (AI) and Machine Learning (ML) techniques, including Linear Regression (LR), Random Forest (RF), Support Vector Regression (SVR), Gradient Boosting Regression (GBR), and Neural Networks (NNs), were applied to accurately emulate human driving behavior and optimize AV trajectories. The results indicate that neural networks achieved the best predictive performance, with R2 values of up to 0.88 for speed prediction, 0.98 for acceleration, and 0.94 for differential distance, significantly outperforming traditional models. GBR and SVR provided moderate improvements over LR but encountered difficulties predicting acceleration and distance variables. AI-driven tools, such as ChatGPT-4, facilitated data pre-processing, model tuning, and interpretation, reducing computational time and enhancing workflow efficiency. A key contribution of this research lies in demonstrating the potential of AI-based trajectory planning to enhance AV navigation, fostering smoother, safer, and more sustainable mobility. The proposed approaches contribute to reduced energy consumption, lower emissions, and decreased traffic congestion, effectively addressing challenges related to urban sustainability. Future research will incorporate real traffic interactions to further refine the adaptability and robustness of the model.
Ključne besede: sustainable mobility, autonomous vehicles, machine learning, roundabouts, artificial intelligence, ChatGPT
Objavljeno v DKUM: 04.04.2025; Ogledov: 0; Prenosov: 1
Celotno besedilo (9,00 MB)
Gradivo ima več datotek! Več...

Opis: This study investigates the predictive modelling of weld bead geometry in wire arc additive manufacturing (WAAM) through advanced machine learning methods. While WAAM is valued for its ability to produce large, complex metal parts with high deposition rates, precise control of the weld bead remains a critical challenge due to its influence on mechanical properties and dimensional accuracy. To address this problem, this study utilized machine learning approaches—Ridge regression, Lasso regression and Bayesian ridge regression, Random Forest and XGBoost—to predict the key weld bead characteristics, namely height, width and cross-sectional area. A Design of experiments (DOE) was used to systematically vary the welding current and travelling speed, with 3D weld bead geometries captured by laser scanning. Robust data pre-processing, including outlier detection and feature engineering, improved modelling accuracy. Among the models tested, XGBoost provided the highest prediction accuracy, emphasizing its potential for real-time control of WAAM processes. Overall, this study presents a comprehensive framework for predictive modelling and provides valuable insights for process optimization and the further development of intelligent manufacturing systems.
Ključne besede: wire arc additive manufacturing, WA AM, predictive modelling, machine learning, weld bead geometry, XGBoost
Objavljeno v DKUM: 13.03.2025; Ogledov: 0; Prenosov: 6
Celotno besedilo (3,54 MB)

Opis: Direct verification of the geometric accuracy of machined parts cannot be performed simultaneously with active machining operations, as it usually requires subsequent inspection with measuring devices such as coordinate measuring machines (CMMs) or optical 3D scanners. This sequential approach increases production time and costs. In this study, we propose a novel indirect measurement method that utilizes motor current data from the controller of a Computer Numerical Control (CNC) machine in combination with machine learning algorithms to predict the geometric accuracy of machined parts in real-time. Different machine learning algorithms, such as Random Forest (RF), k-nearest neighbors (k-NN), and Decision Trees (DT), were used for predictive modeling. Feature extraction was performed using Tsfresh and ROCKET, which allowed us to capture the patterns in the motor current data corresponding to the geometric features of the machined parts. Our predictive models were trained and validated on a dataset that included motor current readings and corresponding geometric measurements of a mounting rail later used in an engine block. The results showed that the proposed approach enabled the prediction of three geometric features of the mounting rail with an accuracy (MAPE) below 0.61% during the learning phase and 0.64% during the testing phase. These results suggest that our method could reduce the need for post-machining inspections and measurements, thereby reducing production time and costs while maintaining required quality standards
Ključne besede: smart production machines, data-driven manufacturing, machine learning algorithms, CNC controller data, geometrical accuracy
Objavljeno v DKUM: 10.03.2025; Ogledov: 0; Prenosov: 6
Celotno besedilo (4,44 MB)
Gradivo ima več datotek! Več...

Opis: This thesis focuses on improving the simulation, estimation, and accuracy of parameter identification in lithium-ion battery models. The key objective was to enhance a previously developed program by transitioning it to an object-oriented design, making it more efficient, user-friendly, and modular. Additionally, efforts were made to optimize the parameter estimation process by upgrading the cost function used during simulations and integrating real-world battery measurement data, specifically for the LGM50 battery type. The first step in the thesis involved reworking the codebase to an object-oriented structure, which improved not only the code’s clarity but also its extensibility and efficiency. With this change, the program was better suited for future improvements and became more accessible for other users through simplified installation procedures. This was accompanied by the implementation of unit testing to ensure the reliability of the code. Experiments were conducted across a range of discharge rates (from 0.05C to 1C) to evaluate the performance of the model under different conditions. These tests helped to identify trends in how the model responded to changes in operational parameters. Additionally, a dynamic pulse test was performed, which allowed for more precise estimation of the parameters. The results of these tests demonstrated the robustness of the methodology, especially under dynamic conditions. A major innovation introduced in this thesis was the development of a new cost function, which led to noticeable improvements in parameter estimation accuracy, particularly under high discharge rates and when estimating multiple parameters simultaneously. This new cost function proved especially effective in more complex scenarios, where the original cost function struggled to maintain the same level of accuracy. The program’s capabilities were further extended by incorporating real experimental data. Using a constant discharge profile for the LGM50 battery, the results showed some challenges when dealing with real-world data, particularly due to issues in measurement or data preprocessing. Nonetheless, the model consistently produced solutions, although the accuracy was influenced by the quality of the input data. The thesis concludes by highlighting the success of the improvements made, both in terms of the program’s structure and the precision of its estimations. However, it also emphasizes the importance of improving the quality of real-world data to fully leverage the model’s potential in practical applications. This work lays a foundation for future developments in battery modeling, providing a framework that is adaptable for further research and practical use.
Ključne besede: Machine Learning, Lithium-Ion Batteries, Parameter Estimation, Uncertainty Quantification, Real-experimental data
Objavljeno v DKUM: 03.03.2025; Ogledov: 0; Prenosov: 25
Celotno besedilo (5,99 MB)