1. Comparative analysis of nonlinear models developed using machine learning algorithmsMaja Rožman, Alen Kišić, Dijana Oreški, 2024, izvirni znanstveni članek Opis: Machine learning algorithms are increasingly used in a vast spectrum of domains where statistical approaches were previously used. Algorithms such as artificial neural networks, classification, regression trees, or support vector machines provide various advantages over traditional linear regression or discriminant analysis. Advantages such as flexibility, scalability, and improved accuracy in dealing with diverse data types, nonlinear problems, and dimensionality reduction, compared to traditional statistical methods are empirically demonstrated in many previous research papers. In this paper, two machine learning algorithms are compared with one statistical method on highly nonlinear data. Results indicate a high level of effectiveness for machine learning algorithms when dealing with nonlinearity.
Ključne besede: machine learning, decision tree algorithm, artificial neural network, predictive models, data characteristics, nonlinear data, artificial intelligence
Objavljeno v DKUM: 02.07.2025; Ogledov: 0; Prenosov: 3
 Celotno besedilo (730,52 KB)
Gradivo ima več datotek! Več... |
2. Integrating Multi-Physics Modeling within Multi-Objective Optimization to Enhance the Performance and Efficiency of Permanent Magnet Synchronous Machines : doktorska disertacijaMitja Garmut, 2025, doktorska disertacija Opis: This Dissertation focuses on the optimization of an Interior Permanent Magnet (IPM) machine for handheld battery-powered tools, aiming to enhance performance and efficiency.
The research integrates multi-physics modeling, including electromagnetic Finite Element Method (FEM) and thermal models, to evaluate machine performance under various operating conditions. The performance is evaluated according to selected Key Performance Indicators (KPIs). Further, different control methods, such as Field Oriented Control and Square-Wave Control, impact the performance significantly and are incorporated into the optimization process.
Due to the computational challenges of FEM-based performance evaluations in Multi-Objective Optimization (MOO), this work utilizes Artificial Neural Network (ANN)-based meta-models, to accelerate the optimization process while preserving accuracy.
The developed meta-models capture nonlinear machine characteristics from the FEM model. These meta-models are then used to evaluate machine performance through a combination of analytical and numerical post-processing methods.
Four MOO scenarios are presented, each aimed at optimizing the cross-sectional design of IPM machines, to enhance performance and efficiency while reducing mass and cost. Additionally, these scenarios modify the machine’s electromagnetic behavior, to ensure better alignment with the selected control method.
By comparing the optimization process of Scenario 1, which uses direct FEM-based evaluation without time reduction measures, to the approach incorporating Artificial Neural Network based meta-models, the total number of individual FEM evaluations decreased from 2.35×10^9 to 2.03×10^5, without almost any loss of accuracy. This reduced the computation time from 297 years to 9.07 days on our standard desktop computer. The obtained ANN-base meta-models can be used further for other optimizations without the need for additional FEM evaluations.
In all four optimization scenarios, the use of meta-models enabled the generation of a Pareto front of the optimal solutions, leading to improved KPIs compared to the reference design. The highest relative improvement occurred in Scenario 1, where the selected optimized machine design achieved a 30% increase in power density compared to the reference design.
Ključne besede: Interior Permanent Magnet (IPM) Machine, Artificial Neural Network (ANN), Meta-Modeling, Multi-Objective Optimization (MOO), Finite Element Method (FEM), Multi-Physics Modeling, Field Oriented Control (FOC), Square-Wave Control (SWC)
Objavljeno v DKUM: 15.05.2025; Ogledov: 0; Prenosov: 66
Celotno besedilo (17,79 MB) |
3. Robot for navigation in maize crops for the Field Robot Event 2023David Iván Sánchez-Chávez, Noé Velázquez-López, Guillermo García-Sánchez, Alan Hernández-Mercado, Omar Alexis Avendaño-Lopez, Mónica Elizabeth Berrocal-Aguilar, 2024, izvirni znanstveni članek Opis: Navigation in a maize crop is a crucial task for the development of autonomous robots in agriculture, with numerous applications such as spraying, monitoring plant growth and health, and detecting weeds and pests. The Field Robot Event 2023 (FRE) continued to challenge universities and other research teams to push the development of algorithms for agricultural robots further. The Universidad Autónoma Chapingo has been developing a robot for various agricultural tasks, aiming to provide a low-cost alternative to work with Mexican farmers in the future. For this edition of the FRE, a navigation algorithm was created using an encoder, an IMU (Inertial Measurement Unit), an RPLIDAR (Rotating Platform Light Detection and Ranging), and cameras to collect data for decision-making. The algorithm was developed in ROS Melodic, dividing the task into steps that were tested to determine the robot's actual movements. The system navigates by using ROIs (regions of interest) and the mass center to guide the robot between maize rows. It calculates the mean of the final orientation values before reaching the end of a row, which is detected using an RPLIDAR. For turns and straight-line movements to reach the next row, the orientation is used as a guide. To detect plants for spraying, lasers located on each side of the vehicle are employed. Obstacle detection relies on a YOLOv5 (You Only Look Once) trained model and a laser, while reverse navigation uses a rear camera. During the competition, the robot faced challenges such as dealing with grass, the small size of the plants, and the need to use a different power source, which affected its performance.
Ključne besede: machine vision, convolutional neural network (CNN), regions of interest (ROI), autonomous navigation
Objavljeno v DKUM: 23.04.2025; Ogledov: 0; Prenosov: 1
 Povezava na datoteko |
4. Using a region-based convolutional neural network (R-CNN) for potato segmentation in a sorting processJaka Verk, Jernej Hernavs, Simon Klančnik, 2025, izvirni znanstveni članek Opis: This study focuses on the segmentation part in the development of a potato-sorting system that utilizes camera input for the segmentation and classification of potatoes. The key challenge addressed is the need for efficient segmentation to allow the sorter to handle a higher volume of potatoes simultaneously. To achieve this, the study employs a region-based convolutional neural network (R-CNN) approach for the segmentation task, while trying to achieve more precise segmentation than with classic CNN-based object detectors. Specifically, Mask R-CNN is implemented and evaluated based on its performance with different parameters in order to achieve the best segmentation results. The implementation and methodologies used are thoroughly detailed in this work. The findings reveal that Mask R-CNN models can be utilized in the production process of potato sorting and can improve the process.
Ključne besede: image segmentation, potato sorting, neural network, mask RCNN, object detection, production process, machine learning, AI
Objavljeno v DKUM: 27.03.2025; Ogledov: 0; Prenosov: 14
Celotno besedilo (5,97 MB)
Gradivo ima več datotek! Več... |
5. Recent applications of explainable AI (XAI) : a systematic literature reviewMirka Saarela, Vili Podgorelec, 2024, pregledni znanstveni članek Ključne besede: explainable artificial intelligence, applications, interpretable machine learning, convolutional neural network, deep learning, post-hoc explanations, model-agnostic explanations
Objavljeno v DKUM: 31.01.2025; Ogledov: 0; Prenosov: 4
Celotno besedilo (1,42 MB) |
6. Predicting the probability of cargo theft for individual cases in railway transportLorenc Augustyn, Małgorzata Kuźnar, Tone Lerher, Maciej Szkoda, 2020, izvirni znanstveni članek Opis: In the heavy industry, the value of cargo transported by rail is very high. Due to high value, poor security and volume of rail transport, the theft cases are often. The main problem of securing rail transport is predicting the location of a high probability of risk. Because of this,the aim of the presented research was to predict the highest probability of rail cargo theft for areas. It is important to prevent theft cases by better securing the railway lines. To solve that problem the authors' model was developed. The model uses information about past transport cases for the learning process of Artificial Neural Networks (ANN) and Machine Learning (ML).The ANN predicted the probability for 94.7% of the cases of theft and the Machine Learning identified 100% of the cases. This method can be used to develop a support system for securing the rail infrastructure.
Ključne besede: rail transport security, supply chain disruption, drones, security support systems, cargo theft, predicting, logistics, artificial neural network, drone monitoring, machine learning
Objavljeno v DKUM: 28.01.2025; Ogledov: 0; Prenosov: 4
Celotno besedilo (1,93 MB)
Gradivo ima več datotek! Več... |
7. Optimizing laser cutting of stainless steel using latin hypercube sampling and neural networksKristijan Šket, David Potočnik, Lucijano Berus, Jernej Hernavs, Mirko Ficko, 2025, izvirni znanstveni članek Opis: Optimizing cutting parameters in fiber laser cutting of austenitic stainless steel is challenging due to the complex interplay of multiple variables and quality metrics. To solve this problem, Latin hypercube sampling was used to ensure a comprehensive and efficient exploration of the parameter space with a smaller number of trials (185), coupled with feedforward neural networks for predictive modeling. The networks were trained with a leave-oneout cross-validation strategy to mitigate overfitting. Different configurations of hidden layers, neurons, and training functions were used. The approach was focused on minimizing dross and roughness on both the top and bottom areas of the cut surfaces. During the testing phase, an average MSE of 0.063 and an average MAPE of 4.68% were achieved by the models. Additionally, an experimental test was performed on the best parameter settings predicted by the models. Initial modelling was conducted for each quality metric individually, resulting in an average percentage difference of 1.37% between predicted and actual results. Grid search was also per formed to determine an optimal input parameter set for all outputs, with predictions achieving an average ac curacy of 98.34%. Experimental validation confirmed the accuracy and robustness of the model predictions, demonstrating the effectiveness of the methodology in optimizing multiple parameters of complex laser cutting processes.
Ključne besede: laser cutting optimization, cut surface quality, dross formation, Latin hypercube sampling, feedforward neural network
Objavljeno v DKUM: 10.01.2025; Ogledov: 0; Prenosov: 26
Celotno besedilo (3,38 MB)
Gradivo ima več datotek! Več... |
8. Interlayer connectivity affects the coherence resonance and population activity patterns in two-layered networks of excitatory and inhibitory neuronsDavid Ristič, Marko Gosak, 2022, izvirni znanstveni članek Opis: The firing patterns of neuronal populations often exhibit emergent collective oscillations, which can display substantial regularity even though the dynamics of individual elements is very stochastic. One of the many phenomena that is often studied in this context is coherence resonance, where additional noise leads to improved regularity of spiking activity in neurons. In this work, we investigate how the coherence resonance phenomenon manifests itself in populations of excitatory and inhibitory neurons. In our simulations, we use the coupled FitzHugh-Nagumo oscillators in the excitable regime and in the presence of neuronal noise. Formally, our model is based on the concept of a two-layered network, where one layer contains inhibitory neurons, the other excitatory neurons, and the interlayer connections represent heterotypic interactions. The neuronal activity is simulated in realistic coupling schemes in which neurons within each layer are connected with undirected connections, whereas neurons of different types are connected with directed interlayer connections. In this setting, we investigate how different neurophysiological determinants affect the coherence resonance. Specifically, we focus on the proportion of inhibitory neurons, the proportion of excitatory interlayer axons, and the architecture of interlayer connections between inhibitory and excitatory neurons. Our results reveal that the regularity of simulated neural activity can be increased by a stronger damping of the excitatory layer. This can be accomplished with a higher proportion of inhibitory neurons, a higher fraction of inhibitory interlayer axons, a stronger coupling between inhibitory axons, or by a heterogeneous configuration of interlayer connections. Our approach of modeling multilayered neuronal networks in combination with stochastic dynamics offers a novel perspective on how the neural architecture can affect neural information processing and provide possible applications in designing networks of artificial neural circuits to optimize their function via noise-induced phenomena.
Ključne besede: neuronal dynamics, coherence resonance, excitatory neurons, inhibitory neurons, neural network, multilayer network, interlayer connectivity
Objavljeno v DKUM: 20.12.2024; Ogledov: 0; Prenosov: 4
Celotno besedilo (6,72 MB)
Gradivo ima več datotek! Več... |
9. Learning physical properties of liquid crystals with deep convolutional neural networksHigor Y. D. Sigaki, Ervin K. Lenzi, Rafael S. Zola, Matjaž Perc, Haroldo V. Ribeiro, 2020, izvirni znanstveni članek Opis: Machine learning algorithms have been available since the 1990s, but it is much more recently that they have come into use also in the physical sciences. While these algorithms have already proven to be useful in uncovering new properties of materials and in simplifying experimental protocols, their usage in liquid crystals research is still limited. This is surprising because optical imaging techniques are often applied in this line of research, and it is precisely with images that machine learning algorithms have achieved major breakthroughs in recent years. Here we use convolutional neural networks to probe several properties of liquid crystals directly from their optical images and without using manual feature engineering. By optimizing simple architectures, we fnd that convolutional neural networks can predict physical properties of liquid crystals with exceptional accuracy. We show that these deep neural networks identify liquid crystal phases and predict the order parameter of simulated nematic liquid crystals almost perfectly. We also show that convolutional neural networks identify the pitch length of simulated samples of cholesteric liquid crystals and the sample temperature of an experimental liquid crystal with very high precision.
Ključne besede: liquid crystal, neural network, artificial intelligence, soft matter
Objavljeno v DKUM: 20.11.2024; Ogledov: 0; Prenosov: 3
Celotno besedilo (1,94 MB)
Gradivo ima več datotek! Več... |
10. Tilt correction toward building detection of remote sensing imagesKang Liu, Zhiyu Jiang, Mingliang Xu, Matjaž Perc, Xuelong Li, 2021, izvirni znanstveni članek Opis: Building detection is a crucial task in the field of remote sensing, which can facilitate urban construction planning, disaster survey, and emergency landing. However, for large-size remote sensing images, the great majority of existing works have ignored the image tilt problem. This problem can result in partitioning buildings into separately oblique parts when the large-size images are partitioned. This is not beneficial to preserve semantic completeness of the building objects. Motivated by the above fact, we first propose a framework for detecting objects in a large-size image, particularly for building detection. The framework mainly consists of two phases. In the first phase, we particularly propose a tilt correction (TC) algorithm, which contains three steps: texture mapping, tilt angle assessment, and image rotation. In the second phase, building detection is performed with object detectors, especially deep-neural-network-based methods. Last but not least, the detection results will be inversely mapped to the original large-size image. Furthermore, a challenging dataset named Aerial Image Building Detection is contributed for the public research. To evaluate the TC method, we also define an evaluation metric to compute the cost of building partition. The experimental results demonstrate the effects of the proposed method for building detection.
Ključne besede: building detection, cost of building partition, deep neural network, remote sensing, tilt correction
Objavljeno v DKUM: 26.09.2024; Ogledov: 0; Prenosov: 1
Celotno besedilo (8,62 MB)
Gradivo ima več datotek! Več... |
