1. Large language models for G-code generation in CNC machining: A comparison of ChatGPT-3.5 and ChatGPT-4oKristijan Šket, David Potočnik, Miran Brezočnik, Mirko Ficko, Simon Klančnik, 2025, original scientific article Abstract: This research explores the viability of producing ISO G-code for 3-axis machining with OpenAI's Chat Generative Pre-Trained Transformer models, particularly ChatGPT-3.5 and the newer GPT-4o. G-code (RS-274-D, ISO 6983) converts human directives into commands that machines can understand, controlling toolpaths, spindle velocities, and feed rates to produce particular aspects of an object. Previously, G-code was generated either by hand or through the use of computer-aided manufacturing (CAM) software along with machine-specific post-processors, both of which may require considerable time and expense. This research aimed to assess the practicality and effectiveness of specific large language models (LLMs) in generating G-code. The assessment took place in three distinct phases on a sample component that required 3-axis machining. These phases included: (1) the self-generated production of G-code for the sample component, (2) the examination of the independently generated G-code in the CAM application, and (3) the recognition and justification of mistakes in the G-code. The outcomes indicated varying abilities with promising findings. This method could accelerate and possibly enhance manufacturing workflows by decreasing reliance on expensive CAM software and specialized knowledge.
Keywords: generative artificial intelligence, intelligent manufacturing, large language models (LLM), ChatGPT, CNC machining, G-code programming
Published in DKUM: 28.11.2025; Views: 0; Downloads: 7
 Full text (4,02 MB)
This document has many files! More... |
2. Improving AGV path planning efficiency using Genetic Algorithms with Hamming distance-based initializationŽiga Breznikar, Janez Gotlih, Ž. Artič, Miran Brezočnik, 2025, original scientific article Abstract: This paper presents a Genetic Algorithm (GA) framework for warehouse navigation as a Travelling Salesman Problem (TSP) variant for Automated Guided Vehicles (AGVs). The warehouse layout is represented as a graph, where pick-up locations serve as terminal nodes. A distance matrix, computed via Breadth-First Search (BFS) enables efficient route evaluation. To promote diversity in the initial population, a Hamming distance-based vectorized initialization strategy is employed, ensuring that the chromosomes are maximally distinct. The GA balances exploration and exploitation by dynamically adjusting the fitness function. Early generations emphasize diversity, while later ones focus on solution refinement, improving convergence and avoiding premature stagnation. Our key contribution demonstrates that the Hamming distance-based approach achieves comparable or better results with significantly fewer chromosomes. This reduces computational cost and runtime, making the method well-suited for real-time AGV routing in warehouses. The framework is adaptable to structured environments and shows strong potential for integration into real-world logistics and robotics applications. Future work will focus on optimizing the algorithm and integrating it into the ROS 2 environment.
Keywords: automated guided vehicles (AGV), warehouse routing, combinatorial optimization, Hamming distance initialization, Robot operating system 2 (ROS 2)
Published in DKUM: 28.11.2025; Views: 0; Downloads: 3
Full text (677,22 KB)
This document has many files! More... |
3. Optimizacije v inženirstvu : reševanje problemov z metahevrističnimi metodami v okolju MATLABJanez Gotlih, Mirko Ficko, 2025, higher education textbook Abstract: Skripta obravnavajo temeljne pristope optimizacije v inženirstvu s poudarkom na uporabi metahevrističnih metod, kot sta genetski algoritem (GA) in algoritem rojev delcev (PSO). Namenjena so študentom in inženirjem, ki želijo razumeti tako teoretično ozadje kot praktično implementacijo optimizacijskih algoritmov v okolju MATLAB. Vključujejo poglavja o enokriterijskih in večkriterijskih optimizacijskih problemih, obravnavajo omejitve, različne ciljne funkcije ter vizualizacijo rezultatov. Vsako poglavje vsebuje strukturirane vaje in naloge za samostojno delo, ki spodbujajo razumevanje delovanja algoritmov, oblikovanje optimizacijskih modelov in interpretacijo rešitev. Poseben poudarek je na razlagi parametrov algoritmov, primerjavi konvergence ter vplivu nastavitev na vedenje optimizacije. Skripta se zaključijo s pregledom značilnih testnih funkcij in primeri Pareto front za večkriterijsko optimizacijo. Zasnovana so tako, da tudi uporabniki brez poglobljenega matematičnega znanja lahko postopoma razvijejo intuicijo za uporabo optimizacijskih pristopov v realnih inženirskih problemih.
Keywords: metahevristične metode, genetski algoritem (GA), algoritem rojev delcev (PSO), eno- in večkriterijska optimizacija, MATLAB, inženirske aplikacije
Published in DKUM: 11.11.2025; Views: 0; Downloads: 1
Full text (5,96 MB)
This document has many files! More... |
4. Strojno učenje za inženirje : koncepti, primeri in uporaba v okolju MATLABJanez Gotlih, Miran Brezočnik, 2025, other educational material Abstract: Skripta obravnavajo strojno učenje z vidika uporabe v inženirstvu, pri čemer temeljne koncepte povezujejo s praktičnimi primeri v okolju MATLAB. Predstavljeni so štirje temeljni pristopi strojnega učenja: nadzorovano učenje, nenadzorovano učenje, učenje z okrepitvijo in prenosno učenje. Za vsak pristop so podani temeljni koncepti, konkretni primeri uporabe ter naloge za samostojno delo. Poseben poudarek je na uporabi orodij, kot so Regression Learner, Classification Learner, Deep Network Designer in Reinforcement Learning Designer, s pomočjo katerih študenti razvijajo modele na podatkih, ki izvirajo iz realnih inženirskih primerov. Med njimi so obraba orodja, vibracije strojev, balansiranje sistemov in prepoznavanje predmetov. Skripta vključujejo tudi eksperimentalne podatkovne množice in praktične napotke za učenje, validacijo in izboljšavo modelov. Namenjena so študentom tehniških smeri ter vsem, ki želijo usvojiti uporabo metod strojnega učenja za reševanje konkretnih inženirskih problemov.
Keywords: strojno učenje, nadzorovano učenje, nenadzorovano učenje, učenje z okrepitvijo, prenosno učenje, MATLAB, inženirske aplikacije
Published in DKUM: 10.11.2025; Views: 0; Downloads: 8
Full text (6,25 MB)
This document has many files! More... |
5. Developing an alternative calculation method for the smart readiness indicator based on genetic programming and linear regressionMitja Beras, Miran Brezočnik, Uroš Župerl, Miha Kovačič, 2025, original scientific article Abstract: The European Union is planning to introduce a new tool for evaluating smart solutions in buildings—the Smart Readiness Indicator (SRI). As 54 energy efficiency categories must be evaluated, the triage process can be long and time-intensive. Altogether, 228 data points (or inputs) about the smartness of the buildings are required to complete the evaluation. The present paper proposes an alternative calculation method based on genetic programming (GP) for the calculation of Domains and linear regression (LR) for the calculation of Impact Factors and the total SRI score of the building. This novel calculation requires 20% (Domain ventilation and dynamic building envelope) to 75% (Domain cooling) fewer inputs than the original methodology. The present study evaluated 223 case study buildings, and 7 genetic programming models and 8 linear regression models were generated based on the results. The generated results are precise; the relative deviation from the experimental data for Domain scores (modelled with GP) ranged from 0.9% to 2.9%. The R2 for the LR models was 0.75 for most models (with two exceptions, with one with a value of 0.57 and the other with a value of 0.98). The developed method is scalable and could be used for preliminary and portfolio-level screening at early-stage assessments.
Keywords: SRI, modelling, genetic programming, linear regression, energy efficient buildings, smart buildings, optimisation
Published in DKUM: 03.11.2025; Views: 0; Downloads: 2
Full text (4,93 MB)
This document has many files! More... |
6. Napredne metode vpenjanja velikih zvarjencev brez klasičnih vpenjalnih sistemov : diplomsko deloŽak Boltavzer, 2025, undergraduate thesis Abstract: Diplomsko delo obravnava uporabo metod ničelnega vpenjanja pri mehanski obdelavi velikih zvarjencev, dimenzij 11 × 3 × 2 m in mase do 25.000 kg, v podjetju ADK, d. o. o. Zaradi omejitev standardnih rešitev je bil razvit lasten sistem z večjo nosilnostjo in ustrezno geometrijo vpenjalnih elementov. Analiza vključuje primerjavo klasičnega in novega sistema vpenjanja ter prikazuje konkretne izboljšave na področjih natančnosti, ponovljivosti in časa priprave. Rezultati potrjujejo, da je sistem ničelnega vpenjanja učinkovita alternativa klasičnim metodam pri obdelavi velikih in zahtevnih obdelovancev. S primerno izbranimi metodami dela sem tako dosegel cilje in namen, ki sem si jih zastavil v okviru diplomskega dela.
Keywords: Ničelno vpenjanje, vpenjalni sistem veliki zvarjenci, mehanska obdelava, ponovljivost.
Published in DKUM: 19.09.2025; Views: 0; Downloads: 6
Full text (4,12 MB) |
7. Reducing scrap in long rolled round steel bars using Genetic Programming after ultrasonic testingMiha Kovačič, Anže Zupanc, Uroš Župerl, Miran Brezočnik, 2024, original scientific article Abstract: At Štore Steel Ltd., continuously cast billets (180 mm × 180 mm) are reheated and rolled after cooling to room temperature. Hot-rolled bars are controlled as they cool to room temperature in specially designed cooling chambers, minimizing residual stresses and the development of pre-existing surface and internal defects. The bar ends can be additionally covered with insulating material. The cooled, rolled bars undergo examination using automated control lines to detect surface and internal defects, which primarily originate from the casting process. Internal defects are identified using ultrasonic testing. Between January 2022 and June 2023, 1550.0 tons of 61SiCr7 rolled bars, with diameters ranging from 53 mm to 72 mm and lengths from 7010 mm to 7955 mm, were examined using ultrasonic testing. The scrap was 109.6 tons (7.07 %). After collecting data on chemical composition (C, Si, Mn, Cr, Mo, Ni content), the casting process (casting temperature, cooling water pressure and flow in the first, second, and third zones of secondary cooling, as well as the temperature difference between input and output mould cooling water), and rolled bar geometry (diameter, length), scrap modelling after ultrasonic testing was carried using genetic programming. The genetic programming model suggested reducing the length of the rolled bar. Due to length multiplication, it was possible to reduce the rolled bar length from the initial lengths of 7010-7955 mm to the current lengths of 4558-6720 mm in June 2023. Based on this adjustment, a new production of rolled bars was established. By August 2024, 1251.9 tons of 61SiCr7 rolled bars were produced with the mentioned length adjustments. These rolled bars were subsequently examined using ultrasonic testing. The scrap was reduced by nearly 14 times, amounting to only 8.1 tons (0.64 %).
Keywords: steel industry, rolling, long bars, ultrasonic testing, scarp, defects, modelling, genetic programming
Published in DKUM: 27.08.2025; Views: 0; Downloads: 3
Full text (681,48 KB)
This document has many files! More... |
8. Hardened workpiece shape prediction using acoustic responses and deep neural networkJernej Hernavs, Tadej Peršak, Miran Brezočnik, Simon Klančnik, 2025, original scientific article Abstract: This study proposes a novel approach to predict the shape of hardened metal workpieces using acoustic responses processed by a deep convolutional neural network (CNN), aiming to advance automated straightening in manufacturing. Tool steel 1.2379 workpieces of varying widths (24 mm, 90 mm, 200 mm) were struck using a custom-built device, with acoustic responses captured and transformed into scalograms via Continuous Wavelet Transform (CWT). A 40-layer CNN predicted 5×9 shape matrices, validated by 3D scans. The dataset (219 shape states, 3396 recordings) was evaluated using leaveone-workpiece-out cross-validation, comparing the CNN against baseline models (linear regression, random forest, shallow CNN, XGBoost). CNN achieved competitive accuracy, demonstrating the feasibility of acoustic-based shape prediction. As a non-invasive, cost-efective complement to 3D scanning, this method ofers innovative potential for multi-modal quality control systems in manufacturing.
Keywords: metal workpiece, hardened, deep neural network, acoustic respons, shape prediction
Published in DKUM: 14.08.2025; Views: 0; Downloads: 9
Full text (1,10 MB)
This document has many files! More... |
9. Tehnološke izboljšave na proizvodni liniji za ovijanje plastičnih profilov : diplomsko deloŽiga Potočnik, 2025, undergraduate thesis Abstract: V diplomskem delu smo predstavili reševanje težave, ki se je pojavljala na proizvodnji
liniji za ovijanje plastičnih profilov. Zaradi visokih temperatur in obratovalnih časov so se
pojavljale motnje v delovanju črpalk, ki so dovajale osnovni premaz na površino profila.
Posledično je prihajalo do zastojev celotne proizvodne linije. Zato smo najprej raziskali
vzroke za zastoje črpalk, nato smo izvedli temperaturne meritve sistema in opravili
hladilni eksperiment, s katerim smo uspeli zmanjšati število zastojev. S tem smo potrdili
tezo diplomskega dela, v kateri smo predvidevali, da lahko z uvedbo preprostega
hladilnega sistema črpalk bistveno zmanjšamo število zastojev na liniji.
Keywords: ovijanje plastičnih profilov, zastoji, črpalke, meritve, temperatura, hladilni sistem
Published in DKUM: 08.07.2025; Views: 0; Downloads: 16
Full text (2,92 MB) |
10. Predictive modelling of weld bead geometry in wire arc additive manufacturingKristijan Šket, Miran Brezočnik, Timi Karner, Rok Belšak, Mirko Ficko, Tomaž Vuherer, Janez Gotlih, 2025, original scientific article Abstract: 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.
Keywords: wire arc additive manufacturing, WA AM, predictive modelling, machine learning, weld bead geometry, XGBoost
Published in DKUM: 13.03.2025; Views: 0; Downloads: 10
Full text (3,54 MB) |
