Wyniki wyszukiwania - BazTech

1

Mechanical performance prediction of asphalt mixtures: a baseline study of linear and non-linear regression compared with neural network modeling

Baldo Nicola, Rondinella Fabio, Daneluz Fabiola, Vacková Pavla, Valentin Jan, Gajewski Marcin, Król Jan

Roads and Bridges - Drogi i Mosty

|

2025

|

Vol. 24, no. 1

27--35

EN

Accurate predictions of asphalt mixtures’ mechanical performance are crucial to improve the conventional mix-design procedures and to optimize both pavements’ performance and service life. This research explores this issue by means of a comparative analysis between different modeling approaches: conventional regressions, both linear and non-linear, and artificial neural networks. The former are widely used but may lack the flexibility to capture complex relationships between testing conditions and the corresponding mechanical behavior. The latter represent promising alternatives due to their capability to successfully model non-linear interactions between variables. This research compares the predictive accuracy of these different modeling approaches using experimental data resulting from 4-point bending tests carried out under several temperatures and loading frequencies. The outcomes suggest that neural networks outperform conventional regression models in capturing complex relationships, highlighting the strengths and limitations of each modeling approach and providing insights for selecting optimal models in road pavement engineering applications.

PL

Dokładne przewidywanie właściwości mechanicznych mieszanek mineralno-asfaltowych jest kluczowe w doskonaleniu konwencjonalnych procedur projektowania mieszanek oraz optymalizacji ich właściwości i trwałości nawierzchni. Niniejsze badania dotyczą pogłębionej analizy tego zagadnienia z wykorzystaniem analizy porównawczej dwóch różnych podejść do modelowania: konwencjonalnymi metodami regresji liniowej i nieliniowej oraz metodą sztucznych sieci neuronowych. Pierwsze podejście z konwencjonalnymi metodami regresyjnymi jest szeroko stosowane, ale może mieć pewne ograniczenia co do zastosowania, szczególnie tam, gdzie należy uwzględnić złożone zależności między warunkami badania, a odpowiadającymi im wyjściowymi właściwościami mechanicznymi. Drugie podejście stanowi obiecującą alternatywę, ze względu na przydatność sztucznych sieci neuronowych w modelowaniu nieliniowych interakcji między zmiennymi. Niniejsze badania porównują dokładność przewidywania różnymi metodami predykcyjnymi właściwości mechanicznych mieszanek mineralno-asfaltowych, wykorzystując dane eksperymentalne uzyskane w badaniu cztero-punktowego zginania przeprowadzonych w różnych temperaturach i częstotliwościach obciążenia. Wyniki analiz wskazują na przewagę sieci neuronowych nad konwencjonalnymi metodami modeli regresyjnych ze względu na złożoność analizowanych zależności. Dodatkowymi efektami przeprowadzonych badań jest wskazanie mocnych i słabych strony każdego podejścia do modelowania oraz praktyczne rekomendacje dotyczące wyboru optymalnych modeli do zastosowania w praktyce inżynierskiej budownictwa drogowego.

2

Forecasting electricity prices in the Polish Day-Ahead Market using machine learning models

Sowiński Rafał, Komorowska Aleksandra

Polityka Energetyczna

|

2025

|

T. 28, z. 2

211--230

EN

Given the constantly changing market situation for electricity prices, driven by shifts in the energy mix, regulatory reforms, and broader socio-economic factors, it is necessary to reassess the understanding of price forecasting periodically. Traditional statistical methods may struggle when faced with heightened volatility, nonlinear dependencies, and rapidly changing input features. In contrast, machine learning models, particularly Artificial Neural Networks (ANNs), can adapt more effectively to complex, non-stationary patterns in price time series. In this study, six distinct artificial neural network (ANN) architectures were developed and trained using eight years of historical Polish Day-Ahead Market electricity price data (2016–2024). Four of these were plain deep learning models: a Multilayer Perceptron (MLP), a Convolutional Neural Network (CNN), a Long Short-Term Memory (LSTM) model, and a Gated Recurrent Unit (GRU) model. Two others were hybrid models combining convolutional layers with recurrent layers. The hybrid architectures, namely CNN+LSTM and CNN+GRU, were designed to leverage the capacity of CNN to automatically extract features from narrower sliding windows of past prices and the LSTM/GRU layers’ ability to capture long-term temporal dependencies. The models’ performances were evaluated using three metrics: Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and the coefficient of determination (R2). The top-performing CNN+LSTM achieved an MAE of 75.21 PLN/MWh, an RMSE of 103.64 PLN/MWh, and an R2 of 0.59. Results were also compared against several models previously reported in the literature. These results may be used to improve price forecasting by indicating the optimal pathways for building forecasting models and, in extension, lead to more efficient power system planning.

PL

Ze względu na stale zmieniające się ceny energii elektrycznej, spowodowane zmianami w miksie energetycznym, regulacyjnymi i innymi czynnikami społeczno-ekonomicznymi, konieczne staje się okre sowe weryfikowanie podejścia do prognozowania cen. Tradycyjne metody statystyczne mogą zawodzić w warunkach nasilonej zmienności, nieliniowych zależności i często zmieniających się cech wejściowych. Modele uczenia maszynowego, a zwłaszcza Sztuczne Sieci Neuronowe (SSN), potrafią skutecznie dostosowywać się do złożonych, niestacjonarnych wzorców w szeregach czasowych. W niniejszym badaniu opra cowano i wytrenowano sześć różnych modeli SSN, korzystając z danych historycznych z Polskiego Rynku Dnia Następnego z lat 2016–2024. Cztery z tych modeli to czyste modele głębokiego uczenia: wielowar stwowy perceptron (MLP), sień konwolucyjna (CNN), długa pamięć krótkotrwała (LSTM) oraz bramkowa jednostka rekurencyjna (GRU). Dwa pozostałe to architektury hybrydowe, oznaczone jako CNN+LSTM i CNN+GRU, łączą zdolność CNN do wychwytywania cech z węższych okien czasowych i umiejętność warstw rekurencyjnych do uczenia się zależności długoterminowych. Wydajność modeli oceniano na pod stawie trzech miar: średniego błędu bezwzględnego (MAE), pierwiastka ze średniego błędu kwadratowego (RMSE) i współczynnika determinacji (R2). Najlepsze wyniki osiągnęła architektura CNN+LSTM, uzy skując MAE na poziomie 75,21 zł/MWh, RSME równe 103,64 zł/MWh i R2 wynoszące 0,59. Wyniki te mogą zostać wykorzystane do usprawnienia procesów prognozowania cen energii elektrycznej poprzez wskazanie wytycznych dotyczących projektowania modeli prognostycznych opartych na uczeniu maszy nowym, co z kolei może wiązać się z wydajniejszym planowaniem działania systemu energetycznego.

3

Springback prediction of sheet metal hydroforming using finite element analysis and artificial neural networks

Fartouh Y., Bouzaffour M., Nassraoui M.

International Journal of Applied Mechanics and Engineering

|

2025

|

Vol. 30, no. 3

42--56

EN

The objective of this paper is to develop a method for the rapid estimating springback in the hydroforming process of circular sheets. First, the springback behavior has been studied with using finite element simulations for various configurations such as sheet thickness, sheet diameter, and deformation pressure. The results obtained shows an excellent correlation with the experimental data. Next, the springback of circular sheets in the setting of hydroforming has been predicted using the artificial neural networks (ANN) approach. Statistical measures, specifically the mean square error (MSE) and the coefficient (R2) are implemented for evaluating this approach. The results reveal that artificial neural networks provide an accurate, high-performance model for predicting the springback of circular sheets.

4

Machine learning-based shear bearing capacity of concrete columns confined by transverse reinforcement subjected to lateral cyclic loading

Hou Chongchi, Lv Yilei, Zheng Wenzhong, Zhang Yichao

Archives of Civil and Mechanical Engineering

|

2025

|

Vol. 25, no. 1

art. no. e7, 2025

EN

The shear bearing capacity of confined concrete columns subjected to lateral cyclic loading is an important mechanical property in investigating seismic behavior of concrete buildings. However, it is still difficult to accurately predict shear bearing capacity of confined concrete columns using traditional analysis methods owing to its complex mechanical principle and indeterminate multivariable interrelationship. In this paper, an experimental study of 15 confined concrete columns subjected to lateral cyclic loading was conducted to explore the seismic behavior of confined concrete columns. Moreover, ANN and SVR models were established to accurately estimate the shear bearing capacity of confined concrete columns based on a reliable test database consisting of 121 specimens conducted in this study and published literatures. Nine key parameters were considered as input variables, including cross-sectional area of core concrete, unconfined concrete compressive strength, shear span ratio, axial compression ratio, volumetric ratio of transverse reinforcement, yield strength of transverse reinforcement, longitudinal reinforcement ratio, yield strength of longitudinal reinforcement, and confinement type. Additionally, the model sensitivity analysis was conducted to investigate the impact of parameters on shear bearing capacity of confined concrete columns. Finally, the ANN and SVR models were evaluated by comparing with five existing predicted methods and experimental results indicating that the ANN and SVM models have enough accuracy and reliability in predicting shear bearing capacity of confined concrete columns subjected to lateral cyclic loading.

5

Integrating experimental data and neural computation for emission forecasting in automotive systems

Zimakowska-Laskowska Magdalena, Orynycz Olga, Kulesza Ewa, Matijošius Jonas, Tucki Karol, Świć Antoni

Advances in Science and Technology. Research Journal

|

2025

|

Vol. 19, no. 9

452--468

EN

This work presents an integrated approach combining experimental testing and mathematical modeling to analyze fuel consumption and pollutant emissions in a spark-ignition engine vehicle. Experimental data were obtained from chassis dynamometer tests under the WLTP driving cycle, including time series of vehicle speed, energy consumption, and CO₂, CO, THC, and other compounds emissions. Two classes of artificial neural networks were implemented to capture the complex, nonlinear relationships between driving dynamics and emission profiles: Multi-Layer Perceptrons (MLP) and Self-Associative Neural Networks (SANN). These models were trained on real-world time series data to predict vehicle speed and energy consumption as functions of emission parameters and vice versa. The models demonstrated high accuracy, especially in the validation phase, confirming their potential for forecasting and environmental performance assessment. The neural network models underwent training, validation, and testing processes, allowing for the assessment of their effectiveness in predicting energy consumption under various system operating scenarios. The results demonstrated high prediction accuracy, confirming the usefulness of ANN as a tool for analyzing complex relationships between emissions and energy efficiency. The study developed models identifying the relationships between emission parameters and energy consumption characteristics, enabling precise modeling of combustion processes. The input data included key emission indicators such as carbon monoxide (CO), carbon dioxide (CO₂), and hydrocarbons (HC, NMHC and CH4), as well as operational parameters of energy systems. Additionally, the observed patterns in energy use were interpreted through a physical lens, considering the thermodynamics and chemical kinetics of combustion processes under different driving conditions. This hybrid methodology – combining data-driven AI with domain-specific physical insight – provides a robust framework for predicting the environmental impacts of internal combustion engines and optimizing their operation. The proposed approach applies to broader engineering contexts, including emission control strategy design, digital twin development for powertrains, and intelligent vehicle energy management systems. The proposed approach represents a significant step toward leveraging modern artificial intelligence methods to improve energy efficiency and develop emission reduction strategies through combustion condition optimization. The obtained results can serve as a foundation for further refining industrial processes in the context of sustainable development and environmental protection.

6

The use of artificial neural networks to the analysis of lubricating performance of vegetable oils for plastic working applications

Szewczyk Marek, Mezher Marwan T., Jaber Tanya Abdulsattar

Advances in Mechanical and Materials Engineering

|

2025

|

Vol. 42, nr 1

5--15

EN

Sheet metal forming is the basic method of processing of deep-drawing quality steel sheets used in the automotive industry. A properly planned technological process of forming should include guidelines for friction conditions, or rather the coefficient of friction. Determination of the coefficient of friction is carried out using various methods. In this article, the strip drawing test was used to analyse the friction of low-carbon DC04 steel sheets. The tests were carried out at different contact pressures and with the use of different vegetable-oil based biolubricants. The most common edible and non-edible oils were selected for the tests: sunflower, rape-seed, moringa and karanja. The analysis of the experimental results was carried out using multilayer artificial neural networks (ANNs). Different learning algorithms and different transfer functions were considered in ANNs. Based on the analysis of experimental data, it was noticed that the coefficient of friction decreased with increasing contact pressure. The lowest values of the coefficient of friction, in the entire range of analysed pressures, were observed during lubrication with karanja oil. It was also found that Levenberg-Marquardt training algorithm with log-sigmoid transfer function provided the lowest values of performance errors and at the same time the highest value of the coefficient of determination R2 = 0.94719.

PL

Kształtowanie blach stalowych głębokotłocznych jest podstawową metodą obróbki blach stosowanych w przemyśle motoryzacyjnym. Prawidłowo zaplanowany proces technologiczny formowania powinien uwzględniać wytyczne dotyczące warunków tarcia, a właściwie współczynnika tarcia. Wyznaczanie współczynnika tarcia przeprowadza się różnymi metodami. W tym artykule do analizy tarcia blach ze stali nisko-węglowej DC04 wykorzystano test przeciągania pasa blachy. Testy przeprowadzono przy różnych naciskach kontaktowych i z wykorzystaniem różnych biosmarów na bazie olejów roślinnych. Do testów wybrano najpopularniejsze oleje jadalne i niejadalne: słonecznikowy, rzepakowy, moringa i karanja. Analizę wyników eksperymentów przeprowadzono z wykorzystaniem wielowarstwowych sztucznych sieci neuronowych. W sieciach neuronowych rozważano różne algorytmy uczenia sieci i różne funkcje przejścia. Na podstawie analizy danych eksperymentalnych zauważono, że współczynnik tarcia zmniejszał się wraz ze wzrostem nacisku kontaktowego. Najmniejsze wartości współczynnika tarcia, w całym zakresie analizowanych nacisków, zaobserwowano podczas smarowania olejem karanja. Stwierdzono również, że algorytm uczenia Levenberga-Marquardta z logarytmiczno-sigmoidalną funkcją przejścia zapewnił najmniejsze wartości błędów sieci neuronowej i jednocześnie największą wartość współczynnika determinacji R2= 0,94719.

7

Acoustic model for the classification of Polish vowels

Pondel-Sycz Karolina

Vibrations in Physical Systems

|

2024

|

Vol. 35, nr 1

art. no. 2024101

EN

The study explored the performance of vowel recognition using an acoustic model built on Audio Fingerprint techniques [1]. The research compares the performance of Support Vector Machines (SVMs), Hidden Markov Models (HMMs), Artificial Neural Networks (ANNs) and k-Nearest Neighbours (k-NN) classifiers in the recognition of isolated and within-word vowels and investigates the importance of different types of acoustic speech features in this process. Temporal, spectral, cepstral, formant, LPC and perceptual features of speech were examined. Importance of features was tested using a random forest classifier. Vowel classification was tested at three confidence levels for feature importance: 90%, 95% and 99%. Two author databases consisting of a total of 1,200 samples from 20 speakers, recorded under household conditions, were used. The classifiers were evaluated by confusion matrix, accuracy, precision, sensitivity and F1 score. A segmentation of words into speech sounds was carried out using a tool based on BiLSTM recurrent neural networks and the BIC criterion. Three most important features were determined: power spectral density, spectral cut-off, and Power-Normalised Cepstral Coefficients. In the isolated vowel recognition task, the SVM classifier was the most effective with a feature significance confidence level of 95% obtaining accuracy = 81%, precision = 81%, sensitivity = 81%, F1 score = 80%. In the task of recognising a vowel within a word, it was verified if the algorithm detected the presence of vowels in the correct segment and if it recognised the correct vowel within it. The best results were obtained by the k-NN classifier (statistical confidence level of feature importance of 99.9%). However, these results were low, correct recognition of the vowel in the word: A, E, U: 20%, I, O: 7%, Y: 23%. This indicates strong influence of the neighbourhood of other speech sounds in speech on the acoustic model of vowels and their recognition.

8

Prediction of the main environmental and energy characteristics of diesel engines using an artificial neural network for pure diesel fuel, pure HVO, and a mixture of these fuels in the ratio 50/50 by volume

Matijošius Jonas, Rimkus Alfredas, Gruodis Alytis

Transport Problems

|

2024

|

T. 19, z. 4

69--82

EN

This research assesses the influence of different quantities of hydrotreated vegetable oil (HVO) in diesel fuel on the performance of the engine and the emissions it produces. The particular areas of interest are the level of smoke emitted and the brake thermal efficiency (BTE). A series of engine experiments were undertaken to quantify emissions and performance characteristics under various operating regimes using three distinct fuel blends: D100 (pure diesel), HVO50 (50% HVO mixture), and HVO100 (pure HVO). The results show a tendency to reduce soot emissions when the amount of HVO in the mixture reaches 50%, but in order to accurately determine the correlation between the amount of HVO and emissions, additional studies with various concentrations of HVO and diesel mixtures are necessary. Similarly, the results show a slight improvement in BTE stability with a 50% HVO blend, but more studies with different percentages of HVOdiesel blends are needed to reliably determine changes in BTE. This indicates a trend of change in engine performance with increasing HVO concentration in diesel fuel. In order to forecast emissions and performance indicators under different operating scenarios, we used artificial neural networks ANNs, which demonstrated excellent prediction accuracy, exhibiting robust linear correlations between the expected and real values for all fuel types. This research emphasizes the advantages of utilizing HVO in diesel engines for both environmental impact and performance. It also emphasizes the usefulness of ANNs in optimizing engine settings to improve efficiency and minimize emissions. The results endorse the further use of HVO as a viable substitute for conventional diesel, leading to less ecological consequences and enhanced engine efficiency.

9

Analysis of the possibilities of the evolutionary algorithm to improve the neural model of the TGE S.A. Day Ahead Market System using selected programming environments

Tchórzewski Jerzy R., Zakrzewski Maciej

Studia Informatica : systems and information technology

|

2024

|

vol. 2 (31)

87--102

EN

The article contains selected research results regarding the analysis of the possibility of using the Evolutionary Algorithm to improve neural models of intelligent systems using selected programming environments. Choosing an appropriate programming language is one of the basic activities in the process of implementing complex algorithms, which include methods of artificial neural networks and evolutionary algorithms. Due to the fact that the object of the researchwas an intelligent Day Ahead Market system operating on the Polish Power Exchange and the modeling methods were artificial neural networks and evolutionary algorithms, it was decided to use very high-level programming languages such as Python, Matlab and C# for implementation and associated development environments. It turned out, among other things, that each of these languages and programming environments has its advantages and disadvantages, but all of them are very useful due to their useful syntax and rich included libraries. A thorough analysis of the implementation shows, among other things, that the choice of programming language affects the efficiency, speed and quality of the obtained implementations of system models. Against this background, the advantages and disadvantages of individual programming languages are shown, especially in the context of implementing evolutionary algorithms. The research results indicate directions for selecting an appropriate programming language and the associated programming environment for system modeling using artificial neural networks and evolutionary algorithms. In addition, the method of analysis, as well as the method of modeling and implementation was shown on the example of a specific system, which was the Day Ahead Market system of TGE S.A.

10

Kolmogorov-Arnold Networks for metal surface defect classification

Krzywda Maciej, Wermiński Mariusz, Łukasik Szymon, Gandomi Amir H.

Studia i Materiały Informatyki Stosowanej

|

2024

|

T. 16, nr 3

52--56

EN

This paper presents the application of Kolmogorov-Arnold Networks (KAN) in classifyingmetal surface defects. Specifically, steel surfaces are analyzed to detectdefects such as cracks, nclusions, patches, pitted surfaces, and scratches. Drawingon the Kolmogorov-Arnold theorem, KAN provides a novel approach compared to conventional multilayer perceptrons (MLPs), facilitatingmore efficient function approximation by utilizingspline functions. The results show that KAN networks can achieve better accuracy than convolutional neural networks (CNNs) with fewer parameters, resulting in faster convergence and improved performance in image classification.

PL

W niniejszej pracy przedstawiono zastosowanie sieci Kolmogorov-Arnold (KAN) w klasyfikacji defektów powierzchni metali. W szczególności badane są powierzchnie stali pod kątem wykrywania takich wad, jak pęknięcia, wtrącenia, łaty, powierzchnie z wżerami i zarysowania. Sieci KAN, oparte na twierdzeniu Kolmogorova-Arnolda, stanowią innowacyjną alternatywę dla tradycyjnych wielowarstwowych perceptronów (MLP), umożliwiając efektywniejsze aproksymowanie funkcji poprzez zastosowanie funkcji sklejanych. Wyniki badań wskazują, że sieci KAN mogą osiągać lepszą dokładność niż konwolucyjne sieci neuronowe (CNN) przy mniejszej liczbie parametrów, co skutkuje szybszą zbieżnością i lepszymi wynikami w klasyfikacji obrazów.

11

Wpływ parametrów sieci typu Tree Parity Machine na bezpieczeństwo Algorytmu wyrównującego wagi

Stypiński Miłosz

Przegląd Telekomunikacyjny + Wiadomości Telekomunikacyjne

|

2024

|

nr 4

81--84

PL

Znaczący postęp w technologii obliczeń z wykorzystaniem komputerów kwantowych jest ogromnym zagrożeniem dla obecnie wykorzystywanej kryptografii asymetrycznej. Sieci neuronowe typu Tree Parity Machine są alternatywnym algorytmem uzgodnienia klucza kryptograficznego. W ostatnim czasie zaproponowano szereg rozwiązań usprawniających właściwości klucz kryptograficznego otrzymanego z wykorzystaniem wspomnianego rozwiązania. Celem tego artykułu jest zbadanie zależności pomiędzy bezpieczeństwem usprawnień sieci Tree Parity Machine, a jej parametrami i rozmiarem.

EN

Significant advancements in quantum computing technology pose a major threat to currently used asymmetric cryptography. Tree Parity Machines are an alternative algorithm for cryptographic key agreement. Recently, several solutions have been proposed to enhance the properties of cryptographic keys obtained using this approach. The aim of this article is to investigate the relationship between the security of Tree Parity Machine network improvements and its parameters and size.

12

Performance of artificial neural networks and traditional methods in determining selected growth parameters of Alburnus sellal Heckel, 1843

Ozcan Ebru Ifakat

Oceanological and Hydrobiological Studies

|

2024

|

Vol. 53, No. 2

153--163

EN

In this study, predictions were made on the growth performance of Alburnus sellal Heckel, 1843 from the Munzur River using back propagation artificial neural networks and ANN algorithms. Statistical growth models used in fish biology and results obtained from artificial neural networks were compared. A total of 239 samples were used in this comparison. It was determined that the population is distributed across age groups 0–VII. The relationship between length and weight was calculated as W = 0.0046L3.198 for all individuals. The von Bertalanffy growth parameters were calculated for all individuals: Lt = 21.93 [1 – e–0.158 (t + 2.11)]; Wt = 102.29 [1 – e–0.158 (t + 2.11)]3.198. The growth performance index (Ф’) value was 1.880 for all individuals. The condition factor varied between 0.479 and 1.115 for females and between 0.533 and 1.076 for males. The Mean Absolute Percent Error (MAPE) statistic was used, which is a widely used method to measure the accuracy of the predictions made. It was determined that ANNs MAPE (%) values were better than MAPE values calculated for the length–weight relationship and von Bertalanffy growth function models for A. sellal. This study shows that ANNs can be used as an alternative useful method for predicting population parameters. ANN models are therefore an effective tool to describe fish growth parameters. They have been found to be a useful predictive tool. The developed models can be used to predict future sustainable fish management.

13

Application of multilayer neural networks for controlling a line-following robot in robotic competitions

Minaya Cesar, Rosero Ricardo, Zambrano Marcelo, Catota Pablo

Journal of Automation Mobile Robotics and Intelligent Systems

|

2024

|

Vol. 18, No. 1

35--42

EN

The paper presents an approach for controlling a line- following robot using artificial intelligence algorithms. This study aims to evaluate and validate the design and implementation of a competitive line-following robot based on multilayer neural networks for controlling the torque on the wheels and regulating the movements. The configuration of the line-following robot consists of a chassis with a set of infrared sensors that can detect the line on the track and provide input data to the neural network. The performance of the line-following robot on a running track with different configurations is then evaluated. The results show that the line-following robot responded more efficiently with an artificial neural network control algorithm than with a PID control or fuzzy control algorithm. At the same time, the reaction and correction time of the robot to errors on the track is earlier by about 0.1 seconds. In conclusion, the capabilities of a neural network allow the line-following robot to adapt to environmental conditions and overcome obstacles on the track more effectively.

14

Method of Predicting Surface Subsidence Caused by Underground Mining: A Review

Nguyen Huy Dinh, Tran Trong Dinh, Le Canh Van

Inżynieria Mineralna

|

2024

|

Vol. 2, no. 1

41--46

EN

In recent years, there has been a global increase in energy demand, with the extraction of underground mineral energy sources such as coal playing a significant role in the energy supply. However, the extraction of these natural resources always faces many challenges and risks. This process has created large voids, causing an imbalance in the original stress state within the earth and resulting in surface terrain deformations. Therefore, ensuring efficient extraction must be accompanied by safety measures. Among these, predicting surface subsidence due to underground mining is a crucial task. This paper presents an overview of the current method of predicting mining subsidence and their application scope. The result synthesizes various methodologies applied to different regions worldwide. Finally, the findings of this research can provide guidelines for establishing essential requirements for the application of surface displacement forecasting technologies due to underground mining.

PL

W ostatnich latach nastąpił globalny wzrost zapotrzebowania na energię, a wydobycie podziemnych mineralnych źródeł energii, takich jak węgiel, odgrywa znaczącą rolę w zaopatrzeniu w energię. Jednak wydobycie tych zasobów naturalnych zawsze wiąże się z wieloma wyzwaniami i ryzykiem. W procesie tym powstały duże puste przestrzenie, powodując brak równowagi pierwotnego stanu naprężeń w ziemi i powodując deformacje terenu na powierzchni. Dlatego zapewnieniu skutecznej ekstrakcji muszą towarzyszyć środki bezpieczeństwa. Wśród nich kluczowym zadaniem jest przewidywanie osiadań powierzchni na skutek eksploatacji podziem-nej. W artykule przedstawiono przegląd dotychczasowych metod prognozowania osiadań górniczych oraz zakres ich zastosowania. W rezultacie dokonano syntezy różnych metodologii stosowanych w różnych regionach świata. Wreszcie, wyniki tych badań mogą dostarczyć wskazówek do ustalenia zasadniczych wymagań dotyczących stosowania technologii prognozowania przemieszczeń powierzchni w wyniku górnictwa podziemnego.

15

Przegląd obecnie istniejących instalacji LNG na świecie, aspekty techniczne pojedynczej stacji oraz prognozowanie zużycia gazu. Część 2 - obliczenia

Cieślik Tomasz, Narloch Piotr, Żero Andrzej

Gaz, Woda i Technika Sanitarna

|

2024

|

Nr 2

2--7

PL

Jednym z warunków prawidłowego zarządzania pracą sieci wyspowej zasilanej z stacji regazyfikacji LNG jest planowanie dostaw opartych o prognozy zużycia. Na podstawie zgromadzonych danych atmosferycznych oraz zużycia gazu z wybranej stacji na przestrzeni dwóch lat określono wpływ czynników atmosferycznych na konsumpcję gazu ziemnego za pomocą sztucznych sieci neuronowych. Wyznaczono wpływ miesiąca oraz dnia (parametr sztuczny) na konsumpcję gazu. Wyznaczono model sieci neuronowych dający najlepsze dopasowania za pomocą współczynnika korelacji.

EN

One of the conditions for proper management of the islanded network supplied by LNG regasification stations is planning deliveries based on consumption forecasts. Based on collected meteorological data and gas consumption from a selected station over a two-year period, the impact of atmospheric factors on natural gas consumption was determined using artificial neural networks. The influence of the month and day (artificial parameter) on gas consumption was identified. A neural network model was developed to achieve the best fits using correlation coefficients.

16

Prediction of Veneer Bonding Strength of Wood-Based Composites Through Soft-Computing Models

Kiliç Kenan, Karaman İbrahim, Kiliç İsmail, Özcan Uğur

Drewno : prace naukowe, doniesienia, komunikaty

|

2024

|

Vol. 67, nr 214

Art. no. 194466

EN

An artificial neural network (ANN) and an adaptive neuro-fuzzy inference system (ANFIS) are used to predict the bonding strength of different wood-based composites and veneers. The dataset used for model creation is obtained from experimental setups. The experiments involved measuring the bonding strength of wood-based composites (Flakeboard, MDF, OSB) and veneer (beech, oak, pine) using different cutting directions and adhesive types. A total of 540 experiments were conducted. The main objective of this study is to propose AI-based models (ANN and ANFIS) that could reduce the cost of experiments and computational time. The ANN model achieved correlation coefficients (R2) of 0.91 and 0.94 for training and testing, respectively. The high R² values for both training and test datasets indicate that the ANN model is well-designed. On the other hand, the ANFIS model yielded R2 values of 0.88 and 0.85 for training and testing, respectively. Based on these results, the ANN models exhibited a stronger correlation than the ANFIS models. Overall, this study demonstrates the effectiveness of using artificial intelligence models, specifically ANN and ANFIS, to predict the bonding strength of wood-based composites and veneer. By employing these models, researchers can reduce the need for extensive experimentation and save computational time, making the process more efficient and cost-effective.

17

Machine learning based event reconstruction for the MUonE experiment

Zdybał Miłosz, Kucharczyk Marcin, Wolter Marcin

Computer Science

|

2024

|

T. 25 (1)

147--167

EN

A proof-of-concept solution based on the machine learning techniques has been implemented and tested within the MUonE experiment designed to search for New Physics in the sector of anomalous magnetic moment of a muon. The results of the DNN based algorithm are comparable to the classical reconstruction, reducing enormously the execution time for the pattern recognition phase.The present implementation meets the conditions of classical reconstruction, providing an advantageous basis for further studies.

18

Development of an object identification algorithm for the forging industry based on standard vision systems

Litwa Adrian, Madej Łukasz

Computer Methods in Materials Science

|

2024

|

Vol. 24, No. 1

5--13

EN

The work aims to develop an algorithm for identifying objects in a forging plant under production conditions. Particular emphasis is placed on the accurate detection and tracking of forgings that are transferred along the forging line and, if possible, detection will also cover employees controlling and supporting the operation of forging machines, all of this with the use of standard vision systems. An algorithm prepared in such way will allow the performance of effective detections that will support activities related to the control of the movement of forging elements, the analysis of safety in workplaces, and the monitoring of compliance with Occupational Health and Safety Regulations by employees, as well as also allowing for the introduction of additional optimization algorithms that will further enrich the presented model, which may prove to be a long-term goal that will form the basis for subsequent work. Three algorithmic solutions with different levels of complexity were considered during the research. The first two are based on artificial neural network solutions, while the last one utilizes classical image processing algorithms. The datasets for training and validation in the former cases were generated based on the recordings taken from standard cameras located in the forging plant. Data were acquired from three cameras, two of which were used to create training and validation sets, and a third one was used to verify how the developed algorithms would work in a variable environment that was previously unknown to the models. The impact of model parameters on the results is presented at this stage of the research. It has been proven that machine learning-based solutions cope very well with object detection problems and achieve high accuracies after a precise selection of hyperparameters. Algorithms show the performance of detections with excellent accuracy of 92.5% for YOLOv5 and 94.3% for Mask R-CNN. However, a competitive solution using only image transformations without machine learning showed satisfactory results that can also be obtained with simpler approaches.

19

Application of M5P Model Tree and Artificial Neural Networks for Traffic Noise Prediction on Highways of India

Mann Suman, Singh Gyanendra

Civil and Environmental Engineering Reports

|

2024

|

Vol. 34, no. 2

45--62

EN

Traffic noise prediction is the fastest growing development that reflects the rising concern of noise as environmental pollution. Prediction of noise exposure levels can help policy makers and government authorities to make early decisions and plan effective measures to mitigate noise pollution and protect human health. This study examines the application of M5P model tree and Artificial Neural Network (ANN) for prediction of traffic noise on Highways of Delhi. In total 865 data sets collected from 36 sampling stations were used for development of model. Effects of 13 independent variables were considered for prediction. Model selection criteria like determination coefficient (R2 ), root mean square error (RMSE), Mean absolute error (MSE) are used to judge the suitability of developed models. The work shows that both the models can predict traffic noise accurately, with R2 values of 0.922(M5P), 0.942(ANN) and RMSE of 2.17(M5P) ,1.95(ANN). The results indicate that machine learning approach provides better performance in complex areas, with heterogenous traffic patterns. M5p Model tree gives linear equations which are easy to comprehend and provides better insight, indicating that M5P model trees can be effectively used as an alternative to ANN for predicting traffic noise.

20

Combined Evaluation of Room Acoustic Descriptors in Different Structural Configurations via ODEON Simulations and Artificial Neural Networks

Do Nascimento Eriberto Oliveira, Zannin Paulo Henrique Trombetta

Archives of Acoustics

|

2024

|

Vol. 49, nr 4

543--556

EN

This study evaluated the combined sensitivity analysis of several room acoustic descriptors: reverberation time (T30), center time (Ts), early decay time (EDT), definition (D50), clarity (C50), useful-to-detrimental sound ratio (U50), and speech transmission index (STI); and also it assessed how these descriptors responded jointly to different acoustic-structural factors. The first-order factors were background noise (A), acoustic ceiling tile sound absorption coefficient (B), confinement (C), and occupancy (D), along with their interaction effects. A novel method is proposed for this joint evaluation of sensitivity factors. This method involves in situ measurements and an unreplicated 2^4 factorial design, which has been validated by ODEON software. The significance of input factors is determined using artificial neural networks (ANN) and the modified profile method (MPM), validated by multiple linear regression (MLR). Three significant correlation groups are identified at p < 0:05: group 1 (EDT, T30, Ts), group 2 (C50, D50), and group 3 (U50, STI). The ceiling material sound absorption (B) is found to affect reverberation (groups 1 and 2), while background noise (A) impacts STI and U50. A weak correlation is found between D50 and STI. These results are confirmed by the MLR and MPM methods.