Browsing by Subject "Artificial Neural Networks"

Now showing 1 - 20 of 42

Access status: Open Access ,
A repeatability study of artificial neural network predictions in flow stress model development for a magnesium alloy
(Wydawnictwa AGH, 2021) Siewior, Hubert; Madej, Łukasz
This work is devoted to an evaluation of the capabilities of artificial neural networks (ANN) in terms of developing a flow stress model for magnesium ZE20. The learning procedure is based on experimental flow-stress data following inverse analysis. Two types of artificial neural networks are investigated: a simple feedforward version and a recursive one. Issues related to the quality of input data and the size of the training dataset are presented and discussed. The work confirms the general ability of feedforward neural networks in flow stress data predictions. It also highlights that slightly better quality predictions are obtained using recursive neural networks.
Access status: Open Access ,
Analysis of the mining torque signal with Continuous Wavelet Transform
(2010) Jonak, Józef; Jedliński, Łukasz; Gajewski, Jakub
This paper presents an analysis of the excavation torgue signal with the use of a Continuous Wavelet Transform. The article also presents results of preliminary research on utilising neural networks to identify excavating cutting tools type used in multi-tool excavating heads of mechanical coal miners. Selected wavelet coefficients were used as data to teach artificial neural network. The research is necessary to identify rock excavating process with a given head, and design adaptation system for control of mining process with such a head. The results of numerical analyses conducted with the use of Neural Networks are presented.
Access status: Open Access ,
Analytical and neural correctors of temperature sensors dynamic errors
(Wydawnictwa AGH, 2010) Jackowska-Strumiłło, Lidia
The paper presents comparison of analytical and neural correctors of temperature sensors dynamic errors. Classical serial correction method using convolution equation is described and also ARX {AutoRegressive with eXogenous variables) model of the corrector is developed. A new correction method by means of Artificial Neural Networks (ANNs), in which an inverse dynamic model of the sensor is implemented by a neural corrector is proposed. Feedforward multilayer ANNs and a moving window method are applied. The described correction techniques are evaluated experimentally for two platinum resistance temperature detectors in sheath, immersed in water. In these working conditions, i.e. for which sensor's dynamic properties can be approximated by linear model the best corrector's performances and the shortest correction time t0,05 are achieved for the ARX corrector with digital moving average filter after the corrector. The results for the neural correctors are only slightly worse, but comparable. However, for the systems without filtering the best corrector's performances are achieved for the neural correctors. Obtained results indicate that the ANN-based method is less sensitive to noise interferences.
Access status: Open Access ,
Artificial intelligence approach for detecting material deterioration in hybrid building constructions
(Wydawnictwa AGH, 2021) Česnokov, Andrej Vladimirovič; Mihajlov, Vitalij Vital?evič; Dolmatov, Ivan Viktorovič
Hybrid constructions include heterogeneous materials with different behaviors under load. The aim is to achieve a so-called synergistic effect when the advantages of particular structural elements complement each other in a unified system. The building constructions considered in the research include high-strength steel cables, fiberglass rods, and flexible polymer membranes. The membrane is attached to the rods which have been elastically bent from the initially straight shape into an arch-like form. Structural materials inevitably deteriorate during a long operational period. The present study focuses on detecting material deterioration using Artificial Neural Networks (ANNs), which belong to the scope of intelligent techniques for data analysis. Appropriate ANN structures and required features are proposed. A semi-supervised learning strategy is used. The approach allows the training of the networks with normal data only derived from the construction without defects. Material degradationis detected by the level of reconstruction error produced by the network given the input data. The work contributes to the field of structural health monitoring of hybrid building constructions. It provides the opportunity to detect material deterioration given the forces in particular structural elements.
Access status: Open Access ,
Artificial neural network with radial basis function in model predictive control of chemical reactor
(Wydawnictwa AGH, 2009) Samek, David; Dostal, Peter
This paper describes the application of artificial neural network with radial basis function as a predictor in model predictive control. Radial basis function neural networks are known for their fast training. Thus, this type of artificial neural networks offers promising way how to reduce computational cost during offline predictor training and eventual online adaptation. The features of this type of artificial neural network are presented in simulations in MATLAB/Simulink on the nonlinear system control. The aim of this paper is to suggest one approach how to solve nonlinear prediction problem using artificial neural network respecting computational demands of the predictor.
Access status: Open Access ,
Artificial neural networks as a tool for supporting a moulding sand control system based on the dependency between selected moulding sand properties
(AGH University Press, 2023) Mrzygłód, Barbara; Jakubski, Jarosław; Opaliński, Andrzej; Regulski, Krzysztof
The article presents the potential for using artificial neural networks to support decisions related to the rebonding of green moulding sand. The basic properties of the moulding sand tested in foundries are discussed, especially compactibility as it gives the most information about the quality of green moulding sand. First, the data that can predict the compactibility value without the need for testing are defined. Next, a method for constructing an artificial neural network is presented and the network model which produced the best results is analysed. Additionally, two applications were designed to allow the investigation results to be searchable by determining the range of values of the moulding sand parameters.
Access status: Restricted ,
Classification of security threats in CAN communication using machine learning methods
(Data obrony: 2020-12-08) Sacha, Filip
Wydział Elektrotechniki, Automatyki, Informatyki i Inżynierii Biomedycznej
Access status: Open Access ,
Development of an object identification algorithm for the forging industry based on standard vision systems
(Wydawnictwa AGH, 2024) Litwa, Adrian; Madej, Łukasz
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.
Access status: Restricted ,
Doskonalenie metod wyznaczania przepuszczalności skał z wykorzystaniem sieci neuronowych
(Data obrony: 2009-12-17) Prętka, Joanna Małgorzata
Wydział Geologii, Geofizyki i Ochrony Środowiska
The absolute rock permeability determination with the use of artificial neural networks (ANN) was a subject of the thesis. Neural models were built on the basis of results from laboratory tests, well logs data and the results of the comprehensive interpretation. At the beginning data were prepared to use them later in ANN training process. It consisted of depth matching of results obtained in laboratory on core samples to results of well logging. Further, database of petrophysical properties was completed. The second part of the thesis consisted of neural models built to describe relation between rock permeability determined in laboratory - K_lab and calculated from borehole measurements K_Zawisza. There were described profiles of ANNs, their quality and criterion for input variable selection, trained on different data sets. Some of generated neural networks were applied on similar data set to show that complicated links between input variable and absolute permeability can be used for prediction of permeability from another data.
Access status: Open Access ,
Ensemble machine learning methods to predict the balancing of ayurvedic constituents in the human body
(Wydawnictwa AGH, 2022) Rajasekar Vani; Krishnamoorthi Sathya; Saračević Muzafer; Pepic Dzenis; Zajmovic Mahir; Zogic Haris
In this paper, we demonstrate the result of certain machine-learning methods like support vector machine (SVM), naive Bayes (NB), decision tree (DT), k-nearest neighbor (KNN), artificial neural network (ANN), and AdaBoost algorithms for various performance characteristics to predict human body constituencies. Ayurveda-dosha studies have been used for a long time, but the quantitative reliability measurement of these diagnostic methods still lags. The careful and appropriate analysis leads to an effective treatment to predict human body constituencies. From an observation of the results, it is shown that the AdaBoost algorithm with hyperparameter tuning provides enhanced accuracy and recall (0.97), precision and F-score (0.96), and lower RSME values (0.64). The experimental results reveal that the improved model (which is based on ensemble-learning methods) significantly outperforms traditional methods. According to the findings, advancements in the proposed algorithms could give machine learning a promising future.
Access status: Open Access ,
Fast automatic configuration of artificial neural networks used for binary patterns recognition
(Wydawnictwa AGH, 2001) Horzyk, Adrian
This paper presents a powerful method of an automatically generated architecture of neural networks used for binary patterns recognition, which can quickly and automatically reduce synapses in a way of minimally reducing a quality of recognition and a quality of generalization. Moreover, this method computes all weights in two runs over a learning sequence, what makes this method very fast. First, the method calculates all binary features for each pattern and then weights are computed. Furthermore, there is a quality of generalization considered because it is one of the most important factors of recognition while using neural networks.
Access status: Open Access ,
Forecasting currency exchange rate time series with fireworks algorithm-based higher order neural network, with special attention to training data enrichment
(Wydawnictwa AGH, 2020) Sahu, Kishore Kumar; Nayak, Sarat Chandra; Behera, Himansu Sekhar
Exchange rates are highly fluctuating by nature, thus, they are difficult to forecast. Artificial neural networks (ANNs) have proven to be better than statistical methods. Inadequate training data may lead the model to reach sub-optimal solutions, resulting in poor accuracy (as ANN-based forecasts are data-driven). To enhance forecasting accuracy, we suggests a method of enriching training datasets through exploring and incorporating virtual data points (VDPs) by an evolutionary method called the fireworks algorithm-trained functional link artificial neural network (FWA-FLN). The model maintains a correlation between current and past data, especially at the oscillation point on the time series. The exploration of a VDP and forecast of the succeeding term go consecutively by FWA-FLN. Real exchange rate time series are used to train and validate the proposed model. The efficiency of the proposed technique is related to other similarly trained models and produces far better prediction accuracy.
Access status: Open Access ,
From Linear Classifier to Convolutional Neural Network for Hand Pose Recognition
(Wydawnictwa AGH, 2017) Rościszewski, Paweł
Recently gathered image datasets and new capabilities of high performance computing systems allowed developing new artificial neural network models and training algorithms. Using the new machine learning models, computer vision tasks can be accomplished based on the raw values of image pixels, instead of specific features. The principle of operation of deep artificial neural networks is more and more resembling of what we believe to be happening in the human visual cortex. In this paper we build up an understanding of convolutional neural networks through investigating supervised machine learning methods suchas K-Nearest Neighbors, linear classifiers and fully connected neural networks. We provide examples and accuracy results based on our implementation aimed for the problem of hand pose recognition.
Access status: Restricted ,
Incorporated mathematical model and artifical neural network method for the electrochemical characterization of solid oxide fuel cells
(Data obrony: 2020-09-29) Gnatowski, Marek
Wydział Informatyki, Elektroniki i Telekomunikacji
Access status: Restricted ,
Integrated framework with Open-CL accelerated components for objects detection in video streams using neural architectures
(Data obrony: 2018-09-27) Zmilczak, Szymon
Wydział Informatyki, Elektroniki i Telekomunikacji
Access status: Open Access ,
Klasyfikacja przypadków medycznych metodą statystyczną
(Wydawnictwa AGH, 2010) Wajs, Wiesław
The problem of data classification with statistical methods is presented in the paper. Described classification method enables calculation of probability of disease incidence. A case of disease incidence is described with two parameters expressed in real numbers. The case can belong to a known set of cases there the disease occurred or to the set where the disease did not occur. There was presented a method for calculating probability with which a given case belongs to the set labeled as »1« or »0«. Source data used in the paper come from medical databases and are original. The algorithm of the method was checked on clinical cases. Correlation method was used for generating respective statistics. The calculated correlation at a level of 0.8 is indicative of disease occurrence, whereas the correlation coefficient at a level of 0.0 is indicative of lack of disease. This property is used in the classification algorithm. It is frequent in the clinical practice that we have one test case and we try to determine whether or not that case describes symptoms of liability to the disease. Classification is related with the occurrence of disease bpd, which is analyzed in a 3 to 4 week period preceding the disease incidence. Bronchopulmonary dysplasia is a chronic lung disease that develops in neonates treated with oxygen and positive pressure ventilation. The majority of bpd cases occur in premature infants, usually those who have gestational ege less than wp = 34 weeks and birth weight less than mu = 1500 g. These babies are more likely to be affected by a condition known as infant Respiratory Distress Syndrome, which occurs as a result of tissue damage to the lungs from being mechanical ventilator for a significant amount of time. Although mechanical ventilation is essential to their survival over time the pressure from the ventilation and excess oxygen intake can injure a newborn's delicate lungs. If symptoms of respiratory distress syndrome persist then the condition is considered Broncho-pulmonary Dysplasia. Important factors in diagnosing bpd are prematurely, infection, mechanical ventilator dependence, and oxygen exposure.
Access status: Restricted ,
Koncepcja zastosowania sztucznej sieci neuronowej w rozpoznawaniu gestów
(Data obrony: 2018-09-26) Szewczyk, Anna
Wydział Elektrotechniki, Automatyki, Informatyki i Inżynierii Biomedycznej
Access status: Open Access ,
Machine learning based event reconstruction for the MUonE experiment
(Wydawnictwa AGH, 2024) Zdybał, Miłosz; Kucharczyk, Marcin; Wolter, Marcin
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.
Access status: Open Access ,
Machine-learning methods for assessing dynamic resistance of existing bridge structures subjected to mining tremors
(Wydawnictwa AGH, 2018) Rusek, Janusz
This paper demonstrates the results of research studies aimed at creating a model that allows to determine the resistance of existing bridge structures to the impact of mining tremors. A database (created by the author of this article) of the dynamic resistance of reinforced concrete bridge structures subjected to seismic excitations commonly occurring in the Legnica-Głogów Copper District (LGOM) formed the basis for the analysis. The dynamic resistance of each structure contained in the database was expressed as the limit values of the acceleration of ground vibrations that may be carried by a given structure without compromising its safety. The study was carried out using the Support Vector Machine (SVM) method in a Support Vector Regression (SVR) approach as well as an Artificial Neural Network (ANN). The models were compared in terms of the quality of the predictions and generalization of the acquired knowledge. This allows to select the most-effective method in evaluating the dynamic resistance of existing bridge structures.
Access status: Restricted ,
Modelowanie profilowania geofizyki otworowej na podstawie innych pomiarów z wykorzystaniem sztucznych sieci neuronowych
(Data obrony: 2019-01-25) Jurek, Krzysztof
Wydział Geologii, Geofizyki i Ochrony Środowiska
Celem projektu było modelowanie wartości czasu interwałowego (DT) w profilowaniu akustycznym i gęstości objętościowej (RHOB) w profilowaniu gamma-gamma gęstościowym w oparciu o dostępne dane otworowe z wykorzystaniem sztucznych sieci neuronowych (SSN). Modelowanie zostało wykonane dla klastycznych (iłowcowo-mułowcowo-piaskowcowych) utworów miocenu, występujących w profilach analizowanych otworów wiertniczych. Dla wstępnego rozpoznania zmienności ośrodka geologicznego oraz optymalnego podziału zbioru danych wejściowych do pracy z SSN wykorzystano metodę k-średnich i algorytm V- krotnego testu krzyżowego. Przetestowano wpływ różnych parametrów na efektywność uczenia SSN i wykonano modelowanie zadanych parametrów w oparciu o optymalnie dobrane sieci neuronowe dla otworu Ci-1. Stwierdzono, że wykonanie wiarygodnego modelowania zależy nie tylko od doboru zmiennych, lecz także od sposobu konstrukcji zbioru uczącego. Stwierdzono również, że pozytywne rezultaty przynosi zastosowanie metody k–średnich do wstępnego opracowania zbioru dostępnych danych, które pozwala na zoptymalizowanie procesu uczenia sieci.