KMITL

Permanent URI for this communityhttps://dspace.kmitl.ac.th/handle/123456789/1

Browse

Settings

Subject: search.filters.subject.artificial neural networks

Search Results

Now showing 1 - 4 of 4
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
    Advancing Masonry Engineering: Effective Prediction of Prism Strength via Machine Learning Techniques
    (2026-04-01)
    Saingam, Panumas
    ;
    Chatveera, Burachat
    ;
    Nawaz, Adnan
    ;
    Ali, Muhammad Hassan
    ;
    Choudhary, Sandeerah
    Masonry buildings have shaped construction history since about 6500 BCE. They offer durability, strength, and cost effectiveness, especially in developing countries. Yet assessing compressive strength during construction remains challenging due to the constituent materials soil, cement, and stone, complicating standardization worldwide. In the present study, an innovative model based on a machine learning algorithm is put forth to predict the compressive strengths of prisms. Some important factors considered as input to the algorithm based on traditional methods are the brick and mortar strengths, prism geometry, mortar bed thickness, and empirically derived height-to-thickness (t) (h/t) ratios. Three different ANN algorithms are coded and trained on the input data, and they are based on the Levenberg–Marquardt algorithm, the resilient backpropagation algorithm, and the conjugate gradient algorithm. The optimal ANN model trained using the conjugate gradient Polak–Ribière algorithm (traincgp) achieves superior performance, with R<sup>2</sup> = 0.9881, R<sup>2</sup> = 0.9927, RMSE = 0.9914 MPa, MAE = 0.6039 MPa, MAPE = 20.9141%, VAF = 0.9881, and WI = 0.9970. Sensitivity analysis shows the height-to-thickness (h/t) ratio is the dominant influence on compressive strength, consistent with structural mechanics. The primary contributions are the systematically curated, richly parameterized dataset and its use to produce robust, physically interpretable predictions with established ANN methods.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
    Localized estimation of electromagnetic sources underlying event-related fields using recurrent neural networks
    (2023-08-01)
    O'Reilly, Jamie A.
    ;
    Zhu, Judy D.
    ;
    Sowman, Paul F.
    Objective. To use a recurrent neural network (RNN) to reconstruct neural activity responsible for generating noninvasively measured electromagnetic signals. Approach. Output weights of an RNN were fixed as the lead field matrix from volumetric source space computed using the boundary element method with co-registered structural magnetic resonance images and magnetoencephalography (MEG). Initially, the network was trained to minimise mean-squared-error loss between its outputs and MEG signals, causing activations in the penultimate layer to converge towards putative neural source activations. Subsequently, L1 regularisation was applied to the final hidden layer, and the model was fine-tuned, causing it to favour more focused activations. Estimated source signals were then obtained from the outputs of the last hidden layer. We developed and validated this approach with simulations before applying it to real MEG data, comparing performance with beamformers, minimum-norm estimate, and mixed-norm estimate source reconstruction methods. Main results. The proposed RNN method had higher output signal-to-noise ratios and comparable correlation and error between estimated and simulated sources. Reconstructed MEG signals were also equal or superior to the other methods regarding their similarity to ground-truth. When applied to MEG data recorded during an auditory roving oddball experiment, source signals estimated with the RNN were generally biophysically plausible and consistent with expectations from the literature. Significance. This work builds on recent developments of RNNs for modelling event-related neural responses by incorporating biophysical constraints from the forward model, thus taking a significant step towards greater biological realism and introducing the possibility of exploring how input manipulations may influence localised neural activity.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
    A Comparison of the Response and Voltage Regulation Performance of the Single-Channel DC/DC Boost Converter Circuit with Artificial Neural Networks, Fuzzy Logic and PID Controllers
    (2023-02-01)
    Chaithanakulwat, Arckarakit
    ;
    Thungsuk, Nuttee
    ;
    Savangboon, Teerawut
    ;
    Ngao-Ngam, Somchai
    ;
    Kanharin, Phatcharaphong
    Fixed-speed wind turbines for generating electricity are also important because they are clean energy and do not pollute the environment. Developing maximum wind energy tracking and increasing DC voltage to optimal values is also an important factor designers must consider so that the power from generators connected to wind turbines can function efficiently. Therefore, in this researcher paper proposed a single-channel dc/dc boost converter control, three forms of algorithms consisting of neural networks, fuzzy algorithms and PID algorithms. The purpose of bringing these algorithms controlled because they wanted to compare the response and voltage control performance of the single-channel DC/DC boost converter to be associated with a three-phase inverter that controls PWM signal modulation with space vector technique. However, the principles and methodologies in this article are presented to simulate the algorithmic response using the MATLAB/Simulink program and compare it with the prototype mechanism. A comparison of the response performance and voltage regulation of the single-channel DC/DC boost converter showed that the three algorithms have different advantages and disadvantages but can be used together to achieve high efficiency.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
    Hybrid algorithm for training feed-forward neural networks using PSO-information gain with back propagation algorithm
    (2012-10-02)
    Sanguanchue, Tanyawat
    ;
    Jearanaitanakij, Kietikul
    This paper proposes a hybrid algorithm for training a feed-forward neural network by combining both Particle Swarm Optimization (PSO) and Information Gain with Backpropagation (BP) algorithm. A conventional neural network training algorithm, i.e. BP, has several drawbacks in its slow convergence and local optima. Although PSO can be applied to search for the near optimal set of weights in the neural network, it may still stuck in the local optima because its fitness function depends merely on the error of the network. By combining the information gain of attributes in the dataset with the fitness function of PSO to train weights in the neural network, we find out that the resulting network has a significant improvement on its recognition rate. The comparisons among other training algorithm on two real-world datasets are provided and discussed. © 2012 IEEE.