Browsing by Author "Mohd Hirzie Mohd Rodzhan, Ph.D"

This study is devoted to investigate the phase diagrams of 3-state Potts model with competing binary, ternary and quaternary interactions on Cayley tree. The 3-state Potts model was classified as the simplest quantum field-theory after Ising model and was an easily defined class of statistical mechanic model. The aim of this study is to analyze the phase diagrams of 3-state Potts model with competing interactions on Cayley tree of order 2 and 3. At vanishing temperature T, the phase diagrams are fully determined for all values of competing binary interactions J 1 , ternary interactions J t and quaternary interactions J q. The results show that there have significant change~ in the phase diagrams for the case of nonzero ]q. Lastly, we analyze in detail the set of modulated phases, with the phase being indicated by many different types of commensurate and incommensurate phases by plotting the wavevectors versus temperature. The Lyapunov exponent is computed to verify the stability of the large period phase in the narrow-interval modulated phase.

High energy physics (HEP) simulation and experimentation data are often high dimensional containing high number of features. A beyond standard model (BSM) dataset that is the supersymmetry (SUSY) event simulation dataset was clustered using self-organising map (SOM) algorithm. SOM clustering is one of the better methods to cluster high dimensional data. To verify the existence of the SUSY event in the clustered dataset, it was visualised through several different methods which are the U-matrix, principal component analysis (PCA) and spectral graph theory. U-matrix is the default representation of SOM that visualises the distance between SOM neurons. PCA reduces the dimensionality of the dataset to only 2-D and 3-D considering only the principal components. Spectral graph connects all the neurons together as a network but the implementation was limited by computational resources due to connecting all the neurons of the high dimensional data requires much more intense computational power. While both U-matrix and PCA are successful in visualising cluster(s) in digit datasets, U-matrix was unsuccessful in showing cluster for the SUSY dataset. PCA on the other hand manages to display cluster existence in the SUSY dataset. This may suggest that U-matrix is limited to a certain number of dimensions and PCA might be a better option for cluster existence verification. Further research needs to be done to probe into the potential of dimensionality reduction of clustered HEP data. The visualisation of cluster existence hints to the potential of the algorithm to be used on actual experimentation dataset.

This research is an extension of Ganikhodjaev et al. (2008), where they had generated and analysed the phase diagram consisting of the first and second nearest-neighbour binary interaction on the three-state Potts model on a Cayley tree. Therefore, in continuing the research, it is in our interest to investigate the effect of the third nearest-neighbour binary interaction to the phase diagrams of the three-state Potts model on a Cayley tree. Therefore, we generate and analyse the phase diagrams of the three-state Potts model, considering prolonged competing binary interaction "J" _"2" and〖" J" 〗_"3" on the same branch of the Cayley tree up to the third nearest-neighbour generations. We derive the recurrence system of equations while considering some ranges of competing parameters. We carry out a numerical procedure by applying several stability conditions and characteristic points into the iteration scheme. For some non-zero parameter "J" _"3" , we found the additional phases of period 5, 6, 9, and 11, with the ferromagnetic, antiphase, paramagnetic, antiferromagnetic and modulated phase. For the modulated phase, we further study the existence of phases with period larger than 12 by conducting a numerical analysis on the variation of wavevector and Lyapunov exponent. This results in the discovery of some phases with period larger than 12, which are the phases of period 13, 16, 23, 26 and 49. From the results obtained as presented in this thesis, it is clear that the third nearest-neighbour binary interaction on the Cayley tree, considering the three-state Potts model, gives significant effect to the generation of the phase diagram.

This project focuses on analyzing sensor response from different types of functionalized reduced Graphene Oxide (rGO) based VOC sensors on three selected VOC gases (acetone, toluene and isoprene), performing feature extraction and employing supervised learning for VOC classification. The rGO as sensing material was functionalized with nanoparticles (NPs); such as gold (Au), silver (Ag) and platinum (Pt) and plasma treatment; such as ammonia (NH3), hydrogen (H2) and Octafluorocyclobutane (C4F8). The sputtering duration and relative frequency power (WRF) are varied for the functionalization of the nanoparticles while the temperature is varied for the plasma treatment functionalization. The sensor response then was measured from the change of resistance signal during the presence of the VOC gas at low concentration, from 1 to 6 parts per million (ppm). Sensors with thin-film from rGO/Au NPs, rGO/Ag NPs, rGO/Pt NPs, rGO/H2 (RT) and rGO/ C4F8 had shown a good response toward the VOC gas while sensor with rGO/NH3 and rGO/H2 (except for the RT recipe) showed poor responses. Sensors that have a good response then proceed with the analysis, feature extraction and machine learning part. Average Resistance value at the presence of clean dry air (CDA) only, Rair and in the presence of the VOC gas, Rgas were extracted from the original sensor signal and manipulated into 10 new features. Then, five supervised learning models such as k-Nearest Neighbors (kNN), Random Forest (RF), Logistic Regression (LR), Support Vector Machine (SVM) and Artificial Neural Network (ANN) were benchmarked for the VOC classification task. The model performances were evaluated using k-Fold Cross- Validation and the prediction of the classification are visualized by using a Confusion Matrix. The results showed that RF and kNN have good performances with a mean of accuracy and standard deviation, 0.813 ± 0.035 and 0.803 ± 0.033, respectively. However, ANN, LR and SVM (Polynomial kernel) showed poor performance with 0.447 ± 0.035, 0.403 ± 0.041 and 0.419 ± 0.035 respectively. Based on the reported performance, it shows that 2 out of the 5 models could deal with the feature selected in the VOC dataset and it is feasible to analyze the classification of VOC gases based on single sensor arrays. It is therefore interesting to explore the analysis of combined sensor arrays for such tasks in future research.