Tuwhera Research Repository
The Tuwhera Research Repository provides sustainable open access and archiving to a broad range of AUT produced research - theses, research outputs and datasets.
Research outputs
Deposit your research output
via Research Elements
via Research Elements
Recent Submissions
Item type:Item, Access status: Open Access , Bayesian Techniques in the Pulsar Timing Array(Auckland University of Technology, 2026) Zahraoui, El Mehdi; Maturana-Russel, Patricio; van Straten, Willem; Meyer, Renate; Gulyaev, SergeiPulsar Timing Arrays (PTAs) utilize the extreme rotational stability of millisecond pulsars to detect the nanohertz stochastic gravitational-wave background (SGWB). Robust detection requires rigorous Bayesian modeling to disentangle the SGWB from complex, highly covariant sources like pulsar, solar wind, and systematic noise. How-ever, calculating the Bayesian evidence within the high-dimensional parameter spaces of modern PTA datasets is computationally expensive, creating a bottleneck as array sizes and model complexities grow. This thesis addresses this limitation by developing highly scalable and accurate evidence-estimation techniques. Specifically, it introduces Generalized Steppingstone Sampling (GSS), which utilizes posterior-informed refer-ence distributions to significantly reduce marginal likelihood estimation costs. Building upon this, the thesis presents the Morph approximation—a novel technique that decom-poses high-dimensional posterior distributions into lower-dimensional, highly correlated blocks to serve as an optimal proposal distribution for bridge sampling. These methodo-logies were evaluated across diverse statistical benchmarks and major gravitational-wave datasets. Applied to recent NANOGrav and European PTA datasets, GSS and MorphZ accurately evaluated Bayes factors and reinforced the statistical evidence for an SGWB while minimizing computational overhead by several orders of magnitude. Furthermore, these tools enabled a comprehensive exploration of the noise-model hypothesis space of the Parkes Pulsar Timing Array (PPTA). This analysis revealed that most pulsars exhibit diffuse support for multiple noise models, demonstrated that data reprocessing fundamentally alters noise sensitivity, and established that Bayesian model averaging is necessary to prevent mis-modeling from obscuring true gravitational-wave signatures. Ultimately, this work provides a computationally tractable Bayesian framework that enables the rapid evaluation of broad hypothesis spaces and ensures the accurate characterization of noise-model uncertainty, which is essential for maximizing PTA sensitivity for future discoveries.Item type:Item, Access status: Open Access , An Analysis of Penalty Kick Behaviour(Auckland University of Technology, 2026) Mistry, Yazdy; Ryan, MatthewThis thesis examines the penalty kick game within a behavioural game theory context to better understand the decision making rationale behind players’ actions. There is currently debate to the existence of goalkeeper ‘Action Bias’ where goalkeepers are shown to dive to the sides rather than staying in the centre, even when staying in the centre more often may be strategically optimal. This research examines the dataset collected by (Bar-Eli et al., 2007) where the researchers find evidence for action bias. This thesis formally parametrises and models action bias, which has not been done before in previous research. This thesis also explores logit best responding through the use of the Quantal Response Equilibrium framework to test if bounded rationality better corresponds to real world play as opposed to perfect rationality assumed under the commonly used Mixed Strategy Nash Equilibrium framework. This research finds evidence of action bias within the primary dataset and that the inclusion of action bias within specified payoffs along with bounded rationality provide the most explanatory power out of all the models tested. Supplementary testing finds no evidence for behavioural heterogeneity. The penalty kick game cannot be treated as a singular homogeneous game across all observations, and all results must be treated as dataset dependent. This research finds there is evidence of goalkeeper action bias within the primary dataset as well as evidence that players are shown to not be acting as perfectly rational utility maximisers.Item type:Item, Access status: Open Access , Control Strategies and Stabilisation Techniques for DC/DC Converter(Auckland University of Technology, 2026) Nduwamungu, Aphrodis; Lie, Tek Tjing; Gunawardane, Kosala; Nair, Nirmal N K; Lestas, IonnisThis research investigates control strategies and stabilisation techniques for dual active bridge (DAB) and quad active bridge (QAB) converters in low-voltage DC (LVDC) microgrids supplying resistive loads, constant-power loads (CPLs) and pulsed-power loads (PPLs). These loads, the coupling between the ports of multiport converters and the nonlinear converter dynamics make fast and stable voltage regulation difficult. The thesis makes six contributions. First, a generalised harmonic model of the four-winding QAB transformer currents, including the dead time, agrees with a detailed switching simulation within 5.2% up to the 15th harmonic and supplies the coefficients of the control model; even harmonics appear only when the two half-cycles are unequal. Second, an adaptive super-twisting sliding mode controller (ASTSMC) with a nonlinear extended state observer and a feedforward decoupler is proposed for the QAB with CPLs, with a Lyapunov proof of practical finite-time stability and bounded adaptive gains. It settled a 48→60 V reference step in 1.14 ms, against 1.50 ms for an SMC with a nonlinear extended state observer and 6.00 ms for a decoupled PID, and gave the smallest dip after a load step (2.33%). Third, three known nonlinear controllers were customised for the QAB with PPLs; for a 100→200 W PPL step, the adaptive suboptimal second-order SMC limited the voltage change to 0.5/0.3 V, against up to 4.2 V for the others, at almost the same efficiency (90.72% against 90.78%). Fourth, for a resonant DAB with CPLs, a variable-gain SMC was derived that remains stable even when its base gain is below the worst disturbance, and it was combined with a physics-informed neural network (PINN) that estimates the unknown disturbance current. The PINN estimate was correct on new runs (R2 = 0.991), and after a 300→900 W CPL step the voltage drop fell from 6.2% without an estimator to 3.7%, close to an SMC with a linear extended state observer (3.1%) and better than a twisting second-order SMC. Fifth, a common test method applies the same tests, limits and tuning to all controllers of a chapter. Sixth, a laboratory DAB prototype with an OPAL-RT interface was built and tested in open loop. The controllers were evaluated in MATLAB/Simulink simulation and run on an OPAL-RT OP5700 simulator (software-in-the-loop); closed-loop tests on hardware remain future work.Item type:Item, Access status: Open Access , MoM: Multi-to-Multi Identification of Decoupled Multifaceted Clustering with Dynamic Representation Fusion(ACM, 2026-11-09) Zhang, Peipei; Zhu, Wenjie; Zhou, Qinli; Yan, Wei QiMultifaceted clustering is dedicated to excavating diverse and heterogeneous facets from complex multimedia data, which is a critical task for multimedia content understanding. Existing approaches primarily focus on decomposing a single dataset into multiple subspaces or sequentially generating alternative clusterings. However, these methods are severely constrained by the inherent limited information contained in the data itself, resulting in suboptimal clustering performance and indistinct facet mining results. To address the issue, we propose a novel paradigm shift from Single-to-Multi to Multi-to-Multi (MoM) clustering for multimedia data, which proactively integrates multiple informative views for decoupled multifaceted discovery, breaking the bottleneck of single-source data information scarcity. The proposed MoM framework consists of two core modules: the Dynamic Representation Fusion (DRF) module and the Decoupled Multifaceted Clustering (DMC) module. Specifically, the DRF module implements an adaptive fusion strategy that generates and dynamically fuses diverse representations guided by the anchors derived from a pretrained model, ensuring both fidelity and diversity. The DMC module leverages a structured self-expression model to jointly learn shared and facet-specific affinity matrices, producing well-separated and interpretable clusters. Unlike existing methods that passively decompose limited data into disjoint subspaces, our end-to-end framework actively employs a clustering-oriented representation fusion strategy to learn decoupled features tailored to each clustering facet. Extensive experiments show that MoM consistently outperforms state-of-the-art multifaceted clustering methods, validating the effectiveness of our MoM framework.Item type:Item, Access status: Open Access , Further Data on the Chemical Profile of Iberian Calendula Species by GC–MS, Including a First Report on C. tripterocarpa(Wiley, 2026-11-16) Marques, Mariza B; Ohse, Silvana; Silveira, Paulo C; Válega, Mónica SGA; Granato, Daniel; Silva, Artur MS; Pinto, Diana CGAThis study aimed to trace the chemical profile of the genus Calendula L. (Asteraceae) in the Iberian Peninsula, including Calendula arvensis, three subspecies of Calendula suffruticosa, and Calendula tripterocarpa. In total, 28 compounds were identified in hexane extracts by GC-MS. Fatty acids, terpenoids, and alkanes were reported for the first time for C. suffruticosa subsp. cinerea and C. suffruticosa subsp. tomentosa. In C. suffruticosa subsp. cinerea alkanes (63.3%) were the most abundant class, followed by terpenoids (25.9%), while C. suffruticosa subsp. tomentosa presented alkanes (66.4%) and terpenoids (21.0%). C. suffruticosa subsp. lusitanica was the most differentiated sample because it was the only one to show alcohols (2.4%), terpenoids (56.8%), carboxylic acids (24.8%), and alkanes (16%). C. arvensis and C. tripterocarpa showed a smaller number of compounds, 11 and 9, respectively. Principal Component Analysis was able to separate the species C. suffruticosa from C. arvensis and C. tripterocarpa.
