Welcome to the Polen - ITU Academic Archive

Recent Submissions

• Loading...

  Item type:Item, Access status: Open Access ,

  Graph-aware fraud detection in review platforms via layered feature engineering

  (Graduate School, 2026-06-09) Tutuk, Mehmet Ali Han; Bahtiyar, Şerif; 504221519; Computer Engineering

  Show more

  Online review platforms shape how users choose products, and the same review and rating signals feed the recommender systems that build on top of them. That same influence makes review platforms a target. Attackers write fake reviews, pay for positive ones, and coordinate rating activity to push specific products up or down; this pulls user decisions and the recommenders trained on these signals in the same direction. This thesis treats the detection of such reviews as a learning problem on review graphs. The difficulty is mostly structural. A fraudulent review almost never appears alone. Fraud accounts target the same products, post within narrow time windows, and produce similar rating patterns, and some imitate legitimate reviewer behavior closely enough to mask single-review evidence. These patterns are only visible when reviews are compared against one another. A natural representation for this kind of signal is a graph that links reviews, users, and products through shared targets, ratings, and timestamps. A review's position in this graph encodes information that no review-level feature can express on its own. Most recent work on this problem uses graph neural networks that operate on the review graph directly. Designs based on graph convolution, attention, and relation-specific message passing have produced consistent improvements on the standard review-fraud benchmarks. There is a cost to these improvements. End-to-end GNN training scales poorly in both nodes and relations, and the techniques used to reduce training cost discard part of the same relational signal that motivated the use of graphs. The trade-off becomes acute on the larger review datasets, where running the published GNN baselines under their original protocol is no longer practical at full scale. A different design is followed in this thesis. The relational signal is extracted from the graph and added to the feature space of a tabular classifier, instead of being processed inside an end-to-end graph model. The feature space is built in three levels: a base level that encodes review, text, and user attributes; a neighborhood level that aggregates the same attributes over local graph neighborhoods defined by the relation layers available in each dataset; and a full level that further appends relation-wise edge counts. Classification is then performed by gradient-boosted tree learners (XGBoost, LightGBM) over the resulting feature matrix. The three levels are designed for layer-by-layer comparison. Two graph-learning benchmark datasets, YelpChi (45,954 review nodes) and Amazon (11,944 user nodes), are used together with two larger datasets constructed from raw Yelp metadata, YelpZip (608,598 review nodes) and YelpNYC (359,052 review nodes). The two constructed datasets do not come with a graph; their relational structure is built in this thesis from the user, target, timestamp, and rating fields of the raw review records. A fixed 40/20/40 train/validation/test split and a 16-regime preprocessing sweep are applied uniformly across all four datasets, so the contribution of the layered representation can be read separately from the contribution of preprocessing. On the benchmarks, the Full layer reaches F1-macro 0.8989 on YelpChi and 0.9347 on Amazon, against the SplitGNN reference of 0.7403 and 0.7352 on the same metric, with consistent gains on AUC and GMean as well. Recall on YelpChi is 0.8016, below the 0.8256–0.8577 range reported by BWGNN, H2-FDetector, and SplitGNN; the corresponding false-alarm rate on the genuine class is 2.36%. The layered improvement is not driven by a single favorable configuration: the Base < Nbr < Full F1 ordering holds in 160/160 rows on YelpChi and 111/160 on Amazon, and Wilcoxon signed-rank tests confirm the same direction at p ≤ 0.010 across all six dataset-transition pairs in the primary 10-seed test. The location of the gain differs across datasets. On Amazon, almost all of the improvement comes from the neighborhood layer; on YelpChi, the structural layer adds a further Recall increase above the neighborhood step. The constructed datasets split this overall pattern by structural form. Eight attack-aware proxy subsets are defined on top of the constructed graphs, derived from structural attack patterns in the review-fraud literature: coordination-driven families on shared-target and burst activity, connectivity- and density-driven families on repeated-user relations and locally dense fraud blocks, and a camouflage family on fraud nodes embedded in genuine-dominant neighborhoods. The Base → Full F1 gain reaches several percentage points on the coordination-driven families, stays smaller but consistent on the connectivity- and density-driven ones, and is weakest on the camouflage regime, where absolute scores stay well below the rest of the families. The RSR hybrid family shows a Base → Nbr step that does not clear a one-sided Wilcoxon test on either constructed dataset, while still benefiting from the structural layer. The dominant layer therefore changes with the structural form of the underlying fraud, and the layered hierarchy makes that dependence readable at the metric level. Some limitations follow from the scope of the thesis. Detection metrics are reported, but a matched-hardware runtime and memory benchmark against the GNN baselines is not, so the lightweight side of the framework is supported by architectural choice and not yet by direct measurement. GNN comparisons on the constructed graphs are also outside the scope of the thesis, since there are no published GNN results on these constructed datasets and re-implementing existing GNN methods at this scale was not part of the present work. The four datasets used here also represent a single temporal snapshot, while real review platforms produce continuously arriving reviews. A fourth limitation concerns the constructed-side analysis: the eight proxy subsets are evaluated under a single seed with fixed predicate thresholds, fixed subset-size caps, and the dataset-native fraud-class proportion, so the family-level numbers should be read as a structural diagnosis rather than as precise comparative measurements. Four follow-up studies grow naturally out of these limitations: a matched-hardware runtime and memory benchmark against the GNN baselines under identical splits, running those same GNN baselines on the constructed graphs, a broader subset-level ablation along seed, threshold, cap, and class-balance axes, and a streaming or dynamic-graph extension in which the relation views are updated as new reviews arrive.

  Show more
• Loading...

  Item type:Item, Access status: Open Access ,

  Fuel cell air supply system control using deep q-network

  (Graduate School, 2026-06-03) Oğuz, Emir; Onat, Ahmet; 504231112; Control and Automation Engineering

  Show more

  The increasing global population and economic growth are causing a continuous rise in energy consumption. Energy production remains an indispensable component of industrial development for every country. However, traditional fossil fuel sources such as coal, oil, and natural gas are causing serious environmental problems due to their rapid depletion as well as greenhouse gas emissions that accelerate global warming. Considering the deteriorating environmental conditions, rising global temperatures, and greenhouse gas emissions, the transition from fossil fuels to renewable energy sources has become imperative. In this context, clean and efficient energy sources such as solar, wind, geothermal, and hydrogen emerge as the most viable alternatives to replace conventional sources. Particularly hydrogen exhibits great potential as both a fuel and an energy carrier and storage medium due to the intermittent nature of solar and wind energy. Through the carbon-free alternatives it offers, hydrogen is increasingly being recognized as a globally viable energy source and fuel alternative. Hydrogen is positioned as the primary fuel for a new generation power generation device that produces electricity, heat, water, and zero pollution using oxidant and hydrogen. The fuel cell (FC) operates through an electrochemical redox reaction: the reaction of hydrogen and oxygen produces energy, water, and heat. At the anode, hydrogen undergoes catalytic decomposition to form protons and electrons; while protons pass through the proton exchange membrane to the cathode, electrons are transferred to the external circuit to provide energy to the power load. At the cathode, oxygen reacts with protons and electrons to produce water and heat. When evaluating the environmental impacts of mobility vehicles, three main propulsion systems emerge: internal combustion engines, battery electric vehicles, and FC electric vehicles. Internal combustion engines produce significant amounts of carbon dioxide, nitrogen oxides, and particulate matter emissions as they are based on the direct combustion of fossil fuels. Battery electric vehicles, while offering zero tailpipe emissions, have disadvantages such as limited range and long charging times. Proton exchange membrane fuel cells (PEMFC) have become a prominent technology, particularly in the automotive sector, due to advantages such as zero emissions, low noise, fast response time, long range, and short refueling time. PEMFC systems are electrochemical devices that directly convert chemical energy into electrical energy and find a wide range of applications from portable power applications to hybrid electric vehicles and FC electric vehicles, with characteristics of high efficiency, high power density, low operating temperature, and minimal pollution. PEMFC systems generally consist of 5 subsystems: thermal management subsystem, water management system, hydrogen supply subsystem, air supply subsystem, and power electronics management subsystem. While the hydrogen supply subsystem uses fast electrical valves to minimize the pressure difference between the anode and cathode, the air supply subsystem provides the necessary oxygen to the cathode through a compressor. Since water and temperature dynamics change much more slowly compared to air flow dynamics, many studies assume that the FC operates under constant temperature conditions. The power electronics subsystem controls load connection and includes application-specific converters. The most critical component of the air supply system is the air compressor. For stable operation of the FC, the presence of sufficient oxygen at the cathode side is mandatory. The oxygen excess ratio (OER, denoted with λ) is defined as the ratio of the amount of oxygen supplied to the cathode to the amount of oxygen consumed in the electrochemical reaction and is one of the fundamental indicators of system efficiency. Due to the slow air flow dynamics of the compressor and air supply manifolds, when the load current changes suddenly, oxygen starvation can occur at the cathode if sufficient oxygen cannot be rapidly supplied. This situation can lead to the formation of hot spots on the membrane surface and irreversible damage to polymeric membranes. Additionally, oxygen starvation causes serious problems such as low power generation, membrane degradation, and even shortening of stack operating life. The manifold system is modeled as a collective volume representing all pipes that supply gas to the stack and drain gas from the stack. The supply manifold pressure depends on downstream cathode dynamics, and the dynamics of cathode variables depend on both the air from the supply manifold and the amount of external current load. Based on the ideal gas assumption, the partial pressures of oxygen and nitrogen can be derived through differential equations. OER control is of great importance for both the economic efficiency and safety of PEMFC systems. The system faces challenges such as nonlinearity, parametric uncertainty, and load disturbances. During rapid load current increase, the chemical reaction within the FC accelerates substantially, and this situation can lead to excessive oxygen consumption and even oxygen starvation that can cause permanent cell damage. The OER at the cathode must be kept greater than 1; however, this should not be interpreted to mean that higher OER always implies better performance. Conversely, excessively high OER causes the air compressor to consume more power and consequently reduces net power output. Net power is defined as the difference between stack power generation and parasitic power consumed by the compressor, and at high loads, these parasitic losses can reach up to thirty percent of the FC's power consumption. Research shows that maximum net power for different stack currents and OER values is obtained when OER is between 2 and 2.5. Therefore, the primary objective of OER control is to prevent oxygen starvation while minimizing compressor power consumption and thus optimizing system efficiency. The optimal OER reference varies depending on load current and is generally found in a value range between 1.5 and 3. One of the greatest challenges encountered in control design of PEMFC systems is the complex and highly nonlinear dynamic structure of the system. Traditional model-based control approaches require an accurate system model; however, PEMFCs exhibit nonlinear and time-varying uncertainties due to aging and degradation. This situation creates significant limitations as the performance of methods such as model predictive control (MPC) is largely dependent on an accurate system model. In this context, reinforcement learning-based approaches offer significant advantages. The Q-learning algorithm is a model-free approach that determines an optimal policy without requiring estimation of the environment's dynamics. This method works in a data-driven manner to address challenges such as system nonlinearity, parametric uncertainty, and load disturbances, and does not require characterization of system dynamics. The findings demonstrate that PI and reinforcement learning collaboration markedly improves transient OER control by increasing response speed and diminishing both safety-critical OER violations and efficiency-reducing deviations, especially when the baseline PI controller is inadequately calibrated. The decline in steady-state accuracy in specific configurations underscores the necessity for precise reward shaping and action constraint design to prevent ongoing corrective behavior approaching equilibrium. The findings indicate that hybrid control performance is fundamentally dependent on configuration and support that reinforcement learning is most effective as a supplementary corrective mechanism rather than a standalone alternative for classical PI control in PEMFC air management systems, where durability and efficiency must be concurrently maintained.

  Show more
• Loading...

  Item type:Item, Access status: Open Access ,

  Environmental and economic analysis of a cruise ship using LNG–MGO as alternative fuels

  (ITU Graduate School, 2026) Altun, Ramazan; Zincir, Burak; 512231028; Maritime Transportation Engineering

  Show more

  As global rules to reduce greenhouse gas (GHG) emissions become stricter, the shipping industry, responsible for approximately 3% of global emissions, faces increasing regulatory pressure. The International Maritime Organization (IMO) aims to reduce CO₂ emissions by 40% by 2030 and achieve net-zero emissions by 2050, encouraging shipowners to consider alternative fuels such as LNG, hydrogen, ammonia, methanol, and biodiesel. Among these, LNG is currently the most widely used due to its technological maturity and lower (SO)x, (NO)x, and particulate emissions. This study examines the operational, environmental, and economic performance of a cruise vessel with dual-fuel engines operating on LNG and MGO over a typical 7-day itinerary. The average weekly consumption amounts to 150 tons of MGO and 570 tons of LNG. While LNG usage on the engine can reduce CO₂ emissions by up to 27% and (NO)x by 79% according to engine exhaust emission data, frequent shifts back to MGO significantly reduce overall benefits. A well-to-wake analysis shows that, including upstream emissions, LNG achieves only a 23.0% reduction in CO₂ emissions relative to MGO. When engine operating durations in MGO and LNG modes are included in the calculation, the CO₂ emission reduction decreases to 19,1%. The use of LNG reduces the CII index by 13,1%. Based on actual consumption data, (CO)2e emission reduction of 1,0% observed with the usage of LNG as an alternative fuel. Methane slip issue in LNG use increases the GFI value, narrowing the gap between LNG and MGO. To further improve LNG's decarbonization potential, additional mitigation pathways were explored. IMO emission measuring methods, including CII, GHG intensity, and GFI were calculated for this example vessel. Cost-benefit and SWOT analyses were conducted to assess LNG's effects on the vessel. Although LNG is an alternative fuel with a 20,1% advantage when consumption data and maintenance costs are included, it provides a 14,6% cost advantage. Operating the engines in LNG mode offers advantages in both fuel costs and carbon-emission reductions. Engines do not operate continuously in LNG mode. It was observed that the engines operate in MGO mode for 12,1% of the total operation time. Operational constraints and unexpected malfunctions lead to increased engine running hours in MGO mode. Operation in MGO mode diminishes the effectiveness of emission reductions, increases fuel expenditures due to higher fuel prices, and results in elevated vessel emission indices and associated carbon tax liabilities. Consequently, the payback period of the LNG system investment has been extended by more than 12.1%. In conclusion, LNG represents a practical short-term transition fuel for meeting IMO decarbonization goals, but long-term compliance will likely require engines capable of using zero-carbon fuels or integration of onboard carbon capture systems.

  Show more
• Loading...

  Item type:Item, Access status: Open Access ,

  A novel current reuse low noise amplifier design operating in 10-18 ghz

  (Graduate School, 2026-05-18) Yavuz, Yavuzhan; Yelten, Mustafa Berke; 504231229; Electronics Engineering

  Show more

  Low-noise amplifiers (LNAs) are essential components extensively utilized in commercial and military applications, including communication, radar, and remote sensing systems. Although their fundamental purpose is signal amplification, the defining characteristic of these structures is the specialized design of the input impedance matching, which prioritizes achieving the minimum noise figure (NF) over maximizing power transfer or minimizing input reflection. While NF remains the primary critical factor in LNA design, the increasing complexity of communication systems, the widespread use of modulated signals, and advancements in radar technology have heightened the demand for front-end designs with high dynamic range, making the linearity parameter equally significant. Furthermore, as satellite technologies advance, constraints on battery life and space-based energy generation necessitate wideband, low-power receiver architectures, thereby imposing significant pressure on the power consumption of LNAs. To meet these evolving demands while maintaining a low noise figure, the current-reuse LNA topology has emerged as a relatively new approach, with ongoing research across GaAs, CMOS, and GaN processes. The current-reuse topology is based on redirecting the DC current used in the output stage to supply the input stage. In this configuration, the sources of the output-stage devices are connected to the ground via DC-blocking capacitors, while the drains of the input-stage transistors are connected to the sources of the output-stage transistors through RF-choke inductors. This situation allows for significant savings in external DC power consumption—typically consumed by blocks where small-signal gain and NF are critical but high linearity is not—without degrading the noise figure. Despite its low power consumption and optimized NF, the current-reuse topology often suffers from linearity issues due to the loss of output-stage voltage headroom. Increasing the current to improve linearity typically causes a deviation from the optimal VGS required for NFmin; thus, the linearity of the structure remains limited by the dimensions and current levels that allow for the minimum noise figure. In the design methodology developed in this thesis, this problem is addressed by implementing a distributed amplifier (DA) structure in the input stage. By adjusting the number of sections in the distributed amplifier, each transistor can be biased at the specific VGS that yields the NFmin value, regardless of the output-stage current for linearity. Consequently, the mandatory trade-off between linearity and noise figure can be adjusted flexibly to meet the design requirements. A methodical approach is also proposed for implementing the developed topology. In this context, the design and layout were performed using a commercial GaAs 150 nm process. The final circuit, including parasitic effects obtained from EM simulations, was analyzed using the Small-Signal and Large-Signal (Harmonic Balance) methods in Advanced Design System (ADS). Post-design results demonstrate a 25 dB gain across the 10-18 GHz band, with a P1dB of 17 dBm and an OIP3 of 29 dBm. While consuming approximately 250 mW (51.8 mA at a 5 V supply), the noise figure remained below 2 dB at the worst-case point. Considering future fabrication and testing phases, the design was finalized with dimensions of 1500 μm × 2300 μm, suitable for production and packaging.

  Show more
• Loading...

  Item type:Item, Access status: Open Access ,

  Cfd-based design and experimental assessment of a gerotor pump for aircraft applications

  (ITU Graduate School, 2026) Höçük, Yunus; Eken, Seher; 511221214; Aeronautics and Astronautics Engineering

  Show more

  A pump is a device that transfers fluid from a low-pressure region to a higher-pressure region by converting mechanical energy into hydraulic energy. Its main functions are to transport fluids, increase pressure, and ensure circulation. Pumps are typically used in liquid systems and are driven by power sources such as electric motors, hydraulic systems, or internal combustion engines. The gerotor pump, classified under gear pumps, is a positive displacement type consisting of two intermeshing gears. The inner gear has fewer teeth than the outer gear, and their eccentric rotation generates the volumetric variations required for suction, compression, and discharge. Fluid enters the cavities of the gerotor through the suction port, becomes pressurized due to the chamber volume change during eccentric rotation, and is subsequently delivered to the system via the discharge port. Gerotor pumps are widely employed in aerospace (power transmission, propulsion system lubrication, and cooling) and automotive industries (fuel and lubrication systems). The design of a gerotor pump must be tailored to the specific requirements of the target system. Key parameters such as flow rate, pressure, temperature, and operating speed are initially defined. Based on these requirements, two-dimensional profiles are developed, ensuring cavitation and filling ratio criteria are satisfied under operational conditions. The inlet and outlet port geometries are then defined in direct correlation with the gerotor profiles. Following this stage, three-dimensional CAD models are generated, and manufacturing is carried out. Experimental testing of the manufactured pump is essential to assess performance and compliance with design requirements. Design deficiencies at this stage may result in financial loss and time delays. Performing Computational Fluid Dynamics (CFD) analyses prior to manufacturing enables prediction of hydraulic performance under operating conditions. CFD provides insight into pump efficiency at specified operating speeds, pressures, and temperatures, thereby minimizing risks of cost and schedule overruns. In CFD studies, three primary leakage mechanisms must be investigated: tip leakage, axial leakage, ,radial leakage and bushing leakage. Each leakage type should be analysed independently for all operating conditions. To validate the numerical models, experimental testing under identical conditions is required. Within the scope of this study, a comprehensive analysis matrix was established to evaluate the hydraulic performance of the gerotor pump under various operating conditions. Both CFD analyses and experimental tests were carried out by systematically varying rotational speed, outlet pressure, and fluid temperature. This approach enabled a detailed investigation of the combined effects of operating parameters on pump performance and internal leakage characteristics. xxiv The results indicate that fluid temperature and outlet pressure are among the most influential parameters affecting pump performance. An increase in operating temperature led to a reduction in fluid viscosity, which significantly increased internal leakage rates and resulted in a noticeable decrease in volumetric efficiency. Similarly, an increase in outlet pressure intensified reverse flow from the high-pressure region to the low-pressure region, further reducing the delivered flow rate. Among the investigated leakage mechanisms, tip leakage was identified as the dominant and most pressure-sensitive leakage type, while axial, radial, and bushing leakages exhibited secondary but non-negligible contributions depending on operating conditions. The numerical results obtained from the CFD analyses showed good agreement with the experimental measurements across the defined analysis matrix. The deviations between numerical and experimental flow rate results remained within acceptable limits, confirming the validity of the developed CFD model. Overall, the findings demonstrate that the proposed CFD-based methodology provides a reliable and efficient tool for predicting gerotor pump performance, supporting design optimization and reducing the need for extensive experimental testing in aerospace pump applications.

  Show more