Kocaeli University Research Information System
Not a member yet
80615 research outputs found
Sort by
Novel hybrid machine learning framework for high-fidelity prediction of fly ash-based geopolymer concrete strength
No full textThe complex, non-linear relationships between mix-design parameters and the mechanical properties of geopolymer concrete (GPC) are not fully understood, presenting a fundamental scientific challenge for accurate strength prediction and mix optimization. This challenge hinders the widespread adoption of GPC, a sustainable alternative to conventional concrete. Existing machine learning models for GPC often lack generalizability and interpretability due to the limited availability of datasets and basic architectures. This research introduces T-BoostNet, a novel hybrid machine learning (ML) framework combining Transformer architectures with XGBoost, designed for superior accuracy and interpretability in predicting GPC compressive strength. Leveraging an unprecedented dataset of 1117 unique GPC mixtures from 77 diverse studies, T-BoostNet effectively captures intricate local and global feature interactions. T-BoostNet consistently outperformed five benchmark ML algorithms, achieving the highest R-2 = 0.848 +/- 0.024 and MAE = 3.56 +/- 0.30 MPa. SHAP analysis provided crucial interpretability, identifying curing period, water content in alkaline solution, specimen age, and curing temperature as the most influential factors. This framework advances sustainable construction by providing a reliable, interpretable tool that accelerates GPC adoption, reduces costly laboratory trials, and aligns with global low-carbon material goals
Hybrid Reinforcement of PLA With Pumice and Basalt Fiber: A Synergistic Approach
No full textThis study investigates the mechanical, thermal, and tribological behavior of PLA-based composites reinforced with basalt fiber (BF) and pumice, along with matrix modification using Joncryl. By systematically combining hybrid reinforcement and matrix modification strategies-typically examined separately in the literature-this study presents a novel approach that reveals their synergistic effects. Pumice was silanized to enhance matrix-filler bonding. FTIR analysis confirmed the successful silanization of pumice particles by the appearance of characteristic -NH2, Si-O, and Si-C bond peaks, indicating effective surface functionalization. The highest tensile strength of 87.1 MPa was achieved in the composite containing 3 wt% pumice and 10 wt% BF, which further increased to 98 MPa with the addition of 2 wt% Joncryl, indicating a significant synergistic effect. In adhesive wear tests, the hybrid composite containing 3 wt% pumice and 10 wt% BF showed no sudden coefficient of friction (COF) increase up to 100 m, indicating a significant improvement in wear resistance. With the addition of 2 wt% Joncryl, the COF values decreased further, confirming enhanced interfacial adhesion and protection of the matrix during sliding. DSC results showed that the addition of 2 wt% Joncryl to the hybrid composites (BF + 3 wt% pumice) increased the glass transition temperature from 51.3 degrees C to 58.7 degrees C while decreasing the relative degree of crystallinity from 101.4% to 79.6%, indicating that enhanced interfacial interactions restricted PLA chain mobility and hindered crystalline ordering. These findings offer a promising route to eco-friendly PLA composites using local materials, with potential for advanced applications
Letter to the editor regarding "ChatGPT delivers satisfactory responses to the most frequent questions on meniscus surgery"
No full textBackground: This letter addresses methodological aspects of a study that evaluated a large language model's responses to frequently asked patient questions regarding meniscus surgery. The original research collected common questions from orthopedic resources, submitted them to the model in separate sessions, and assessed the answers using an adequacy and clarification framework. The purpose of this correspondence is to highlight methodological limitations in the prompting strategy and evaluation procedures that may have influenced the study's findings. Methods: The study used zero-shot prompting without specifying audience level, communication role, or expected response style. Each response was rated by a single orthopedic specialist using a qualitative scale. This letter reviews these methods and discusses how alternative approaches could enhance validity and reproducibility. Results: Zero-shot prompts without audience targeting or role-based instructions can lead to general, non-specific outputs rather than patient-focused explanations. Structured prompting techniques, such as defining the audience or providing one-shot or few-shot examples, often improve clarity, consistency, and alignment with patient needs. In addition, assessment by a single rater increases subjective bias, as individual interpretation may influence scoring. Multi-rater evaluation with standardized agreement metrics would provide a more reliable and objective assessment of accuracy. Conclusions: The study offers useful preliminary insight into the potential of artificial intelligence tools for patient education. However, limitations in prompting design and evaluator methodology restrict the strength of the conclusions. Future studies employing structured prompts and multi-rater assessments may yield more robust and clinically meaningful evidence. (c) 2025 Elsevier B.V. All rights are reserved, including those for text and data mining, AI training, and similar technologies
Testing the usability of metal mesh diffusion layer in the Direct Borohydride Fuel Cell (DBHFC)
No full textSince fuel cell systems are an emerging technology, studies to develop alternative materials for fuel cell components are of great importance. In this study, the usability of stainless-steel metal mesh material was tested as diffusion layer (DL) for DBHFC, experimentally. The catalyst coated diffusion layer was obtained by coating the stainless steel meshes at different wire diameters, surface area, and opening ratio with palladium by the electrodeposition method. The newly developed DL was tested in a single cell experimental setup, which has a 2 × 2 cm2 active area. It has been observed that the cell with DL, where the wire diameter is high, has a relatively higher performance. Metal mesh DL, which is quite inexpensive compared to carbon-based DL material, achieves power outputs of up to 3.14 mW/cm2. Results showed that metal mesh material could be used as an alternative DL material for the fuel cell
Development of Pyridine Polybenzimidazole/Boron Nitride Composite Proton Exchange Membranes for Potential Use on High-Temperature Fuel Cells
No full textThe incorporation of amine functionalized hexagonal boron nitride nanofillers (NH2-h-BN) at various amounts (10–20 %wt) to pyridine-based polybenzimidazole (Py-PBI) for the next-generation high temperature proton exchange membrane fuel cells was studied. Fourier transform infrared (FTIR) spectroscopy, X-ray diffraction (XRD), scanning electron microscopy (SEM) and thermogravimetric analysis (TGA) were used to examine the structure, morphology and thermal properties of the pristine Py-PBI and Py-PBI/NH2-h-BN composite membranes. Moreover, the proton conductivity, chemical stability and ion exchange capacity of the membranes were examined. Structural and morphological analyses confirmed uniform dispersion of NH2-h-BN through Py-PBI membrane. The synergistic system showed significantly enhanced proton conductivity, acid retention, and oxidative stability through strong acid-base and hydrogen-bonding interactions between amine, pyridine, and imidazole nitrogen sites. The 10 wt% addition of NH2-BN to Py-PBI membrane exhibited the highest proton conductivity of 0.31 S cm−1 at 180 °C indicating efficient Grotthuss-type proton transport
Enhanced detergent removal from hospital laundry wastewater using air-assisted electrocoagulation and peroxi-coagulation: Process optimization via Box-Behnken Design and a comparative approach
No full textThis study evaluated the detergent removal efficiencies of laundry wastewater from high bed capacity hospitals using air-assisted electrocoagulation (AA-EC) and peroxi-coagulation (PC) methods for the first time. In the AA-EC process, when aerated and non-aerated conditions were compared, the detergent removal efficiency increased from 18 % to 48 % with air assistance at the optimal airflow rate of 1.25 L/min. The AA-EC and PC processes were optimized using a Box–Behnken Design for detergent removal, yielding optimum conditions of pH 7.7, a current density of 14.23 mA/cm2, and 60 min for AA-EC, and pH 3, a current density of 20.40 mA/cm2, and 55.62 min for PC, with corresponding removal efficiencies of 75.63 % and 98.31 %, respectively. The operating costs were calculated as 3.94 /m3 for the PC process. Sludge characterization and scavenger analyses were used to identify the detergent removal mechanisms in the AA-EC and PC. In addition, QA/QC validation confirmed MBAS measurements were reliably quantified, with LOD and LOQ of 0.007 and 0.023 mg/L, respectively. This study significantly advances the literature by utilizing real hospital laundry wastewater, developing innovative electrocoagulation strategies, and optimizing the processes, thereby offering more effective and practical alternatives to conventional methods
Some new Boole’s inequalities with respect to the middle ending point using k-conformable fractional integral operators via different classes of differentiable functions
No full textThe objective of this study is to formulate a new definition of k-conformable fractional operators and to elucidate their boundedness and semigroup characteristics. Furthermore, we introduce an innovative identity for differentiable h-convex functions that employs k-conformable fractional operators. By employing this identity, we have formulated novel inequalities applicable to bounded functions as well as Lipschitzian functions. In the subsequent section, we utilized h-convexity in conjunction with the power-mean inequality and Hölder’s inequality, respectively
Cerrahi Hemşirelerinin Yapay Zekâya Yönelik Kaygı ve Tutumlarının Değerlendirilmesi: Çok Merkezli Çalışma
No full textAdLU: Adaptive double parametric activation functions
No full textActivation functions are critical components of neural networks, introducing the necessary nonlinearity for learning complex data relationships. While widely used functions such as ReLU and its variants have demonstrated notable success, they still suffer from limitations such as vanishing gradients, dead neurons, and limited adaptability at various degrees. This paper proposes two novel differentiable double-parameter activation functions (AdLU1 and AdLU2) designed to address these challenges. They incorporate tunable parameters to optimize gradient flow and enhance adaptability. Evaluations on benchmark datasets, MNIST, FMNIST, USPS, and CIFAR-10, using ResNet-18 and ResNet-50 architectures, demonstrate that the proposed functions consistently achieve high classification accuracy. Notably, AdLU1 improves accuracy by up to 5.5 % compared to ReLU, particularly in deeper architectures and more complex datasets. While introducing some computational overhead, their performance gains establish them as competitive alternatives to both traditional and modern activation functions