Endüstri Mühendisliği Bölümü Koleksiyonu
Permanent URI for this collectionhttps://hdl.handle.net/20.500.11779/1942
Browse
21 results
Search Results
Article Robust HMM-Based Remaining Useful Life Estimation Using a Ridge-Regularized EM Algorithm(MDPI, 2026-02-18) Kucukdag, Halime Beyza; Kirkil, Gokhan; Hekimoglu, MustafaEstimating the remaining useful life (RUL) of engineering systems is crucial for maintenance planning and the reliability of complex mechanical units. Accurate RUL predictions support timely interventions and help to prevent unexpected failures. This study proposes a statistically robust framework that models degradation signals up to the end of life using a hidden Markov model (HMM) with a simple-failure structure and an absorbing terminal state. The proposed method estimates state-dependent linear emission parameters and transition probabilities using a ridge-regularized expectation-maximization (EM) algorithm. The ridge penalty stabilizes slope estimates under limited data, while a robust Huber-based scale estimator reduces sensitivity to outliers in the sensor-derived health indicator. RUL is computed as a weighted expected time to absorption, combining transient-state survival characteristics with smoothed posterior-state probabilities obtained via the forward-backward algorithm. This yields a low-variance state-aware estimator that preserves the probabilistic structure of the HMM. Simulation studies show that the proposed ridge-regularized EM significantly reduces parameter variance and improves predictive accuracy compared with the baseline weighted least squares EM (WLS-EM). A real-data case analysis demonstrates further improvements in RUL estimation accuracy and smoother, more reliable prediction trajectories. Overall, the framework provides a robust and interpretable approach for practical prognostics applications.Article Empowering Electric Vehicle Adoption: Innovative Strategies for Optimizing Charging Station Placement Based on Projected Demand(Wiley, 2025-01-01) Cekyay, Bora; Kabak, Ozgur; Ozaydin, Ozay; Isik, Mine; Toktas-Palut, Peral; Topcu, Y. Ilker; Ulengin, FusunElectric vehicles (EVs) are pivotal for reducing transportation-related emissions; however, the lack of adequate charging infrastructure remains a significant barrier to their widespread adoption. This study presents a comprehensive methodology for optimizing EV charging station placement. It combines a gravity model, scenario analysis, and mixed-integer linear programming (MILP) to ensure a thorough and robust approach. The model aims to maximize accessibility by ensuring both path-level and overall system demand coverage across diverse scenarios, providing reassurance about the validity of the findings. The methodology is tested on the Bursa-& Idot;zmir motorway in Turkey, a strategic intercity route with rapidly growing EV penetration. Results reveal that the optimal configuration involves locating charging stations in seven of the nine service areas. This allocation secures a minimum path coverage ratio of 0.903, meaning 90.3% of the route is covered by charging stations, and an overall demand coverage ratio of 0.935, indicating that 93.5% of total demand is covered across all scenarios. A sensitivity analysis further shows that increasing the network to 45 chargers elevates reachability levels to above 97%, indicating the infrastructure scale required for reliable service quality. The findings underscore the practical applicability of the proposed framework, providing policymakers and infrastructure planners with robust, data-driven guidance for charging network expansion. By integrating demand forecasting with resilient optimization, this study advances both methodological and empirical insights, empowering the audience to make informed decisions for sustainable EV adoption.Article A Comparative Study of Branch-And Algorithms for Vehicle Routing With Time Windows and Waiting Time Costs(Wiley, 2026-02-09) Michelini, Stefano; Kucukaydin, Hande; Arda, YaseminBranch-and-price is one of the most commonly used methodologies for solving routing problems. In recent years, several studies have investigated advanced labeling algorithms to solve the related pricing problem, which is usually a variant of the elementary shortest path problem with resource constraints. Such algorithms include efficient techniques such as decremental state space relaxation, ng-route relaxation, and several hybridizations of these two relaxation methods. In this study, we compare the performance of these labeling algorithms in a branch-and-price framework when applied to the vehicle routing problem with time windows and a variant of this problem in which waiting times have a linear cost. For the latter problem, we also propose an appropriate label structure with associated resource extension functions and dominance rules. We perform these comparisons by using a rigorous methodology, which consists of parameterizing several features of these algorithms, obtaining a good parameter configuration for each algorithm, and analyzing the performance of these configurations on benchmark instances. In order to obtain good configurations, we make use of irace, which is a tool for automated parameter tuning, while statistical tests are used for performance comparisons. Our results show that a class of hybrid algorithms with certain features based on ng-route relaxation outperforms all the others.Article Anomali Tespiti ve Suistimal Önleme: Telekomünikasyon Sektöründe Bir Uygulama(2025-12-19) Koç, Utku; Bulut, Özgür; Özalanyalı, ÖzgeBu çalışmada, telekom sektöründeki satış kanallarında ortaya çıkan anomalilerin tespitine ve suistimal olabilecek durumların engellenmesine yönelik istatistiksel bir yöntem geliştirilmiştir. Yöntemin geliştirilmesi ve test edilmesi sürecinde 371 farklı satış kanalına ait 9 aylık tüm satış bilgileriyle 340 binden fazla gerçek veri noktası kullanılmıştır. Anomali tespitinde en çok karşılaşılan engellerden biri yöntemin anomali olarak işaretlediği noktaların gerçekten anomali olup olmadığının teyit edilmesindeki zorluktur. Her bir kanalın kendi kontrol grubunu oluşturduğu bu çalışmada ise yöntemin anomali olarak işaretlediği noktaların gerçekten bir anomali olup olmadığı ilgili iş birimi tarafından değerlendirilmiş ve teyit edilmiştir. Her bir satış kanalı için günlük güven aralıkları ayrı ayrı hesaplanmış ve bu aralığın dışına çıkan durumlara hızlı tepki veren bir yöntem kullanılarak olası suistimallerin önüne geçilmiştir. Elde edilen bulgular, önerilen yöntemin anomali tespitinde başarılı olduğunu ve satış süreçlerindeki potansiyel suistimallerin önüne geçtiğini ve dolayısıyla müşteri memnuniyetini artırdığını göstermektedir. Geliştirilen yöntem yüksek performans ve ölçeklenebilirliği sağlamak için çoklu mimari yapısında uygulamaya alınmıştır. Geliştirilen yöntem ve uygulama, güvenlik ve veri bütünlüğü konularında da önemli avantajlar sunmaktadır. İlgili iş birimlerinin hızlı ve etkili kararlar alabilmesi, organizasyonun genel risk yönetimi stratejisine büyük katkı sağlamaktadır. Bu sayede, potansiyel tehditler zamanında tespit edilerek işletmenin güvenlik standartları korunmakta ve sürdürülebilir bir operasyonel çevre yaratılmaktadır. Ayrıca, projenin teknik yapısı anomali tespit sisteminin sürekli iyileştirilmesi hem yazılımın performansını artıracak hem de daha ileri düzeyde veri analizi imkanı sunacaktır. Sonuçlar, telekom şirketlerinin stratejik karar alma süreçlerine önemli katkılarda bulunarak rekabet avantajı sağlamalarına yardımcı olmaktadır.Article Citation - WoS: 3Citation - Scopus: 5Qubo Formulations and Characterization of Penalty Parameters for the Multi-Knapsack Problem(IEEE-Inst Electrical Electronics Engineers Inc, 2025) Guney, Evren; Ehrenthal, Joachim; Hanne, ThomasThe Multi-Knapsack Problem (MKP) is a fundamental challenge in operations research and combinatorial optimization. Quantum computing introduces new possibilities for solving MKP using Quadratic Unconstrained Binary Optimization (QUBO) models. However, a key challenge in QUBO formulations is the selection of penalty parameters, which directly influence solution feasibility and algorithm performance. In this work, we develop QUBO formulations for two MKP variants-the Multidimensional Knapsack Problem (MDKP) and the Multiple Knapsack Problem (MUKP)-and provide an algebraic characterization of their penalty parameters. We systematically evaluate their impact through quantum simulation experiments and compare the performance of the two leading quantum optimization approaches: Quantum Approximate Optimization Algorithm (QAOA) and quantum annealing, alongside a state-of-the-art classical solver. Our results indicate that while classical solvers remain superior, careful tuning of penalty parameters has a strong impact on quantum optimization outcomes. QAOA is highly sensitive to parameter choices, whereas quantum annealing produces more stable results on small to mid-sized instances. Further, our results reveal that MDKP instances can maintain feasibility at penalty values below theoretical bounds, while MUKP instances show greater sensitivity to penalty reductions. Finally, we outline directions for future research in solving MKP, including adaptive penalty parameter tuning, hybrid quantum-classical approaches, and practical optimization strategies for QAOA, as well as real-hardware evaluations.Article Citation - WoS: 1Facial Emotion Recognition Using Residual Neural Networks(Aves, 2024-11-08) Kırbız, SerapFacial emotion recognition (FER) has been an emerging research topic in recent years. Recent automatic FER systems generally apply deep learning methods and focus on two important issues: lack of sufficient labeled training data and variations in images such as illumination, pose, or expression-related variations among different cultures. Although Convolutional Neural Networks (CNNs) are widely used in automatic FER, they cannot be used when the number of layers is large. Therefore, a residual technique is applied to CNNs and this architecture is named residual neural network. In this paper, an automatic facial emotion recognition method using residual networks with random data augmentation is proposed on a merged FER dataset consisting of 41,598 facial images of size 48 × 48 pixels from seven basic emotion classes. Experimental results show that ResNet34 with data augmentation performs better than CNN with a classification accuracy of 81%.Article Citation - WoS: 1Citation - Scopus: 3Evaluation of Learning Management Systems Using Interval Valued Intuitionistic Fuzzy-Z Numbers(Anadolu Üniversitesi, 2023-10-01) Ucal Sarı, İrem; Sergi, Duygu; Sari, Irem UcalThe use of online education tools has increased rapidly with the transition to distance education caused by the pandemic. The obligation to carry out all activities of face-to-face education online made it very important for the tools used in distance education to meet the increasing needs. In line with these needs, radical changes have occurred in the learning management systems used in distance education. Therefore, in this study, it is aimed to determine the features that the systems used in distance education should have and to compare the existing systems according to these features. For this purpose, a novel fuzzy extension, interval valued intuitionistic fuzzy Z-numbers, is defined for modeling uncertainty, and AHP and WASPAS methods using proposed fuzzy numbers are developed to determine the importance of decision criteria and compare alternatives.Article Citation - WoS: 9Citation - Scopus: 15Mixcycle: Unsupervised Speech Separation Via Cyclic Mixture Permutation Invariant Training(IEEE, 2022) Karamatlı, Ertuğ; Kırbız, SerapWe introduce two unsupervised source separation methods, which involve self-supervised training from single-channel two-source speech mixtures. Our first method, mixture permutation invariant training (MixPIT), enables learning a neural network model which separates the underlying sources via a challenging proxy task without supervision from the reference sources. Our second method, cyclic mixture permutation invariant training (MixCycle), uses MixPIT as a building block in a cyclic fashion for continuous learning. MixCycle gradually converts the problem from separating mixtures of mixtures into separating single mixtures. We compare our methods to common supervised and unsupervised baselines: permutation invariant training with dynamic mixing (PIT-DM) and mixture invariant training (MixIT). We show that MixCycle outperforms MixIT and reaches a performance level very close to the supervised baseline (PIT-DM) while circumventing the over-separation issue of MixIT. Also, we propose a self-evaluation technique inspired by MixCycle that estimates model performance without utilizing any reference sources. We show that it yields results consistent with an evaluation on reference sources (LibriMix) and also with an informal listening test conducted on a real-life mixtures dataset (REAL-M).Article Citation - WoS: 5Citation - Scopus: 12Predicting Cash Holdings Using Supervised Machine Learning Algorithms(Springer, 2022-05-18) Özlem, Şirin; Tan, Ömer FarukThis study predicts the cash holdings policy of Turkish firms, given the 20 selected features with machine learning algorithm methods. 211 listed firms in the Borsa Istanbul are analyzed over the period between 2006 and 2019. Multiple linear regression (MLR), k-nearest neighbors (KNN), support vector regression (SVR), decision trees (DT), extreme gradient boosting algorithm (XGBoost) and multi-layer neural networks (MLNN) are used for prediction. Results reveal that MLR, KNN, and SVR provide high root mean square error (RMSE) and low R2 values. Meanwhile, more complex algorithms, such as DT and especially XGBoost, derive higher accuracy with a 0.73 R2 value. Therefore, using advanced machine learning algorithms, we may predict cash holdings considerably.Article A Lot-Sizing Problem in Deliberated and Controlled Co-Production Systems(Taylor and Francis, 2022-02-11) Kabakulak, Banu; Ağralı, Semra; Taşkın, Z. Caner; Pamuk, BahadırWe consider an uncapacitated lot sizing problem in co-production systems, in which it is possible to produce multiple items simultaneously in a single production run. Each product has a deterministic demand to be satisfied on time. The decision is to choose which items to co-produce and the amount of production throughout a predetermined planning horizon. We show that the lot sizing problem with co-production is strongly NP-Hard. Then, we develop various mixed-integer linear programming (MILP) formulation of the problem and show that LP relaxations of all MILPs are equal. We develop a separation algorithm based on a set of valid inequalities, lower bounds based on a dynamic lot-sizing relaxation of our problem and a constructive heuristic that is used to obtain an initial solution for the solver, which form the basis of our proposed Branch & Cut algorithm for the problem. We test our models and algorithms on different data sets and provide the results.
- «
- 1 (current)
- 2
- 3
- »