Sürdürülebilir Kalkınma Amaçları

• 7
• 1
• 1
• 1

• MakaleAçık Erişim2026Scopus SJRQ1/r/munzur/1397
  A Metric-Driven Evaluation Framework for Remaining Useful Life Prognosis with Quantified Uncertainty

  Vashishtha, Govind; Chauhan, Sumika; Ertarğın, Merve

  This paper introduces a novel metric-driven evaluation framework for Remaining Useful Life (RUL) prognosis in rotating machinery, featuring robust uncertainty quantification. Accurate RUL prediction is vital for optimizing maintenance and preventing failures, but existing methods often struggle with complex nonlinear degradation or lack reliable uncertainty estimates. Our proposed framework integrates a probabilistic Deep State Space Model (DSSM) with a variational inference approach to model complex, non-linear degradation trends and inherent aleatoric uncertainty. A key innovation is the use of the Slime Mold Algorithm (SMA) for efficient hyperparameter optimization, ensuring maximum accuracy. Furthermore, an online adaptation mechanism, governed by a heuristic reinforcement learning agent, allows the model to continuously update its knowledge and adapt to concept drift in real-time. Experimental validation on the IMS bearing dataset demonstrates superior RUL prediction accuracy, evidenced by the lowest Root Mean Square Error (RMSE) of 8.1829 cycles, and a PICP of 0.59416. This dual capability makes the framework highly suitable for real-world predictive maintenance, enhancing safety and reliability.

  • 
  İncele
• MakaleAçık Erişim2026Scopus SJRQ1/r/munzur/1382
  A smartphone-based vibration dataset for induction motor fault diagnosis under different speed and load conditions

  Ertarğın, Merve; Orhan, Ahmet; Yıldırım, Özal; Chauhan, Sumika; Vashishtha, Govind

  Induction motors are foundational components across industrial applications, valued for their inherent robustness, operational simplicity, and cost-efficiency. Maintaining their reliable function is paramount for a wide range of mechanical and electrical systems. However, they are prone to various mechanical and electrical faults, such as bearing defects, rotor issues, and voltage imbalances, which can significantly impair their performance and reliability. This study presents a novel vibration dataset for induction motor fault diagnosis, uniquely acquired using a smartphone-based inertial sensor rather than conventional industrial accelerometers. Vibration signals were recorded along three orthogonal axes (gx, gy, gz), alongside gravity-compensated acceleration components (guserx, gusery, guserz), enabling detailed analysis of both raw and gravity-free vibration characteristics. Data were collected under diverse conditions, including healthy operation and several fault types, across varying rotational speeds and load states. The dataset features long-duration vibration recordings sampled at 100 Hz, suitable for both time-domain analysis and window-based feature extraction. Its inclusion of multiple operating speeds and load conditions is ideal for studying the impact of operational variability on fault signatures. By leveraging low-cost and readily accessible smartphone sensors, this dataset supports practical and accessible vibration data acquisition for supporting the development, benchmarking, and validation of data-driven fault diagnosis methods. This resource is expected to significantly advance research in condition monitoring of induction motor, particularly for machine learning and signal processing applications using vibration data.

  • 
  İncele
• MakaleAçık Erişim2026Scopus SJRQ2/r/munzur/1051
  A Unified and Interpretable Benchmark of Classification Models for DGA-Based Power Transformer Fault Diagnosis

  Doğan, Kıvanç; Ertarğın, Merve; Çalışkan, Abuzer; Bakay, Ayşenur

  Power transformers are among the most critical components of electric power transmission and distribution systems, and unexpected failures can lead to substantial economic losses and prolonged power outages. Dissolved Gas Analysis (DGA) is the most widely used diagnostic technique for transformer fault diagnosis and involves interpreting gases dissolved in insulating oil. However, conventional interpretation methods, such as the Rogers ratio, the Doernenburg ratio, the IEC 60599 ratio method, and the Duval Triangle, rely heavily on expert knowledge, may produce inconsistent diagnoses for the same oil sample, and may fail to provide a diagnosis in certain cases. In this study, 12 classification models were evaluated using the publicly available Power Transformers Fault Detection and Diagnosis (FDD) and Remaining Useful Life (RUL) dataset and a unified evaluation protocol. Model performance was assessed using Accuracy, Balanced Accuracy, and Macro-F1 score, while model interpretability was investigated through Shapley Additive Explanations (SHAP) analysis. The results showed that ensemble tree-based methods achieved the best overall performance. LightGBM and Random Forest both attained an Accuracy of 0.969 and a Macro-F1 score of 0.924, while LightGBM further achieved the highest Balanced Accuracy of 0.940. XGBoost exhibited the most stable performance under cross-validation. SHAP analysis revealed that engineered relative concentration features, particularly the CO/H2 ratio and the combined gas ratio, were among the most influential features for fault classification. These findings demonstrate that, for datasets of this scale, ensemble tree-based models combined with well-designed features provide strong and interpretable performance for imbalanced DGA-based fault diagnosis, highlighting the effectiveness of feature-based ensemble learning for small- to medium-sized datasets.

  İncele
• MakaleAçık Erişim2025/r/munzur/6001
  A comparison of cnn and transformer-based models for algal species classification

  Ertarğın, Merve; Şimşek, Gökçe Kendirlioğlu

  In this study, the classification of Spirulina platensis, Tetraselmis suecica, and Porphyridium cruentum algal species was conducted using RegNetY-008, ConvNeXt-Tiny, and ViT-Tiny models. The experiments were performed on a small and imbalanced dataset, where all models achieved an accuracy of 100%. To assess model performance, training and validation losses were analyzed, confirming the absence of overfitting. Furthermore, Grad-CAM was applied to visualize the decision-making processes of the models. The results indicate that although all models demonstrated high accuracy, the ViT-Tiny model exhibited the most interpretable visualizations by effectively focusing on algal cells. This study highlights the capability of transformer-based models to achieve high performance even on small and imbalanced datasets, offering a promising alternative for biological image analysis.

  İncele
• MakaleAçık Erişim2025/r/munzur/5999
  Predicting daily energy consumption in ev charging stations using tree-based models

  Ertarğın, Merve

  Electric vehicles (EVs) have emerged as a cornerstone in the transition toward sustainable transportation, offering a cleaner alternative to conventional fossil fuel-based mobility. Their integration into modern transportation systems not only supports global decarbonization efforts but also requires robust energy management solutions to address the growing demands on charging infrastructure. Ensuring efficient energy consumption forecasting is vital for optimizing EV charging station operations and maintaining grid stability. This study aims to predict the daily energy consumption of an EV charging station using a public dataset. Key features such as charging duration, vehicle count, mean and median connection time were utilized for the prediction. Both the original dataset and its z-score-cleaned version were employed to evaluate the models' performance under different data conditions. Decision Tree (DT), Random Forest (RF), Extreme Gradient Boosting (XGBoost) and Light Gradient Boosting Machine (LightGBM) models were implemented, with hyperparameter optimization conducted via GridSearchCV. The results revealed that all four models achieved high accuracy in predicting daily energy consumption across both datasets, with R2 values exceeding 0.97. Specifically, for the z-score-cleaned dataset, LightGBM demonstrated the best performance with the lowest RMSE (37.35 kWh) and MAE (28.93 kWh), as well as the highest R2 value (0.9727) followed by XGBoost, RF and DT.

  • 
  • 
  • 
  • 
  AtıfOpenAlex1

  İncele
• BildiriYalnız künye2025Scopus SJRQ4/r/munzur/2309
  A Hybrid Machine Learning Architecture for Acoustic Fault Diagnosis in Electric Motors: Offline and Online Modes via Mobile Phone

  Yordanov, Nikolay; Zhilevski, Marin; Mikhov, Mikho; Ertargin, Merve

  This paper proposes a hybrid ML architecture for non-invasive acoustic fault diagnosis in electric motors and machines using a.NET MAUI cross-platform mobile app for short audio capture (5-20s), audio-to-image transformations and dual-mode inference: embedded ML.NET/ONNX models for offline preliminary detection and Azure ML-hosted deep CNN/RNN for online probabilistic multi-fault analysis. The primary objective of this paper is to demonstrate the benefits of the proposed architecture and to show the specific tools which can be used for implementation. Architectures (embedded, cloud, hybrid) are evaluated by listing main advantages and disadvantage, highlighting hybrid approach that combines offline resilience, reduced latency and keeping the possibility for complex analysis by using online hosted model. Sample mobile application design is added to showcase the proposed user experience.

  İncele
• BildiriAçık Erişim2025/r/munzur/1908
  Health Status Classification of Electric Motors Using CNN-Based Models and SDP Images Under Varying Noise Conditions

  Ertarğın, Merve; Yordanov, Nikolay; Zhilevski, Marin

  The increasing application of electric motors across various industrial sectors requires effective monitoring and early diagnostics to prevent potential failures. This study explores the use of Convolutional Neural Network (CNN)-based models together with Symmetrized Dot Pattern (SDP) sound representations for classifying the health status of electric motors under different noise conditions. Acoustic data from a brushless DC motor were transformed into SDP images, which were then used to train CNN models. The dataset included recordings of motors in “Good”, “Broken” and “Heavy Load” conditions, captured under various noise environments such as pure, talking, white noise, atmospheric, and stress test conditions. The classification tasks were conducted under three conditions: assessing motor health status, evaluating both motor health status and noise types, and excluding the stress test noise type for a balanced dataset. The results demonstrated that the CNN models achieved high accuracy rates in classifying motor health status, with the Custom CNN model performing best in simpler tasks and MobileNet excelling in more complex scenarios. The study highlights the feasibility of using SDP images with CNN-based models for fault classification in motors and suggests future research directions for improving classification accuracy through advanced feature extraction techniques and multimodal data representations.

  • 
  AtıfScopus1

  İncele
• MakaleAçık Erişim2025Scopus SJRQ1/r/munzur/1379
  Automated fault classification of asynchronous motor using mobile phone accelerometer and Parallel Residual CNN-GRU

  Ertargin, Merve; Orhan, Ahmet; Yildirim, Ozal; Gurgenc, Turan

  In this study, a fault detection system that classifies motor faults using deep learning without requiring any sensor connections inside or on the motor setup is proposed. The focus is on the electrical and mechanical faults of asynchronous motors, which are widely used in the industry. A mobile application developed using Flutter can collect vibration data using the three-axis accelerometer sensor in smartphones. By placing a smartphone with the application installed on the motor for which fault detection is desired, vibration data on three axes (x, y, z) is collected to train deep models. Subsequently, test data is used to identify the fault status and class of the electric motor. Different conditions and scenarios were created to test the adequacy of smartphone accelerometer sensors in detecting motor faults. Fault classification with vibration data obtained under different motor operating speeds in loaded and unloaded conditions achieved accuracy rates of over 99%.

  • 
  AtıfScopus18·OpenAlex15

  İncele
• MakaleAçık Erişim2025Scopus SJRQ2/r/munzur/986
  Species classification of microalgae using a CNN-based deep learning approach under optimal cultivation conditions

  Simsek, Gokce Kendirlioglu; Ertargin, Merve; Pezzolesi, Laura; Pistocchi, Rossella; Yildirim, Ozal; Cetin, A. Kadri

  Microalgae possess significant potential in a wide range of applications due to their valuable bioactive compounds. They are utilized in biofuel production to reduce dependence on fossil fuels, in wastewater treatment for the biological removal of heavy metals and pollutants, in carbon capture for mitigating climate change, and in the pharmaceutical and nutraceutical industries as supplements. As their applications expand, the accurate and efficient identification of microalgal species becomes increasingly important. Traditional classification methods are time-consuming and rely heavily on expert knowledge. The main aim of this study is to develop a reliable, fast, and expert-independent deep learning-based approach for the classification of microalgae species using microscopic images. In this context, deep learning techniques, specifically convolutional neural network (CNN)-based models were employed to classify microalgal species. Four widely used pre-trained CNN architectures (ResNet152, DenseNet201, MobileNetV2, and EfficientNetB0), along with a custom-designed CNN, were implemented. The models were trained and tested on a labeled dataset consisting of microscopic images of Chlorella vulgaris, Scenedesmus acutus, and Haematococcus pluvialis. The classification models achieved accuracy rates ranging from 96.87 % (Custom CNN) to 100 % (DenseNet201), demonstrating the potential of CNN-based approaches in automating and improving microalgae species identification.

  AtıfScopus5·WoS5·OpenAlex5

  İncele
• MakaleAçık Erişim2025Scopus SJRQ2/r/munzur/904
  A Comparative Study of Time–Frequency Representations for Bearing and Rotating Fault Diagnosis Using Vision Transformer

  Orhan, Ahmet; Yordanov, Nikolay; Ertarğın, Merve; Zhilevski, Marin; Mikhov, Mikho

  This paper presents a comparative analysis of bearing and rotating component fault classification based on different time–frequency representations using vision transformer (ViT). Four different time–frequency transformation techniques—short-time Fourier transform (STFT), continuous wavelet transform (CWT), Hilbert–Huang transform (HHT), and Wigner–Ville distribution (WVD)—were applied to convert the signals into 2D images. A pretrained ViT-Base architecture was fine-tuned on the resulting images for classification tasks. The model was evaluated on two separate scenarios: (i) eight-class rotating component fault classification and (ii) four-class bearing fault classification. Importantly, in each task, the samples were collected under varying conditions of the other component (i.e., different rotating conditions in bearing classification and vice versa). This design allowed for an independent assessment of the model’s ability to generalize across fault domains. The experimental results demonstrate that the ViT-based approach achieves high classification performance across various time–frequency representations, highlighting its potential for mechanical fault diagnosis in rotating machinery. Notably, the model achieved higher accuracy in bearing fault classification compared to rotating component faults, suggesting higher sensitivity to bearing-related anomalies.

  AtıfScopus29

  İncele