Browsing by Author "Uzel, Hasan"

Now showing 1 - 5 of 5

Acoustic-Based Fault Diagnosis of Electric Motors Using Mel Spectrograms and Convolutional Neural Networks
(Nature Portfolio, 2025) Uzel, Hasan; Ozupak, Yildirim; Alpsalaz, Feyyaz; Aslan, Emrah; Zaitsev, Ievgen
This study presents a comprehensive deep learning framework for diagnosing acoustic faults in electric motors. The framework uses Mel spectrograms and a lightweight convolutional neural network (CNN). The method classifies three motor states, engine_good, engine_broken, and engine_heavyload, based on audio recordings from the IDMT-ISA-ELECTRIC-ENGINE dataset. To prevent data leakage and ensure a robust evaluation, the study employed file-level splitting, session separation, 5-fold cross-validation, and repeated trials. The raw audio signals were transformed into Mel spectrograms and processed through a CNN architecture that integrates convolutional, pooling, normalization, and dropout layers. Quantitative metrics, including THD, spectral entropy, and SNR, further characterize the acoustic distinctions between motor states. The proposed model achieved a test accuracy of 99.7%, outperforming or matching state-of-the-art approaches, such as ResNet-18, CRNN, and Transformer classifiers, as well as traditional MFCC-based baselines. Noise robustness and sensitivity analyses demonstrated stable performance under varying SNR conditions and preprocessing settings. Feature-importance maps revealed that low-frequency regions (0-40 Mel bins) were key discriminative components linked to physical fault mechanisms. Computational evaluation confirmed the model's real-time feasibility on embedded hardware with low latency and a modest parameter count. Though primarily validated on one motor type, external-domain testing revealed strong adaptability. Future work may incorporate transfer learning or multimodal fusion. Overall, the proposed framework provides a highly accurate, interpretable, and efficient solution for real-time motor fault diagnosis and predictive maintenance in industrial environments.
Citation - WoS: 6
Citation - Scopus: 9
Classification of Maize Leaf Diseases With Deep Learning: Performance Evaluation of the Proposed Model and Use of Explicable Artificial Intelligence
(Elsevier, 2025) Alpsalaz, Feyyaz; Ozupak, Yildirim; Aslan, Emrah; Uzel, Hasan
Maize leaf diseases pose significant threats to global agricultural productivity, yet traditional diagnostic methods are slow, subjective, and resource-intensive. This study proposes a lightweight and interpretable convolutional neural network (CNN) model for accurate and efficient classification of maize leaf diseases. Using the 'Corn or Maize Leaf Disease Dataset', the model classifies four disease categories Healthy, Gray Leaf Spot, Common Rust, and Northern Leaf Blight with 94.97 % accuracy and a micro-average AUC of 0.99. With only 1.22 million parameters, the model supports real-time inference on mobile devices, making it ideal for field applications. Data augmentation and transfer learning techniques were applied to ensure robust generalization. To enhance transparency and user trust, Explainable Artificial Intelligence (XAI) methods, including LIME and SHAP, were employed to identify disease-relevant features such as lesions and pustules, with SHAP achieving an IoU of 0.82. The proposed model outperformed benchmark models like ResNet50, MobileNetV2, and EfficientNetB0 in both accuracy and computational efficiency. Robustness tests under simulated environmental challenges confirmed its adaptability, with only a 2.82 % performance drop under extreme conditions. Comparative analyses validated its statistical significance and practical superiority. This model represents a reliable, fast, and explainable solution for precision agriculture, especially in resource-constrained environments. Future enhancements will include multi-angle imaging, multimodal inputs, and extended datasets to improve adaptability and scalability in realworld conditions.
Citation - WoS: 2
Citation - Scopus: 4
Hybrid Deep Learning Model for Maize Leaf Disease Classification With Explainable AI
(Taylor & Francis Ltd, 2025) Ozupak, Yildirim; Alpsalaz, Feyyaz; Aslan, Emrah; Uzel, Hasan
This study presents a hybrid learning model that integrates MobileNetV2 and Vision Transformer (ViT) with a stacking model to classify maize leaf diseases, addressing the critical need for early detection to improve agricultural productivity and sustainability. Utilising the 'Corn or Maize Leaf Disease Dataset' from Kaggle, comprising 4,062 high-resolution images across five classes (Common Rust, Grey Leaf Spot, Healthy, Northern Leaf Blight, Not Maize Leaf), the model achieves an impressive accuracy of 96.73%. Transfer learning from ImageNet, coupled with data augmentation (rotation, flipping, scaling, brightness adjustment), enhances generalisation, while a 20% dropout rate mitigates overfitting. The key advantage of the hybrid model lies in its ability to combine the strengths of MobileNetV2's localised feature extraction and ViT's global context understanding, enhanced by the stacking model's ability to reduce the weaknesses of either model. Explainable AI techniques, including SHapley Additive exPlanations (SHAP), Local Interpretable Model-agnostic Explanations (LIME), and Grad-CAM, provide transparent insights into model decisions, fostering trust among agricultural stakeholders. Comparative analysis demonstrates the model's superiority over prior works, with F1-scores ranging from 0.9276 to 1.0000. Despite minor misclassifications due to visual similarities, the model offers a robust, interpretable solution for precision agriculture.
Hybrid Deep Learning with Attention Fusion for Enhanced Colon Cancer Detection
(Nature Portfolio, 2025) Alpsalaz, Suheyla Demirtas; Aslan, Emrah; Ozupak, Yildirim; Alpsalaz, Feyyaz; Uzel, Hasan; Bereznychenko, Viktoria
This study introduces a hybrid deep learning model integrating EfficientNet-B3 and Vision Transformer with an Attention Fusion mechanism for automated colon cancer detection using the Kvasir endoscopic dataset. The model leverages EfficientNet-B3's strength in capturing fine-grained local textures and Vision Transformer's ability to model global contextual relationships. A multi-head attention-based fusion block harmonizes these features, achieving comprehensive representations and enhanced classification stability. Model optimization was guided by the Matthews Correlation Coefficient (MCC), alongside evaluations of accuracy, F1-score, and Brier Score. Experimental results demonstrate a 96.2% accuracy and an MCC of 0.961, surpassing standalone baselines and existing benchmark architectures. Cross-validation confirmed robust generalization, while Grad-CAM analyses improved interpretability by visualizing salient histopathological regions influencing predictions. Despite slight overfitting tendencies, the model maintained strong performance across all eight image classes. These findings highlight the model's ability to address limitations of single-architecture approaches by combining local and global feature extraction, offering rapid, objective, and reliable diagnostic support. The proposed framework shows significant promise for integration into computer-aided colonoscopy systems, paving the way for enhanced clinical diagnostics and reduced pathologist workload through AI-driven precision medicine.
Optimized ANN-RF Hybrid Model With Optuna for Fault Detection and Classification in Power Transmission Systems
(Nature Portfolio, 2025) Uzel, Hasan; Ozupak, Yildirim; Alpsalaz, Feyyaz; Aslan, Emrah
This study proposes a hybrid machine learning approach that integrates Artificial Neural Networks (ANN) and Random Forest (RF) classifiers, enhanced by Optuna hyperparameter optimization, for fault detection and classification in power transmission networks. The model is trained on a synthetic dataset generated from MATLAB/Simulink simulations of an 11 kV multi-generator system, incorporating three-phase current (Ia, Ib, Ic) and voltage (Va, Vb, Vc) signals under fault scenarios such as line-to-ground (LG), double line-to-ground (LLG), and three-phase symmetrical (LLLG) faults. To address class imbalance, the Synthetic Minority Oversampling Technique (SMOTE) is applied, ensuring balanced representation of rare fault categories. The ANN-RF model achieves superior performance, with 99.8% accuracy, 99.5% precision, and 99.4% recall, consistently outperforming traditional classifiers including K-Nearest Neighbors, Bagging, AdaBoost, and Gradient Boosting. Its effectiveness arises from ANN's non-linear feature extraction, RF's ensemble robustness, and Optuna's hyperparameter tuning, with SMOTE improving detection of rare fault types. Compared with advanced models such as Modified InceptionV3 (98.93% accuracy) and Extreme Learning Machines (99.60% accuracy), the proposed approach provides a balanced trade-off between sensitivity and specificity, offering a reliable solution for fault identification. Nonetheless, challenges in computational efficiency and reliance on simulated data highlight the need for validation with real-world measurements and further optimization for real-time smart grid applications.