Akademik Veri Yönetim Sistemi
Signal, Image and Video Processing, cilt.19, sa.18, 2025 (SCI-Expanded, Scopus)
Predictive maintenance has become a critical component of modern manufacturing systems, aiming to minimize unexpected downtimes and optimize operational efficiency through early fault detection. For this reason, this study presents a deep learning-based approach for the multi-class classification of machine faults in a manufacturing system. CNN (Convolutional Neural Network)-LSTM (Long Short Term Memory) and CNN-GRU (Gated Recurrent Unit) models are developed to exploit both spatial and temporal dependencies in the data. Bayesian optimization is applied to determine optimal hyperparameters for each architecture then models are validated using tenfold cross validation. The proposed models demonstrated superior classification performance compared to traditional methods such as Decision Trees, Support Vector Machines, and basic Artificial Neural Networks found in the literature. Both CNN-LSTM and CNN-GRU achieved an average F1-scores of 0.96 and overall classification accuracy of 0.98, outperforming classical approaches across all fault types. These results indicate that hybrid deep learning models are highly effective for real-time fault diagnosis and predictive maintenance in smart manufacturing systems.