Machine failures in manufacturing industries lead to operational disruptions and increased maintenance costs. Identifying failure mechanisms in electronic machining components is crucial for ensuring reliability and minimizing unexpected breakdowns. This study investigates the use of machine learning, particularly the transformer model, for fault classification via vibration signal data from the CWRU dataset. Model performance was improved by using data preparation methods such as feature extraction, normalization, and noise reduction. The proposed transformer model achieves superior results, with 99.4% accuracy, 99% precision, recall, and F1 score, surpassing those of the MLP (92.4%) and Bi-LSTM (81.7%) methods. The performance evaluation through confusion matrix analysis and learning curves confirmed the model’s ability to detect failure patterns effectively. The results show that deep learning is useful for predictive maintenance, which in turn decreases downtime and increases efficiency in industrial systems. This research shows that AI-driven methods have the capacity to improve operational dependability in contemporary industrial settings via real-time monitoring and problem diagnostics.