MFD-Net: Mechanical Fault Detection Using Deep Feature Extraction for Manufacturing Industries

In manufacturing industries, mechanical faults contribute to nearly 40% of equipment downtime, with annual losses exceeding billions globally. Despite efforts to implement preventive maintenance, many current fault detection systems struggle with real-time accuracy and adaptability across different fault types. This study introduces a novel Mechanical Fault Detection Network (MFD-Net) that enhances fault detection precision and efficiency in manufacturing environments. MFD- Net integrates advanced data preprocessing with a deep learning convolutional neural network (DLCNN) for effective feature extraction, paired with a machine learning-based Categorical Boosting (MLCB) classifier for optimal classification. Leveraging the Stator with Permanent Magnet Synchronous Machines (PMSM) dataset, which includes two classes (Fault and Normal), the model achieved remarkable performance metrics: an accuracy of 99.95%, precision of 99.93%, recall of 99.95%, and F-score of 99.94%. These results affirm MFD-Net's capability to deliver high accuracy and reliability, marking a substantial advancement in MFD technology for complex industrial environments.