Process classification in a lathe machine using vibration analysis and transfer learning

Industrial lathe machines play a critical role in modern manufacturing, particularly in high-precision industries such as aerospace, automotive, and medical devices. The classification of machining processes is essential for optimizing production, reducing costs, and ensuring process stability. Traditional vibration analysis methods, widely used for machine condition monitoring, often struggle with real-time classification due to environmental noise, varying operational conditions, and sensor placement challenges. This thesis explores a novel approach to lathe process classification by integrating vibration analysis with deep learning techniques. Vibration signals are transformed into frequency-domain representations using the Fast Fourier Transform (FFT), and convolutional neural networks (CNNs) are employed to classify different machining operations. A transfer learning approach leveraging the VGG-19 model is adopted to enhance classification accuracy despite limited data availability. Additionally, data augmentation techniques—including noise injection, intensity scaling, and frequency shifting—are introduced to improve model generalization and robustness. The proposed methodology effectively automates process classification without prior mechanical knowledge, ensuring real-time monitoring and enhancing industrial automation in alignment with Industry 4.0 principles. Experimental results demonstrate the superiority of deep learning-based classification over traditional methods, achieving high accuracy even in the presence of noise and variable machining conditions. This work contributes to smarter manufacturing processes by enabling efficient supervision, optimizing operational efficiency, and laying the groundwork for future research in intelligent machining diagnostics.