This paper provides a detailed review of vibration-based fault detection methods used for CNC (Computer Numerical Control) spindle bearings, which are essential in machining operations. The review looks at traditional signal processing methods like Fast Fourier Transform (FFT) and wavelet analysis, along with new machine learning models such as Long Short-Term Memory (LSTM) networks, Convolutional Neural Networks (CNNs), and hybrid approaches. It also examines newer diagnostic frameworks like digital twins and predictive maintenance systems for their potential to improve fault diagnosis and prognosis. Key findings indicate that machine learning models significantly boost fault detection accuracy, but challenges remain for real-world use in industries. These challenges involve the need for real-time prediction, combining multi-modal sensor data, and making advanced methods scalable in operational settings. The paper highlights these research gaps and suggests future directions to address these issues. This includes validating techniques in real environments, merging multi-sensor data, and developing IoT-enabled fault detection systems.
Patil et al. (Thu,) studied this question.