Comparative Analysis of Deep Learning Models for Vibration Signal-based Fault Diagnosis of High-speed Spindle Motor

This study investigates vibration signal-based fault diagnosis for high-speed spindle motors operating at 10,000 60,000 RPM, which exhibit vibration patterns distinct from conventional lower-speed motors. The analysis focuses on vibration characteristics, signal preprocessing, and deep learning model performance. Vibration signals are transformed into the frequency domain using the fast fourier transform (FFT), while time-domain features such as root mean square (RMS) and peak-to-peak amplitude are extracted. To address data distributional variations, normalization techniques, including MinMaxScaler, StandardScaler, and RobustScaler, are applied before model training. Subsequently, multiple architectures are evaluated, including convolutional neural networks (CNN), long short-term memory networks (LSTM), and hybrid variants such as CNN-LSTM, LSTM-CNN, bidirectional LSTM (BiLSTM), CNN-BiLSTM, and BiLSTM-CNN. Experimental results highlight that CNN-LSTM and CNN-BiLSTM yield the highest classification accuracy with efficient inference times. These findings demonstrate the efficacy of combining advanced signal preprocessing with deep learning for fault diagnosis, offering valuable contributions to the development of robust predictive maintenance strategies, enhancing the safety, reliability, and efficiency of industrial systems employing high-speed spindle motors.