Comparative Analysis of CNN and LSTM For Bearing Fault Mode Classification and the Causality Through Representation Analysis

This study examines how the clarity of frequency-domain characteristics in vibration signals influences the performance of deep learning models for bearing fault classification. Two datasets were used: the CWRU benchmark dataset, which exhibits distinct and easily separable spectral signatures across fault modes, and a custom low-speed bearing dataset in which small defects do not significantly alter the frequency spectrum. To enable a clear and interpretable comparison, we deliberately employed simplified CNN and LSTM ar-chitectures with a single core layer. This design choice allows us to directly attribute per-formance differences to the inherent learning mechanisms of each architecture rather than the complexity of the models. Our representation analysis reveals that LSTM-F achieves the highest accuracy when the dataset contains clearly distinguishable spectral patterns, as in the CWRU case. In con-trast, CNN-S outperforms both LSTM models in the experimental datasets, where fault-induced frequency characteristics are weak or ambiguous. Representation analyses further reveal that LSTM-F relies on consistent frequency-indexed patterns, whereas CNN-S captures more complex time–frequency interactions, making it more robust under low-separability conditions. These findings demonstrate that the optimal deep learning architecture for bearing fault classification depends on the degree of frequency separability in the data. LSTM-F is pref-erable for severe faults with distinct spectral features, while CNN-S is more effective for minor defects or systems exhibiting complex, weakly discriminative frequency behavior.