We propose a phase-aware complex-spectrogram autoencoder (AE) for preprocessing raw vibration signals of rotating electrical machines. The AE reconstructs normal components and separates fault components as residuals, guided by an input-phasor phase-orthogonality regularization that defines parallel/orthogonal residuals with respect to the local signal phase. We use a U-Net-based AE with a mask-bias head to refine local magnitude and phase. Decisions are based on residual features—magnitude/shape, frequency distribution, and projections onto the normal manifold. Using the AI Hub open dataset from field ventilation motors, we evaluate eight representative motor cases (2.2–5.5 kW: misalignment, unbalance, bearing fault, belt looseness). The preprocessing yielded clear residual patterns (low-frequency floor rise, resonance-band peaks, harmonic-neighbor spikes), and achieved an area under the receiver operating characteristic curve (ROC-AUC) = 0.998–1.000 across eight cases, with strong leave-one-file-out generalization and good calibration (expected calibration error (ECE) ≤ 0.023). The results indicate that learning to remove normal structure while enforcing phase consistency provides an unsupervised front-end that enhances fault evidence while preserving interpretability on field data.

This study presents an LSTM autoencoder-based vibration anomaly detection system developed using normal operating data for the servomotor driving the ball screw in a CNC machine tool. The system comprises wireless vibration data acquisition and storage, preprocessing, time series feature generation, model training, and identification of potential anomalies based on reconstruction error (RE). Input features, model architectures, and loss configurations were compared using normal operating data from CNC-A. Vibration RMS and Moving Average were selected as the final input features, together with a loss function incorporating an additional RMS reconstruction error term. The existing standardization parameters and trained models were applied without further adjustment to 29,001 test sequences generated from normal operating data from CNC-B, which had not been used for training or model selection. Across three random seeds, the coefficient of determination (R²) for Vibration RMS was 0.938569 ± 0.000682 across all time steps and 0.931685 ± 0.005062 at the final time step. Potential anomalies were identified using the 99 th percentile (P99) and 99.5 th percentile (P99.5) of the RE distribution derived from the CNC-A normal validation data. The proportion of CNC-B test sequences exceeding P99 ranged from 2.121% to 2.262%, and 567 sequences (1.955%) were identified by all three models. This study demonstrates the implementation and application of a system that transfers models trained on normal operating data and validation-derived thresholds to a different CNC machine tool to quantify reconstruction discrepancies and identify sequences requiring further examination.