FreqCondNorm Transformer for Cross-Domain Predictive Maintenance

FreqCondNorm uses a FiLM-style frequency-conditioned normalization layer to unify heterogeneous time-series data spanning 1 Hz to 100 kHz. Pretrained on five public datasets, it achieved 99.2% accuracy on CWRU and 82.1% zero-shot accuracy on MFPT, though it did not improve remaining useful life predictions.