This study explores a model for identifying failure factors in a gantry crane gearbox using the VG-350 sensor with a 1㎐ sampling rate. We analyzed statistically significant factors through experimental design, with misalignment identified as the most significant. The machine learning results demonstrated high performance, with the data augmentation model achieving cross-validation AUCs above 99.7% and CAs above 97.8% for all factors. This suggests that machine learning can accurately identify even those factors perceived as less significant. The lower performance observed in FFT and PCA pre-processing models is attributed to the low sampling rate. In contrast, the data augmentation model, trained with the same row count as the FFT and PCA pre-processing model, exhibited superior accuracy, emphasizing the potential of low sampling rate sensors for precise long-term fault diagnosis. The study underscores the effectiveness of machine learning in discerning subtle factors and the potential of using low sampling rate sensors for accurate and cost-effective maintenance in the field of gantry crane gearboxes.