Crack faults in the planetary gears are critical since it grows suddenly to the tooth breakage, resulting in whole system loss, but is hard to detect for too many gears in use. To identify this, various sensor signals such as motor current and position (built-in) or vibration (add-on) are acquired for gear with cracks from length 0 to 100% in the interval of 25%. Fault diagnosis procedure is developed, in which a proper signal processing, feature engineering, and classification algorithms are introduced for efficient diagnosis capability.