The invention provides a fault diagnosis method for parts on a vehicle engine. The method comprises the steps that a wireless acceleration sensor is used for acquiring a time domain vibration signal generated by a to-be-tested part on the vehicle engine, and the time domain vibration signal is forwarded in a wireless mode; and a fault diagnosis apparatus receives the time domain vibration signal sent by the wireless acceleration sensor, and performs three-step processing on the signal: 1, first-period fuzzy C-means clustering filtering is carried out on the signal to eliminate background noises and strong natural periodic pulses; 2, the signal subjected to the first-period filtering is subjected to second-period fuzzy C-means clustering filtering to select out the frequency band with the strongest periodic impact, namely, the frequency band with the most concentrated fault features; and 3, the signal subjected to the double-period filtering is demodulated by adopting a Hilbert envelope, and is compared with a preset fault frequency to identify a fault type. By implementing the fault diagnosis method, the problem of complex wiring in a traditional fault diagnosis technology can be solved, and the fault can be rapidly and effectively diagnosed by improving a fault feature extraction algorithm.