In order to detect the occurrence and location of the faults in the life test of the gearbox bench as soon as possible and to avoid causing damage to the test equipment and personnel, a gearbox online fault diagnosis system is developed. The system uses acceleration sensors to detect the vibration signals of the box and realizes real-time monitoring and fault early warning of the gearbox test through time-frequency domain analysis. The system performs time-frequency domain analysis on time-domain signals by using ZFFT refined spectrum method to process frequency-domain signals, and analyzing the characteristics of typical fault signals to realize fault type determination. The test results show that the system can realize gearbox failure prediction and accurately identify typical failures.