In order to identify sudden failures in industrial robot gearboxes, a fault identification method based on time-frequency domain characteristics of Acoustic Emission (AE) signals is proposed, using the pencil-lead breaking method to simulate fault sources. Firstly, a DS5-16C AE instrument is used to build a gearbox AE acquisition platform, with the resonant frequency of the AE sensor set at 150 kHz and a sampling frequency of 3 MHz. Then, according to ASTM AE testing standards, sudden failures are simulated on the surface of a normally operating gearbox using pencil-lead break. Finally, based on calculating time-domain indicators such as the absolute mean, pulse index, and margin index of the raw AE signals, the time-frequency characteristic of the AE signals are further obtained through continuous wavelet transform, thus proposing a time-frequency domain identification method for sudden failures in gearboxes. The research results show that although the characteristic frequency of the same AE signal on the gearbox and the support are 170 kHz and 100 kHz respectively, the pulse index and margin index of the AE signals at different locations can significantly identify sudden failures.