This paper primarily explores the potential of denoising diffusion probabilistic models in generating vibration signals for planetary gear systems, producing data under two different health states and three rotating speed conditions. Results demonstrate that the diffusion model can effectively reconstruct time-domain waveforms and frequency-domain characteristics of the original signals, attributed to constraints from time-domain loss and FFT loss. However, envelope demodulation analysis reveals a critical limitation: the diffusion model currently fails to learn implicit signal modulation phenomena inherited in original vibration signals. This shortcoming highlights the need for future improvements to enable the model to reproduce modulation effects widely present in mechanical vibration signals.

denoising diffusion probabilistic models,data generation,planetary gearboxes,vibration signals