Abstract: Rotary vector (RV) reducers are widely used at heavy-duty joints, which directly affect the working accuracy and stability of industrial robots. Different from the general rotating machines, when the industrial robot performs various tasks, the RV reducer at the joint rotates at reciprocating variable speed. The change of rotation speed will cause time-varying effects on characteristic frequency and statistical indicators, which makes it difficult to extract fault information and thus brings challenges for fault diagnosis. A fault diagnosis method for RV reducer based on stationary condition data capturing is proposed, combined with the motion characteristics of the industrial robot RV reducer in this paper. Firstly, a squeezing area constraint factor is introduced to obtain a clear time-frequency spectrogram. Then, ridge extraction based on fast path optimization is performed and the required stationary condition data is obtained by using the sliding window peak-to-peak and mean indicators. Finally, the fault diagnosis is realized by the envelope spectrum analysis of the stationary condition data. The proposed method is verified by the vibration data from RV reducers under the reciprocating variable speed condition. The experimental results show that this method can capture the stationary condition data accurately, overcome the influence of variable speed conditions and successfully extract the fault information.