Abstract: Under normal conditions, bearings often operate at time-varying speed, which makes fault diagnosis technology of bearings at constant speed impossible to apply. Aiming at this problem, a multiple time-frequency curve extraction method for bearing fault diagnosis based on improved fast path optimization algorithm is proposed. Using the learning characteristic of fast path can effectively prevent frequency jump and extract time-frequency curve to represent ridge line in time-frequency domain expression more accurately. In addition, applying fast path optimization to iteration of time-frequency domain expression can effectively extract multiple time-frequency curves. Then the ratio of average curve to time-frequency curve is compared with fault characteristic coefficient. Realize fault diagnosis. The experimental results show that the method can effectively realize bearing fault diagnosis under unknown time-varying speed.