Abstract: Aiming at reconstruction error and slow calculation speed in the frequency tracking of time domain synchronous average, an improved TSA based on three spline resampling is proposed in this study. Firstly, the algorithm uses the peak search algorithm to collect time-scale signals, secondly, the data are divided into multiple data segments using the obtained time-scale signals, and finally a cubic spline function is used to resample each data segment. Experimental results show that the computational resources required for the proposed algorithm are reduced from the square level to the linear level. The proposed algorithm suppresses noise and uninteresting frequencies, and the performance of fault feature extraction increases 10% compared with that of the conventional one.