<p>The separation of approximately periodic weak impulses remains a significant challenge in fault diagnosis for rotating machinery. To address this issue, an enhanced algorithm, referred to as improved time synchronous averaging by evaluating energies of local maximum, is proposed. Firstly, the reasons for the limited extraction effect of existing algorithms such as time–frequency representations and Fourier transform are analyzed: the instantaneous amplitude amplification and periodicity of fault signals generated by steady speed rotating machines. To overcome these limitations, the time synchronous averaging (TSA) technique is introduced. In the TSA approach, the local maximum within a specific interval replaces the signal value at a fixed sequence number. This replacement serves to suppress the impact of aperiodicity caused by minor jitter errors or slippage between contacting parts. Furthermore, a masking operation is performed synchronously on the signal to further enhance the visibility of the fault component. To mitigate the potential influence of large-amplitude noise points, an evaluation of energy intensity rankings of the local maxima is proposed as a solution. Additionally, a criterion is introduced to determine whether the filtered impulses originate from a fault. The proposed method is then applied to both the original signal and its corresponding negative value, in order to address cases where the local maximum amplitudes of the fault components are negative. Finally, the proposed algorithm is validated using both simulation signals and two sets of experimental data. The results demonstrate that the proposed method is effective in extracting impulses from signals with a low signal-to-noise ratio, showcasing its robustness and potential for practical applications in fault diagnosis.</p>