Vibration from rotating machines with impulse-like fault sources are studied.

MED convolution fix is proposed to remove discontinuity-related spurious impulses.

MOMEDA is proposed as a non-iterative deconvolution problem to deconvolve impulses controlled by a target vector.

Spectrums of impulse train targets can be solved simultaneously to study the fault level according to the fault period.

On our setup, results show that AR model preprocessing before deconvolutions had no significant effect on fault detection result.