For the sound field of cylindrical or compact sources, the current sound field separation technique (FST) based on statistically optimized near-field acoustic holography (SONAH) uses the planar wave to separate and has not studied the effects of different wave functions. However, the non-conformal problem between the hologram surface and the sound sources will have the risk of accuracy reduction. To solve this problem, a FST for nonplanar sources based on SONAH is proposed. First, according to the characteristic that field prediction of near-field acoustic holography will make errors attenuate, by moving the reconstruction surface to the prediction area, noise immunity of FST is effectively improved in a noisy environment. Subsequently, combining the shape of the sound source, the appropriate conformal wave function is determined and then substituted into the improved method in the first step to separate the free-field sound pressure. Finally, the simulations and the experiments of loudspeakers and a cylindrical experiment platform further proves the effectiveness of the proposed method, which can improve adaptability and reliability of FST in nonplanar sound sources in mechanical systems.