In response to the problems of frequency aliasing and large positioning errors in the fault location of mining belt conveyor rollers, a method for locating the sound source of mining belt conveyor roller faults based on function deconvolution optimization is proposed. The cross-spectral matrix function and point spread function within conventional fault acoustic source localization imaging functions are reconstructed and analyzed through this methodology. By optimizing the sound source localization function, the resolution of the fault signal is improved, effectively enhancing the positioning accuracy and ensuring accurate identification of the fault location of the roller. Firstly, calculate the peak cross spectral matrix function caused by sidelobes in the initial positioning sound source image, and use the corrected cross spectral matrix function after removing the peak cross spectral matrix as the target cross spectral matrix function for reconstruction optimization. Secondly, perform eigenvalue decomposition on the reconstructed cross spectral matrix function, calculate its eigenvectors and eigenvalues, and construct unitary and diagonal matrices. Finally, hadamard power operations are applied to the acoustic source localization function, and Hadamard root operations of corresponding orders are performed on the cross-spectral matrix, yielding the reconstructed and analytically derived point spread function. This process leads to the formation of a functionally deconvolved optimization-based fault acoustic source localization equation system. The accelerated greedy iterative update strategy is introduced to obtain fault sound source localization information. To address the issue of frequency aliasing in faulty sound sources, an improved sparrow algorithm optimized variational mode decomposition and wavelet improved multi-scale dual threshold multi-level joint filtering denoising method are adopted to enhance the extraction of fault characteristic frequencies in sound sources. This joint filtering approach not only effectively suppresses noise components but also enhances the extraction of characteristic fault frequencies from the acoustic source signal.The experimental results show that compared with the traditional method for locating the sound source of roller faults, the noise sidelobe interference in sound source imaging maps can be effectively removed and the localization accuracy of idler fault sound sources is improved by this method; The optimized algorithm achieved a fault localization accuracy of 94.42%, with a 88.25% increase in signal-to-noise ratio and a 56.06% increase in peak signal-to-noise ratio for multi-stage joint denoising. It also increased the computation speed by 3.8 times, demonstrating good accuracy and efficiency gains.