The invention relates to a reciprocating compressor fault diagnosis method based on fine multi-fractal. The reciprocating compressor fault diagnosis method based on fine multi-fractal comprises the following steps of: collecting machine body surface vibration acceleration signals of a sensitive measuring point of a reciprocating compressor; decomposing selected signals by using a parameter optimization time-varying filtering empirical mode decomposition algorithm, and extracting intrinsic-mode function components; calculating the kurtosis values of the obtained intrinsic-mode function components, preferably selecting the main intrinsic-mode function component, performing signal reconstruction, and realizing the noise reduction processing of the collected vibration acceleration signals; performing fine multi-fractal calculation on the reconstructed signal, and describing the structural characteristics and local dynamics behaviors of the signals through a multi-fractal singular spectrum;extracting parameters of the fine multi-fractal singular spectrum, forming eigenvectors of the fine multi-fractal singular spectrum, and diagnosing the fault of the reciprocating compressor; and inputting the eigenvectors of the vibration signals into a recognizer of a support vector machine, and judging the fault type of the vibration signals. The method can describe the fractal characteristicsof the signals in more detail, and can diagnose the fault type more accurately.