Reciprocating machinery (such as diesel engines, reciprocating compressors, etc.) has a wide range of applications in the petrochemical, power equipment, and other fields. Conducting vibration monitoring on them is beneficial for fault early warning. The vibration acceleration signals of reciprocating machinery exhibit multi-source impacts in the time domain. Faults such as abnormal valve clearances in diesel engines or valve plate fractures in reciprocating compressors can easily introduce new weak fault impacts and increases in impact amplitude. To adaptively extract fault impact signals, this paper proposes a genetic algorithm optimized variational time-domain decomposition (GAOVTDD) method. GAOVTDD constructs a search objective for the number of decompositions based on the variational time-domain decomposition method, incorporating a genetic algorithm to enhance its parameter adaptability. Meanwhile, the length of the impact extraction window is optimized. GAOVTDD can locate the time-domain positions of impacts, extract each impact component. Finally, the method has yielded favorable test results in detecting abnormal valve clearances in diesel engines and valve plate fracture fault vibration signals in reciprocating compressors.