Abstract: Metal magnetic memory testing (MMMT) can detect early damage represented by stress concentration, and has broad application prospects in the field of damage detection of ferromagnetic components. However, the MMMT signals are very weak and nonlinear, and it is prone to have no warning or false warning applied the MMMT is applied in engineering application. The machine learning method is suitable for data processing and analysis in MMMT for its good self-learning ability and adaptability for characteristic patterns finding. This paper focused on the application of machine learning methods in metal magnetic memory testing. The traditional feature extraction methods of MMMT signals were discussed, and the research status of machine learning methods in feature extraction, quantitative recognition and residual life prediction of metal magnetic memory testing was analyzed. The difficulties and future development direction of machine learning methods in the field of metal magnetic memory testing were discussed.