The correlation between software failure characteristics and software failure directly determines the predictive performance of the failure prediction model. The extraction of software fault features is crucial for building equipment software fault prediction models, and is an important process to ensure accurate prediction. However, the software fault feature data extraction method is often complicated to use and has no pertinence to the selected software fault feature data, and it takes a lot of time to complete the extraction steps. This paper summarizes software metrics and software defect types based on research at home and abroad, and selects software metrics and software defect types that are suitable for equipment software. Using regular expression technology and CSV technology research the automatic extraction way of software fault features, and finally constructs a fault data set that can be used for software fault prediction models.