Confidential information, such as the reconnaissance system and mission information in a military drone, mayleak if a crack occurs in the blade during operation and the drone crashes. Major accidents, such as thoseinvolving humans, can happen when a commercial drone crashes. Therefore, minimizing the damage by earlyidentification of blade cracks before a crash is necessary. In this study, the failure mode was analyzed usingthe PHM(Prognostics and Health Management) method, RPN(Risk Priority Number) was identified through theFMECA(Failure Mode Effect and Critical Analysis) of the drone, and differences between the conditions weredetermined after obtaining vibration data under the normal and abnormal conditions identified. In addition,data features were extracted using a statistical method. A method to minimize drone loss was also proposedby predicting the vibration state of the drone according to the length of the blade crack using deep learning