The present invention relates to a precise predictive maintenance method for a driving unit, the method comprising: a first base information collection step (S10) of collecting change information of energy size in accordance with a time measured in a normal driving state of a driving unit by dividing into a peak period and a mean period; a second base information collection step (S20) of collecting change information of energy size in accordance with a time measured in a driving state of the driving unit before a malfunction of the driving unit is generated, by dividing into a peak period and a mean period; a setting step (S30) of setting a mean fault of the mean period on the basis of the information collected in the first and second base information collection steps (S10, S20); and a detection step (S40) of collecting change information of energy size in accordance with a time measured in a real-time driving state of the driving unit by dividing into a peak period and a mean period, and detecting the driving unit as being in an abnormal state when an energy value of the collected mean period exceeds the mean fault of the mean period set in the setting step, such that driving information of the driving unit in a normal state and driving information of the driving unit, which is shown before a malfunction is generated, are collected, and peak/mean fault and peak/mean warning of the peak period and the mean period are set on the basis of the collected information, and then, driving information collected in real-time through a driving of the driving unit is compared with the set peak/mean fault and peak/mean warning of the peak period and the mean period and when a condition suspected to show abnormality of the driving unit is satisfied, a warning is provided so as to induce the driving unit to be modified and replaced at the right time, thereby preventing a huge loss caused by malfunction of the driving unit. In addition, various detection conditions are presented in order to search for various abnormal signs which can be generated in a driving unit, and when the detection conditions are satisfied, a warning is provided to a user such that the various abnormal signs generated from the driving unit can be easily detected and excellent reliability for the detection results can be ensured.