In this work, a self-powered fault diagnosis system using vibration energy harvesting was constructed to verify the accuracy of state identification and abnormality detection for a monitored target. A vibration energy harvester and a sensor were attached to an air compressor to be monitored, and the sensor signal was transmitted wirelessly using the energy acquired. Using the supervised machine learning of the k-nearest neighbor algorithm with the vibration sensor signal and the wireless transmission interval, the algorithm was able to identify three states (normal, unstable, and overturned) with a maximum accuracy of 99%. In addition, by using the local outlier factor algorithm as unsupervised learning, it was possible to achieve abnormality detection with a maximum accuracy of 98%. The accuracy of fault diagnosis was improved by analyzing not only the sensor signals but also the wireless transmission interval as machine learning features. It was found that the frequency of wireless transmission by energy harvesting is valuable information for determining the status of the monitored target.