Abstract: Active magnetic bearing is a new type of bearing that uses electromagnetic force to support the rotor, which has the advantages of no lubrication, small energy consumption, high speed and wide adjustable speed range, and is more and more used in various types of rotating equipment. Auxiliary bearing is an important part and the last safety barrier of active magnetic bearing. After an overload or drop accident, whether to replace the auxiliary bearing is an important issue for the user to consider. If the auxiliary bearing is in a serious state of damage, it may fail in the next fall accident, resulting in significant losses to the unit. Due to the high cost of disassembling the machine and the difficulty to assess degradation directly, an online degradation assessment method is urgently needed. This article first carried out the highspeed drop tests for active magnetic bearing until a certain type of auxiliary bearing failed completely. Before each highspeed drop, a lowspeed drop experiment was performed and the displacement signal of the rotor was recorded. Only the state of the auxiliary bearing varies between lowspeed drop experiments, so auxiliary bearing degradation can be evaluated using lowspeed drop data. First, the displacement signal was noisereduced using complementary ensemble empirical mode decomposition and spearman correlation coefficients, and the shortterm Fourier transform was used to obtain a speed drop curve. Then, the time domain characteristic parameters of the signal were calculated, and the four principal components containing most of the information were selected by principal component analysis. Based on the four principal components above, multiscale permutation entropy, Mahalanobis distance and multiscale fuzzy entropy were applied to study the degradation pattern of auxiliary bearings as the number of highspeed drop test increases. The concept of last safe fall is that the protective bearing can withstand one fall before the last safe fall; after the last safe fall, the protective bearing will be fail on the next fall. In the experiment, the low-speed signal after the last safe fall was the third lowspeed drop. Compared with the difference between the second fall signal and the first fall signal, the method should reflect that the difference between the third fall signal and the first fall signal significantly increases. The results show that multiscale fuzzy entropy can reflect the performance degradation of auxiliary bearings due to highspeed drop tests. The Mahalanobis distance can be used to assist assessment, but the effect is not as good as the multiscale fuzzy entropy. Multiscale permutation entropy is not suitable for this application. Therefore, it is recommended to use multiscale fuzzy entropy to evaluate the degradation of auxiliary bearings and use Mahalanobis distance for auxiliary judgment.