The fault diagnosis of a gearbox is crucial to ensure its safe operation. Entropy has become a common tool for measuring the complexity of time series. However, entropy bias may occur when the data are not long enough or the scale becomes larger. This paper proposes a gearbox fault diagnosis method based on Refined Time-Shifted Multiscale Reverse Dispersion Entropy (RTSMRDE), t-distributed Stochastic Neighbour Embedding (t-SNE), and the Sparrow Search Algorithm Support Vector Machine (SSA-SVM). First, the proposed RTSMRDE was used to calculate the multiscale fault features. By incorporating the refined time-shift method into Multiscale Reverse Dispersion Entropy (MRDE), errors that arose during the processing of complex time series could be effectively reduced. Second, the t-SNE algorithm was utilized to extract sensitive features from the multiscale, high-dimensional fault features. Finally, the low-dimensional feature matrix was input into SSA-SVM for fault diagnosis. Two gearbox experiments showed that the diagnostic model proposed in this paper had an accuracy rate of 100%, and the proposed model performed better than other methods in terms of diagnostic performance.

In the process of bearing profile detection, an analog displacement sensor is used, but there is noise interference in the process of data acquisition. In order to suppress noise interference, the paper designs three-order and fifth-order anti-aliasing filters according to the actual interference signal source. The resistance value and capacitance value are calculated according to the transfer function. The real object is made by the porous simulation board. The comparison between the input and output signals shows that the fifth-order anti-aliasing filter is better than the third-order anti-aliasing filter. Experiments show that the higher order anti-aliasing filter has a certain suppression effect on the noise signal, but the effect is not obvious. The signal distortion would happen and the system would become more and more complicated. For this reason, the fifth-order anti-aliasing filter is finally selected, and the actual sampling is completed, and the effect reaches the requirements of the detection system.