Predictive maintenance of planetary gearboxes using FFT and machine learning technique

Abstract

Planetary gearboxes are widely used in manufacturing processes, and non-destructive assessment is becoming increasingly important for monitoring their state. We outlined a fine-tuned random decision tree (FT-RDT) in this study for classifying and fault-finding the gearbox via signals generated by vibrations. This approach concentrates on the identification of worn gears, consequently distinct classes—healthy gears, ringed gears containing damaged tooth faces, and planetary gears featuring damaged tooth faces—were established. Each of the categories consists of 150 specimens, divided into two separate sets of 50 specimens for the testing data and 100 specimens for the data used for training. The Fast Fourier Transform (FFT) was used to convert the temporal signals to frequencies. The next step was to gather 24 statistical characteristics from frequency data. The retrieved characteristic was fed into the fault classification procedure (FT-RDT). Combining these methods yields rates of classification accuracy across train and test data of 92.75% and 91.50%, demonstrating the excellent reliability and capability of the problem identification solution that is created.