In order to verify the effectiveness of the proposed method, the experimental study is carried out on the zjs50 comprehensive test-bed as shown in the figure. In the experiment, the sensor is installed on the bearing seat of the gearbox, the fault gear is located on the input shaft (gear tooth number G = 32), and the roller bearing model is 1007em (roller number Z = 16, contact angle α = 0, roller diameter d = 6.75 mm, pitch diameter D = 47.5 mm).
In order to verify the recognition effect of the proposed method, the data set contains seven working conditions, as shown in Table 1 (“-” means normal), which are bearing outer ring fault (of), bearing inner ring fault (if), bearing rolling element fault (BF), gear broken and bearing outer ring fault (GoF), gear broken and bearing inner ring fault (if) under the condition of gearbox speed of 600 R / min and fault size of 0.5 mm（ GIF), gear broken and bearing rolling element failure (GBF) and normal state of gearbox (n). In this study, 6000 data points are taken as the data input length, and the length is taken as a sample length to randomly intercept 1000 segments of each state signal. Finally, all samples are divided into training set, verification set and test set according to the ratio of 0.7 ∶ 0.2 ∶ 0.1.