This paper uses wavelet packet analysis technology to decompose the current signal of a motor with broken rotor bars. Experiments show that changes in the wavelet packet decomposition coefficients can indicate whether a rotor bar breakage fault has occurred. 1. Introduction When a rotor bar breakage fault occurs, additional current components with specific frequencies (s is the slip, ν is the supply frequency) will appear in the stator current. However, the absolute amplitude of these components is very small. If Fourier spectrum analysis is performed directly, the leakage of these components due to the picket fence effect may overwhelm the frequency components. Adaptive filtering and Hilbert transform are two methods that can effectively solve the problem of detecting frequency components and are currently the most representative online detection methods for broken rotor bars in asynchronous motors. However, these two methods are only suitable for operating conditions where the motor drives a stable load. When the load fluctuates significantly, satisfactory results cannot be obtained, which affects the accuracy of fault diagnosis. Wavelet packet analysis not only decomposes the low-frequency part of the signal but also the high-frequency part, adaptively determining the resolution of the signal in different frequency bands, making it more precise than wavelet analysis in fault diagnosis. This paper attempts to use wavelet packet analysis to study the fault diagnosis problem of broken rotor bars in motors. 2. Experimental Data Acquisition The current signals of two three-phase squirrel-cage induction motors, one in normal operation and the other with broken rotor bars, were collected and analyzed. The rated power of the three-phase asynchronous squirrel-cage motor is 3KW, and the rated speed is 1430 rpm. Under the conditions of slip s=5.6%, sampling frequency =1000Hz, and sampling points, virtual instruments were used to collect the current signals. Figure 1 shows the stator current signal and its spectrum when the asynchronous motor is running normally, and Figure 2 shows the stator current signal and spectrum when one bar of the asynchronous motor is broken. As can be seen from Figures 1 and 2, under the same load conditions, it is difficult to judge the fault characteristics of the motor with broken bars from the spectrum, and only slight changes can be seen. As can be seen from Table 1, there is no significant change in the values ​​under fault conditions and normal conditions. Therefore, it is difficult to extract fault information from these characteristic parameters. 3. Experimental Analysis The following analysis uses wavelet packet transform to analyze the motor current signal and extracts the wavelet packet decomposition coefficients of each node under the optimal wavelet packet basis. The changes of the wavelet packet decomposition coefficients of each node are analyzed and compared. According to the frequency band division characteristics of wavelet packet decomposition, based on the rotor fault characteristic frequency (44.4HZ, 55.6HZ), the current signal can be decomposed into 5 layers of wavelet packets, and the wavelet basis function is selected as sym4. Under the optimal wavelet packet basis shown in Figure 3, each node corresponds to a wavelet packet decomposition coefficient, and the appropriate wavelet packet decomposition coefficient of the node can reflect the characteristics of the signal. As shown in Figures 4 and 5, it can be seen that there are obvious differences in the wavelet packet decomposition coefficients of nodes (5,4), (5,5) and (5,6) of the normal signal and the fault signal without considering the influence of edge effects. Based on this, the following conclusions can be drawn: (1) The state of the motor can be simply judged based on the wavelet packet decomposition coefficients. (2) When the motor is normal, the wavelet packet decomposition coefficients of nodes (5,4), (5,5) and (5,6) are approximately zero and stable. (3) When the motor bar breaks, the wavelet packet decomposition coefficients at nodes (5,4), (5,5), and (5,6) exhibit certain amplitude oscillations. The nodes under the optimal wavelet packet basis are then reconstructed, and the differences between normal and fault conditions are analyzed based on the waveform of the reconstructed signal. Figures 6 and 7 show that the reconstructed signal shows little difference in the approximate part, but significant differences in the detailed part, especially at nodes (5,4) and (5,6). Therefore, it can be concluded that whether the motor has experienced rotor bar breakage can be easily determined from the wavelet packet decomposition coefficients and the reconstructed waveform. 4. Conclusion This paper decomposes the current signals of the motor under normal conditions and when the rotor bar is broken using wavelet packets, and extracts and reconstructs the wavelet packet decomposition coefficients of each node. The changes in the wavelet packet coefficients can be used to determine whether the motor has experienced a fault. References: [1] Xu Boqiang, Li Heming. A preliminary study on the application of wavelet analysis to online detection of broken bars in the rotor of a squirrel-cage motor [J]. Proceedings of the CSEE, 2001, 21(11): 24-28. [2] Zhang Zhengping, Ren Zhen, et al. A new method for rotor fault detection of motor based on wavelet ridges [J]. Proceedings of the CSEE, 2003, 23(1): 79-82. [3] Wei Yunbing, Huang Jin, Niu Faliang. Fault feature extraction of squirrel-cage asynchronous motor based on wavelet ridges. Journal of Electrical Engineering, 2003, 18(4): 123-126. [4] Cao Zhitong, Chen Hongping, He Guoguang. Fault diagnosis of asynchronous motor based on wavelet reconstruction [J]. Journal of Electrical Engineering, 2002, 17(2): 80-83. [5] Feisco Technology Product R&D Center. Wavelet Analysis Theory and MATLAB 7 Implementation [M]. Beijing: Electronic Industry Press. Author Biography: Li Tao (1983-9-) is a master's student at Henan Polytechnic University, specializing in intelligent control and information processing technology.