Abstract: A fault diagnosis method based on the combination of wavelet analysis and SOM network is proposed. Wavelet analysis technology is used to acquire tractor gear fault feature signals, and then SOM neural network is used to model and diagnose the acquired fault data. The experiments show that this method can effectively improve the accuracy of gear fault diagnosis. Keywords: wavelet analysis; SOM network; fault diagnosis 1 Introduction Gears are the most commonly used transmission components for changing rotational speed and transmitting power. They are an important component of mechanical equipment and also a component prone to failure. Their operating status has a great influence on the overall performance of the machine. Once a failure occurs, it often leads to serious consequences. If the fault can be diagnosed and eliminated in time, the accident can be avoided, and the reliability of machine operation can be improved, and the utilization rate of the machine can be further improved. At present, gear fault diagnosis in China is still mainly based on manual analysis. With the development of diagnostic technology, relying on computers and software to carry out diagnosis is the general trend of mechanical equipment fault diagnosis technology [1]. This paper attempts to introduce wavelet analysis technology into the analysis of gear fault signals, use it to extract the characteristic signals when gear faults occur, and use SOM neural network to perform fault diagnosis modeling on the acquired signals in order to achieve better results. 2 Signal extraction based on wavelet transform When a gear fault occurs, the measured signal contains non-stationary components or time-varying components, and these components often directly reflect the running state of the gear. Wavelet transform has the characteristics of time-frequency domain localization [2], which is suitable for feature extraction of non-stationary signals and time-varying signals. In particular, continuous wavelet transform can extract the components of the required frequency band that change with time in the signal. It is not only suitable for the feature extraction of equipment steady-state signals, but also suitable for the process of state change, making the signal feature extraction very effective. For details, please click: Application of wavelet neural network in gear fault diagnosis