Abstract: Accurately collecting various signals and extracting features is the key to achieving automatic control of mining equipment. The vibration signal of motor bearings is one of the important signals for automatic identification of coal and rock in mining equipment. It is severely affected by environmental noise and component friction under complex working conditions, resulting in blurry signal characteristics and affecting the signal characteristics of mining equipment. This study proposes a joint wavelet denoising method based on Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(ICEEMDAN)and Multiscale Permutation Entropy(MPE)optimized by a genetic algorithm, and evaluates its effectiveness through signal-to-noise ratio, mean square error, and denoising error ratio. The research results show that compared to traditional methods such as EEMD-MPE, CEEMDAN-MPE, and ICEEMDAN-MPE, the proposed joint wavelet denoising method has the highest signal-to-noise ratio, minimum mean square error, and maximum denoising error ratio in simulated signals and mechanical equipment bearing vibration datasets. This method not only exhibits excellent noise suppression capabilities, but also effectively preserves the feature information that characterizes the mechanical state. By studying the motor bearing signals of coal mining equipment, it can provide preliminary research for studying the signal characteristics of the entire mining equipment, and lay a certain foundation for the subsequent automatic recognition of coal and rock and the automation and intelligence of working condition equipment.