In order to improve the fault diagnosis ability of cold rolling mill, comprehensively using the noise reduction technology of Ensemble Empirical Mode Decomposition (EEMD) algorithm and Discrete Wavelet Transformation(DWT), a joint noise reduction technology by EEMD-DWT was designed, and it was ensured that the useful features were fully preserved under the premise of removing noise. Then, the smooth curve signal was obtained by using EEMD-DWT mode for denoising, which showed excellent denoising effect. The application signal results show that when denoising with EEMD-DWT, it includes obvious impact characteristics, which effectively reduces the noise component in the region where the amplitude is near zero, and ensures that the original vibration characteristics are effectively reserved when the noise is fully removed. When DWT or EEMD method is used alone for denoising, a large amount of effective information is lost. When the EEMD-DWT joint noise reduction method is used for processing, the ideal noise reduction effect is achieved, and the useful parameters contained in the signal are fully preserved. Thus, this research can accurately identify different fault degrees of cold rolling mill, which has good theoretical support significance for improving the fault diagnosis level of similar mechanical transmission equipment.