Bearing is a vulnerable component in food processing machinery and packaging machinery. Bearing status is related to the safe operation of machinery. In this study, bearing fault diagnosis method based on EMD and AR models was adopted to solve the problem of bearing fault diagnosis, ensuring the maximum quality and efficiency of food production. A bearing fault dynamics model integrating outer ring, inner ring, and rolling were built to obtain the bearing vibration signal during the operation of the bearing. Therefore, the bearing fault diagnosis method based on EMD and AR model was adapted to decompose the bearing vibration signal into IMF components. After constructing AR model, it was used to construct the bearing comprehensive determination distance between the autoregressive parameters and the variance of the residuals, and the fault diagnosis of the bearing was completed according to the state corresponding to the minimum comprehensive discrimination distance. The analysis and research showed that the outer ring fault and the inner ring fault were diagnosed with the normal bearing. Moreover, the error of the diagnosis result was small with low complexity, and this helped to achieve the goal of high efficiency diagnosis of bearing fault.