In order to solve the problems of inconsistent data distribution and insufficient samples in the predictive modeling of the bearing health status of the injection molding machine, this paper proposes a deep learning algorithm based on the transfer learning of the bearing feature data domain adaptive processing, using the VGG-16 network to predict The training model is used as the basic model of migration learning to fine-tune the high-level parameters of the VGG model without changing the low-level network parameters, so as to achieve accurate prediction of the differential data generated in different environments. Comparative experiments show that the migration learning method proposed in this paper has better prediction effects, and the optimized model has a higher prediction rate than the unoptimized model, which further verifies that the model proposed in this paper has stronger feature learning Generalization.