A large amount of labeled data is a necessary condition for model training of intelligent fault diagnosis methods. However, this condition is difficult to be met in some industrial application scenarios. It is difficult to collect enough labeled data, especially in the fault state. This limits the industrial application of the intelligent fault diagnosis method. To solve this problem, an intelligent fault diagnosis method for mechanical equipment based on feature knowledge transfer is proposed. The characteristic knowledge contained in the sufficient amount of labeled data collected by the experimental equipment or other related equipment is transferred to the intelligent model deployed by an industrial field device. The feature knowledge transfer of monitoring data between different mechanical devices is completed. In this way, the intelligent fault diagnosis of mechanical equipment under unlabeled data can be achieved. First, the proposed method constructs the onedimensional deep convolutional neural network to realize the depth mapping from the raw vibration signal to the mechanical equipment fault category. Then, the domain adaptation constraint is added to the deep convolutional neural network to realize the deep transfer adaptation of the feature knowledge between different mechanical equipment monitoring data. Finally, the health status of the mechanical equipment is identified by a fully connected neural network. To evaluate the effectiveness of the proposed method, the transfer fault diagnosis experiment is implemented by the monitoring data collected from the bearings of the two mechanical devices under different performance conditions. Experimental results show that the proposed method realizes the transfer and adaptation of the monitoring data feature knowledge between different devices. Compared with the traditional intelligent diagnosis method, the proposed method improves the recognition rate of migration fault diagnosis between two data sets more than 20%.