Abstract: In order to achieve the accurate detection of aero-engine blade damages, a Zeiler-Fergus(ZF) network improvement method based on the faster region-based convolutional neural network(Faster R-CNN) model is proposed. Firstly, the aero-engine blades damage detection is abstracted as the object detection problem, and aiming at the problem, a ZF network is used to detect the damage location and type. Then, some convolutional layers are added into the original ZF network according to the detection effect. Hyper -parameters such as kernel size in the convolutional layers and the stride size in the pooling layers are adjusted, and the added convolutional layers are included in the shared convolutional layers to extract more detailed damage features. Finally, the existing dataset is divided into three sub-datasets, on which the network performance before and after improvedment is experimentally compared. Results show that the mean average precision of the improved ZF network is up to 5.5% compared with that of the original network, which proves that the improved ZF network can predict engine blades damage more accurately

Key words: engine blades, Faster R-CNN, Zeiler-Fergus(ZF) network, damage detection