Research On Damage Identification Of Aeroengine Blades Based On Borescope Data

As an important part of the aircraft,aeroengine blades are directly related to the flight safety.Due to the hard working environment,the blades are easy to cause damage such as cracks,missing materials,dents and so on.The timely and accurate daily detection is of great significance for maintaining the blades in good condition and ensuring the airworthiness of the aircraft.At present,the detection of blades damage based on borescope technology is mainly manual,and the detection results are largely affected by human factors.Therefore,the automatic identification and measurement of blades damage have practical application value for reducing labor intensity and improving detection precision.Considering that the internal structure of the engine is complicated,the traditional image recognition method and the object detection technology in the field of deep learning have their own advantages,but we hope to realize the automatic recognition and positioning of the target by autonomously learning the object features.Therefore,this paper uses the YOLOv3 model and the faster R-CNN(ZF)model to detect damage and localization of the collected borescope images firstly.Then,the YOLOv3-dense model and faster R-CNN model(Improved ZF)are obtained based on the improvement of the blades damage feature extraction capabilities of the above models;Finally,training and testing are performed on the same dataset,and the evaluation results are obtained by comparison of detection precision and recall rate which are compared to select the model with the best performance.After the damage identification and localization is completed,the size of the damage needs to be measured.In this paper,the type of missing material is the research target.Firstly,the characteristic points of the damage measurement were identified;then the contour curve of the damaged blades was scanned by combining the eight-chain code and pseudo-edge removal methods;finally,Hermite second-order interpolation method is used to fit the contour curve to obtain the feature points by using the derivative relation,and the result is compared with the actual value to verify the correctness of the method.Although the damage detection model and measurement method in this paper cannot reach accuracy of 100%,it provides a practical solution for the automatic detection of aeroengine blade defects and theoretical support for related product development.