Key Smart Technologies For Fluorescent Magnetic Particle Inspection Of Turbine Blades

With the significant improvement and popularization and application of major high-end equipment such as gas turbines,aero-engines,and naval deep submarine equipment in our country,its turbine blades and other components show a trend of complexity,large-scale and integration,and their service conditions are increasingly extremely demanding,which puts forward stringent requirements for their safety and economic operation and maintenance testing.In recent years,the rapid and efficient non-destructive evaluation and intelligent detection of surface damage defects of turbine blades have become increasingly urgent,which has become the key to equipment operation and maintenance.However,there is still a lack of fast,accurate and efficient non-destructive testing methods for turbine blades,which rely heavily on theoretical research and technological breakthroughs in automated,intelligent,and lowcost non-destructive testing.Fluorescent magnetic particle detection technology has the advantages of fast detection speed,low cost,and no need for special protection,and has been widely used in the manual maintenance and inspection of complex curved components in the process of manufacturing and service.However,for the highefficiency detection requirements of blade automation and intelligence in the operation and maintenance process,a lot of research work still needs to be carried out in the aspects of fluorescent magnetic particle imaging detection,intelligent defect recognition and robot-assisted automatic detection.Therefore,in view of the defects such as cracks and corrosion in the process of manufacturing and service of turbine blades,three types of defects such as pitting corrosion,linear cracks and network cracks were prepared in this paper,the research on intelligent defect recognition algorithms of SSD,YOLOv3,Faster RCNN+FPN and Retina Net has been carried out.Since most of the failures of turbine blades originate from the surface,first of all,the different failure mechanisms of turbine blades are continuously analyzed,and the types and characteristics of surface defects of turbine blades in their life cycle are extracted to prepare typical surface defects such as pitting,linear cracks and reticular cracks;secondly,an automatic fluorescent magnetic particle inspection equipment was used and experimental research was carried out,and a large number of fluorescent magnetic particle inspection images were obtained based on the automatic image acquisition system;Defect dataset suitable for object detection algorithms.Finally,the location and category of defects detected by fluorescent magnetic particle were marked,and a defect data set suitable for target detection algorithm is made.For the detection and classification of surface defects of turbine blades,the application research of SSD and YOLOV3 one-stage target detection algorithm is carried out.The detection results are compared,and the recognition effects of the two algorithms on pitting,linear cracks and mesh cracks are analyzed.The results show that the average detection accuracy of the detection model based on the YOLOv3 algorithm reaches 95.77 %,which is better than the 87.2% detection accuracy of the SSD algorithm model.The YOLOv3 algorithm can not only accurately detect surface defects,but also achieve accurate defect classification.In order to improve the detection accuracy and efficiency of actual workpiece surface defects,Faster RCNN algorithm and Retina Net algorithm based on FPN multiscale feature fusion are further studied in surface defect recognition and classification.The study found that the detection performance of the Faster RCNN algorithm and the Retina Net algorithm for typical defects such as pitting,linear cracks,and reticular cracks is better than that of the YOLOv3 algorithm.The average detection accuracy of the Faster RCNN algorithm is 96.74%,and the Retina Net algorithm is 97.85%.Meanwhile,FPN network structure can effectively improve the detection algorithm’s recognition of the characteristics of fluorescent magnetic particle defects,and the Faster RCNN algorithm with FPN structure can significantly improve the recognition accuracy of typical defects to 98.62%.In summary,this paper carries out comparative analysis and research on the detection accuracy and detection speed of the above four intelligent algorithms for three typical defects of turbine blade pitting,linear crack and network crack.Based on FPN multi-scale feature fusion detection algorithm,the actual defect detection of turbine blades and other complex curved surface components is carried out.It is found that Faster RCNN+FPN and Retina Net target detection algorithms can accurately detect the location of defects and achieve accurate classification,and obtain good practical application effect.It is of great significance for intelligent detection and maintenance of fluorescent magnetic particles in turbine blades.