Pipeline of turbine blade defect detection based on local geometric pattern analysis

Blade defects such as scratch and deformation always cause performance degradation or even failure of gas turbine, and furtherly threat energy safety. Therefore, defect detection of individual blades is important in gas turbine maintenance. Currently, defect detection is mostly manually implemented, so it is necessary to design an automatic method to reduce labor costs. However, most available detection methods require CAD models or prior shapes of blades, which might not always be acquirable for gas turbine users. Besides, those methods are not suitable for practical use because of their high demands for detection equipment. In this paper, a simple but efficient 3D vision-based defect detection process using point clouds is proposed to detect scratches and deformation accurately. First, sliding sampling window is used to reduce the computation burden in each detection process. Second, a pipeline of detection algorithms based on underline surface analysis is proposed to extract local geometry features and predict potential defects. Last, a filtering algorithm is introduced to reduce false detections. The whole process is carried on with real defective blades, and its effects are both intuitively and quantitatively evaluated. The results show that a multi-step detection process based on FPFH is able to detect both scratch and deformation accurately, which will be suitable for practical application.