Abstract

Improving gas turbine design and repair through advanced manufacturing requires sophisticated computer-aided inspection tools that can help us ensure an accurate assessment of component defects. Among key elements of gas turbines are airfoil blades. To guarantee the functionality of the blades, tight tolerances are applied to their ideal basic geometry. Blade manufacturing requires a high-accuracy inspection to verify the conformity of the manufactured blades to the specified geometric tolerances and the blade repair necessitates high-accuracy inspection to carry out reliable defect detection. The practice of evaluating the geometric quality of a turbine blade requires the definitions of several pre-specified blade sections and the coordinate inspection data for these airfoil sections. A coordinate measuring machine (CMM) with a contact probe has been traditionally used to collect the coordinate data. The CMM data acquisition process is lengthy as the data is collected through probe contacts with the part surface. This slow data acquisition makes the inspection process a bottleneck of the blade production and makes it almost impossible to acquire holistic inspection data for repair planning. Blade manufacturers and MRO companies would rather use the high-speed optical 3D scanning systems that can complete the inspection of the blade surface in seconds, with hundreds of thousands or even millions of data points. The collected point cloud via 3D scanning is, however, distributed all over the blade surface rather than at the desired specific sections. In fact, no point of the point cloud is located exactly on the sectional planes. The desired sectional data has to be extracted from the nearby data points. In this talk, I will present an algorithm to extract the desired sectional inspection data from the nearby data points of the sectional plane by projecting the points along a curvilinear trajectory. This is achieved in this work by fitting a local quadric surface to the neighboring points of the point of interest and projecting the data points along the fitted surface curved contour. A systematic approach to establishing a balanced set of neighboring points is employed to avoid bias in fitting the local quadric surface as well as to guide the selection of points to be projected onto the sectional plane. The algorithms were validated using both synthetic point clouds and real scanned point clouds. The developed software realizes accurate and fully automatic extraction of sectional inspection data of the blades from the massive, unorganized 3D scanned point cloud data. The extracted sectional data also has significant application in the reconstruction of the damaged blade model for generating the repair volume for repairing the blades via direct energy deposition (DED) process. Another challenge of the blade inspection is the alignment of inspection data to the nominal CAD model. The point cloud captured by the scanner is located in the measurement coordinate system, while the CAD model is in the design coordinate system. To compare the two models for inspection, they must be brought to a common coordinate system. Traditionally, the best matching by Iterative Closest Point (ICP) algorithm is performed to transform the measurement coordinate frame in order to have both CAD and the scanned point cloud in a common coordinate frame. The ICP algorithm iteratively minimizes the distance from the points in the point cloud to the corresponding points on the CAD model in a least-squares sense, a process that averages out the geometric errors in different regions on the blade surface for the sake of minimizing the least-squares objective function. This best matching scheme leads to wrongful inspection results, especially in the case of damaged blades in repair applications. In this talk, an algorithm will be proposed that can fine-tune the alignment in order to avoid the averaging out issue. The validation of the algorithm using numerical and experimental point clouds will also be presented with several case studies on damaged blades.