Automated eigenmode classification for airfoils in the presence of fixation uncertainties

Automated structural design optimization should take into account risk of failure which depends on eigenmodes, since eigenmode shapes determine failure risk by their characteristic stress concentration pattern, as well as by their specific interaction with excitations. Thus, such a process needs to be able to identify eigenmodes with low error rate. This is a rather challenging task, because eigenmodes depend on the geometry of the structure which is changing during the design process, and on boundary conditions which are not clearly defined due to uncertainties in the assembly and running conditions. The present investigation aims to find a proper classification method for eigenmodes of compressor airfoils. Specific data normalization and data dependent initialization of a neural network using principle-component directions as initial weight vectors have led to the development of a classification and decision procedure enabling automatic assignment of proper uncertainty bands to eigenfrequencies of a specific eigenmode shape. Application to compressor airfoils of a stationary gas-turbine with hammer-foot and dove-tail roots demonstrates the high performance of the proposed procedure.