Data-Driven Prediction of Blade-Disk Contact Mistuning From Modal Characteristics

The dynamic behavior of bladed disk assemblies is strongly influenced by the contact conditions at the blade–disk interface. Small variations and assembly imperfections lead to different contact footprints at each blade root, producing sector-to-sector variations known as contact mistuning. This work presents a data-driven procedure to identify the unknown contact conditions at each blade root by using the modal information. First, a basis of possible contact patterns is generated by coupling different sets of nodes at the root of a single blade attached to the disk. For each imposed contact configuration, a modal analysis of the blade–disk pair is performed, producing a dataset that links each pattern to its corresponding natural frequencies. To identify an unknown contact condition, the natural frequencies of the blade–disk pair are computed and compared with the database using a first-nearest-neighbor machine learning approach. The closest frequency set provides the most likely contact pattern. Repeating the process for all sectors allows the reconstruction of the full bladed disk with identified contact conditions. The predicted natural frequencies and forced response of this reconstructed assembly match closely with the reference ones, demonstrating that the generated basis of patterns is representative and that the method can accurately recover realistic contact mistuning.