Prediction of Blade–Disk Contact Mistuning Patterns Using Probabilistic and Machine Learning Methods

Bladed disks with detachable blades are widely used in turbomachinery, where the blade–disk joint critically influences the dynamic response. Manufacturing tolerances and assembly uncertainties introduce non-uniform contact at the dovetail interface, giving rise to contact mistuning. Although pressure-sensitive films provide accurate measurements of local contact conditions, obtaining such data for every blade root becomes impractical for large assemblies. This work introduces a generalized framework to statistically characterize contact mistuning in families of bladed-disk assemblies using only a limited subset of experimentally measured contact patterns. Two reconstruction strategies are proposed. The first is a probabilistic method that derives node-level contact probabilities from the measured subset and generates complete synthetic assemblies by sampling full contact maps. The second is a machine-learning approach based on Principal Component Analysis, which identifies dominant spatial features of the measured patterns and synthesizes fully predicted, physically coherent contact maps. Consequently, three families of bladed-disk assemblies are generated: one using experimental patterns, one using probabilistic predictions, and one using PCA-based predictions. Their dynamic behavior is evaluated by comparing the mean natural frequencies, allowing identification of the most suitable reconstruction method. Finally, for the selected approach, the maximum and minimum envelopes of the forced-response functions are used to characterize the admissible vibration range of a family of bladed disks with statistically similar contact patterns