Non deterministic approaches for the evaluation of the mistune effects on the rotor dynamics

In this job new approaches to the mistune effects evaluation on turbomachine are presented that cannot be classified neither deterministic neither statistical: they are based on the employment of artificial neural networks and of genetic algorithms. In the first case, a net is trained to predict the blade displacements when the rotor is mistuned, correlating the structural response to the introduced mistune pattern. The target is that the net learns the rotor dynamics with an amount of examples sufficient to make fast and reliable predictions on the configurations not included in the learning database. In the second case, a genetic algorithm has been developed to find the structural mistuned configuration that leads to the maximum value of blade vibration amplitude: the target is to perform a smart search of the worst mistuned configuration. In both analyses, a simple test case, representing a twenty blades disc, has been employed and one thousand forced frequency response analyses have been performed. The investigation has demonstrated the applicability of this new possible engineering approaches to the study of system with uncertain properties.