Automated optimization of output only modal parameter identification

The automation of modal parameter identification is important for processing big data with repeatability and without time consuming user interaction. This becomes even more important for complex structures whose modal parameters must be monitored in operation. During modal identification optimal poles have to be found out of many possibilities. For most real world examples with a huge amount of candidate poles the optimal solution cannot be calculated in reasonable time. This paper describes a method for the autonomous selection of optimal parameters within reasonable time constraints. The method modifies the idea of error propagation usually applied in training artificial neural networks. Poles are selected pseudo randomly and an error is calculated based on the difference between the synthesized and measured Cross Power Spectral Densities (CPSDs). This overall error is spread over the selected poles based on the poles influence on the error, i.e. the poles are assigned weights depending on their goodness. This random selection is repeated iteratively until the weights converge. The poles with best weights are selected as the final solution. The method was tested on simulated data and laboratory data from the DLR AIRcaft MODel (AIRMOD) structure and compared to other pole selection methods. The method provides equally good estimations of eigenfrequencies and better damping estimates than the other tested methods