Parameter identification from spectral measurements using mechanical prior

Abstract

This paper presents a systematic approach for identifying physically plausible parameters of multi-mass drivetrains based on measured torsional natural frequencies. The approach combines a Lanczos-based reconstruction of the Jacobi matrix with a constrained optimization that incorporates prior model information and physical relations, such as symmetries arising from the repeated use of identical components. The method is applied to an industrial turbine test bench developed by ABB and CEL for Rolls-Royce, where significant discrepancies between the calculated and the measured TNFs were observed during commissioning. The proposed identification algorithm successfully detects and localizes inconsistencies between the prior model and the real system. Technical verification of the identified deviations confirms the results and demonstrates that the methodology provides a reliable and transparent tool for model validation, commissioning support, and fault diagnosis in industrial rotor systems.