Performance assessment of modal parameters identification methods for timber structures evaluation: numerical modeling and case study

The capacity to detect changes in modal properties caused by structural response from those resulting from noises (environment, test conditions, etc.) is a major issue in vibration analysis. For timber-based structures monitoring, high uncertainty ratios in the measurements may prevent efficient modal parameter identification. Therefore, assessment of large timber structures needs to rely on realistic measurement technologies and adequate post-processing tools for improving engineering practices. In this paper, four of the most common identification methods were presented in a benchmark study with respect to the modal parameters identification efficiency of timber elements under different numerical and experimental configurations. These are the Least-Squares Complex Exponential (LSCE) method, the Ibrahim Time Domain (ITD) method, the Frequency-Domain Direct Parameter Identification (FDPI) method and the Least-Squares Complex Frequency-Domain method (PolyMax). All these methods have advantages and disadvantages in terms of computational efficiency, statistical bias, or variance reduction. Therefore, a careful selection of the modal analysis method is a vital step in dynamic data evaluation. Experimental vibration tests combined with a finite element model were conducted. First, a numerical versus experimental efficiency benchmark was performed. Second, the robustness of the selected algorithms for investigating the influences of input waveforms complexity and external noise to the performance of the algorithms was investigated. The robustness of the selected algorithms for estimating the influences of input waveforms complexity on natural frequencies shifts was analyzed. When comparing various algorithms, the simulation and experimental results give a specific direction for the choice of the adapted modal analysis algorithm in timber-based structures engineering applications. The simulation and experimental results show that, for the same experimental data, the PolyMax algorithm has better performance, while the LSCE is worst. Besides, the PolyMax method gives a nonlinear dimensionality reduction algorithm for processing high dimensional information