Comparative analysis of typical mathematical modelling methods through model updating of a real-life bridge structure with measured data

Mathematical modelling method (MMM) is widely applied in engineering issues. But the performance of MMMs is rarely compared, especially with high dimensional variables. This research focused on the comparison among three mostly used MMMs, i.e., quadratic polynomial (QPMM), kriging (KMM) and neural network (NNMM), based on model updating of a bridge with 13 variables and in-situ data. Firstly, 200 indetermined samples by Latin Hypercube sampling were generated and the relevant responses were computed by finite element model (FEM). Secondly, explicit expressions of responses by MMMs were established. Finally, with the optimization program and explicit expressions, updated variable groups were optimized. From the process and results of FEM updating, it shows that all the MMMs lead to an acceptable result as the discrepancies were reduced sharply. In terms of accuracy, KMM and NNMM are better than QPMM, but in terms of efficiency, KMM is time-consuming.

Automated summary

This research compared the performance of quadratic polynomial, kriging, and neural network mathematical modelling methods in bridge model updating. The results showed that kriging and neural network outperformed quadratic polynomial in accuracy, but kriging was time-consuming in terms of efficiency.