Structural Multimodal Analysis of Bridges Using Adaptive Neuro-Fuzzy Inference System and Artificial Intelligence Metamodels

Abstract

This study investigates using the Adaptive Neuro-Fuzzy Inference System (ANFIS) and machine learning methods for predicting the natural vibration properties of a bridge modeled as a three-degree-of-freedom system. The ANFIS method, which integrates Fuzzy Logic and Neural Networks, was used to analyze a dataset generated from four input and three output variables representing the first three natural vibration periods. A Monte Carlo simulation created a probability density distribution for these variables. The study aims to enhance the precision of bridge response predictions using artificial intelligence, offering a more adaptable alternative to conventional analytical approaches for structural dynamic analysis. The results demonstrate the effectiveness of ANFIS in predicting natural mode periods, with a comparative analysis revealing that ANFIS models, particularly those using triangular membership functions, provide accurate predictions. The findings highlight the importance of dataset partitioning and membership function selection in optimizing ANFIS performance. A comparative evaluation with Artificial Neural Networks (ANN) and the Response Surface Method (RSM) shows that ANFIS and ANN closely match the reference values from algebraic modal analysis, while RSM exhibits some deviations. The study concludes that ANFIS is a viable method for real-world applications in structural engineering, offering a balance between computational efficiency and interpretability.