Artificial intelligence-based damage identification in linear structural elements

Resumo

This paper presents a structural health monitoring framework for identifying damage in linear steel structural elements (compressed bars), using artificial intelligence techniques, specifically hybrid modeling, machine learning, and digital twin methodologies. The framework integrates a hybrid physics-based and data-driven model with supervised machine learning methods to construct a digital twin. Damage is modeled as a localized reduction in axial stiffness (EA) and the structural parameters considered include the bar’s length, axial stiffness (EA), boundary conditions, and material density. representing typical degradations. Various damage intensities (5%, 10%, 15%, and 20%) and locations, as well as different noise levels, are simulated to evaluate the robustness of the framework. The governing equations of motion of the healthy structure, discovered through hybrid modeling, are used to simulate the system's dynamic response under each damage scenario. These simulations generate a dataset to train machine learning classifiers covering both healthy and damaged states. The resulting digital twin can detect the presence, location, and severity of damage, supporting engineering decisions. The recall metric achieved 83.33% for 5% damage, 89.33% for 10%, 92.47% for 15%, and 93.73% for 20% damage. For 20% damage with noise, up to 13% noise level still allowed damage identification with 62%–73% accuracy.