Ensemble classification method for structural damage assessment under varying temperature

Abstract

Vibration-based damage assessment approaches use modal parameters, such as frequency response functions, mode shapes, and natural frequencies, as indicators of structural damage. Nevertheless, these parameters are sensitive not only to damage but also to temperature variations. Most civil engineering structures are exposed to varying environmental conditions, thus hindering vibration-based damage assessment. Therefore, in this article, a new damage assessment algorithm based on pattern recognition is proposed to scrutinize the healthy state of a structure in the presence of uncertainties such as noise and temperature. The algorithm adopts a combination of couple sparse coding and deep neural network as an ensemble system to assess damage. The proposed method is validated using a numerical model of a truss bridge and experimental data of the I-40 bridge. The results demonstrate its efficiency in the localization and quantification of damages under varying temperature conditions.

Deep neural networks–based damage detection using vibration signals of finite element model and real intact state: An evaluation via a lab-scale offshore jacket structure

“…investigating the natural frequencies for damaged (D4) indicates that there is an increment in the first resonant frequency, which can be caused by uncertainties such as measurement errors and environmental noise during the test. 17 NMA on FE model and eigenvalues method. In the NMA of the offshore jacket structure based on FE model, the natural frequencies and mode shapes of the structure are extracted by eigenvalues method 43 and using the Block Lanczos technique.…”

“…The results showed that CNN was effective in detecting intact, single damaged, and multiple damaged structure with high accuracy. Fallahian et al 17 investigated an ensemble classification method for the assessment of structural damage under varying temperatures. The mentioned method presented a combination of sparse coding and a DNN as an ensemble system.…”

Ensemble learning‐based structural health monitoring by Mahalanobis distance metrics

“…The limitation of their research study is the necessity of measuring the excitation data for obtaining the frequency response functions, which is not suitable for an output‐only SHM problem under ambient vibration. Fallahian and Khoshnoudian 35 presented an ensemble classification method using deep neural network and couple sparse coding for damage detection in the presence of temperature changes. They adopted PCA to extract features from frequency response functions, which served as raw measured data.…”

Damage detection of bridges monitored within one cluster based on the residual between the cumulative distribution functions of strain monitoring data

“…In view of this problem, based on the fact that all the bridges monitored within one cluster bear similar environmental temperature loads during the same monitoring period, the effects of the environmental temperature on strain monitoring data acquired during the same monitoring period are indirectly mitigated. In addition, with a cluster analysis algorithm, [25][26][27] the strain monitoring data for multiple bridges are classified into different classes, and all the monitoring data of each class follow a similar cumulative distribution function (CDF). Then, for each class, the strain monitoring data of all the bridges within the cluster are comparatively analysed to detect damage, which cannot be achieved with monitoring data obtained from a single bridge.…”