924 publications from this institution
With recent ground-breaking advances, machine learning (ML) has been applied widely in numerous fields in this day and age. However, because of the application of backpropagation algorithms based on gradient descent (GD) techniques, the network of ML may be trapped in local minima, especially if its starting point is not on the same side of the global best or the network contains too many local minima. This drawback may reduce the accuracy and effectiveness of ML. To transcend these limitations of ML, numerous researchers have employed algorithms based on global search techniques to eliminate initial local minima of the network by looking for a beneficial starting point. Nevertheless, those solutions are only valid under certain circumstances when the network only contains a few local minima and they are distributed on the same side. With complex problems such as structural health monitoring (SHM), the network always exists of different error surfaces with numerous widely distributed local minima. The approach of the selection of a good starting position for the network may no longer be useful. Therefore, this paper proposes a novel machine-learning based on an evolutionary algorithm, namely Cuckoo search (CS) to solve the local minimum problem of ML in the most radical way. CS algorithm based on the global search technique is employed to work parallel with ML during the process of training the network. This win-win approach has both advantages of GD techniques (fast convergence) and stochastic search techniques (avoiding being trapped in local minima). The core idea of the proposed method is recapped as follows: (1) ML using the GD technique is first applied to speed up convergence; (2) if the network gets stuck in local minima, CS with global search capability is applied to assist particles in escaping from local minima; (3) the GD technique is applied again to increase the convergence speed. Steps 2 and 3 are repeated until the target is achieved. Additionally, to handle the large amount of data used to train the network, we also apply a vectorization technique for the data of the objective function, which significantly reduces the computational cost. This is another contribution of this work. To assess the performance of the proposed approach, both numerical and experimental models with different damage scenarios are considered. The results showed that the proposed approach completely outperforms CS, ML, and other hybrid ML in terms of accuracy and considerably reduces calculational costs compared to CS.
No abstract is provided for this article.
No abstract is provided for this article.
No abstract is provided for this article.
As one of the key elements of the Brite EuRam BE96-3157 SIMCES Project a series of tests were conducted on the bridge Z24 in Switzerland. The aim of the tests was to serve as basis for a feasibility proof and the development of procedures for damage detection in civil engineering structures consisted of two main parts. First the influence of the environment on the dynamic characteristics was measured by a long term environmental monitoring system. Then a series of full modal surveys directly before and after applying a damage pattern to the Z24 bridge were carried out. Each damage pattern was increased in progressive steps until it could safely be detected or the safety of the bridge was endangered. The paper describes the instrumentation, test setup, damage patterns and the respective considerations made. Finally conclusions are drawn with regard to the instrumentation for long term monitoring of possible bridge candidates.
No abstract is provided for this article.
Vibroacoustic metamaterials are used to improve the sound insulation of a host panel by periodically attaching local resonators.While most vibroacoustic metamaterials exploit single-mode resonators for narrowband improvements, the use of multi-modal resonators has recently emerged to extend the frequency band of effectiveness.However, significant modelling and computational challenges need to be tackled when designing adequate resonator layouts that exploit multiple modes to influence broadband sound transmission.In this work, we show that analytical effective medium models can accurately predict the performance of vibroacoustic metamaterials, simply requiring the extraction of modal parameters from the analysis of the fixed-based resonator.Effective medium models are used to analyze diffuse field sound transmission loss both for infinite structure dimensions, through analytical formulas, and for finite size, through combination with the hybrid deterministic -Statistical Energy Analysis.The accuracy of the proposed prediction methodologies is validated for two different multi-modal metamaterial layouts, that target the suppression of the broadband insulation dip due to coincidence in orthotropic host panels.
This paper presents a technique for offline time synchronization of data acquisition systems. The technique can be applied when real-time synchronization of data acquisition systems is impossible or not sufficiently accurate. It allows for accurate synchronization based on the acquired dynamic response of the structure only, without requiring a common response or the use of a trigger signal. The synchronization is performed using the results obtained from system identification, and assumes linear dynamic behavior of the structure and proportional damping of the structural modes. A demonstration for a laboratory experiment on a cantilever steel beam shows that the proposed methodology can be used for accurate time synchronization, resulting in a significant improvement of the accuracy of the identified mode shapes.
This paper develops a sensitivity-based updating method to identify the damage in a tested reinforced concrete (RC) frame modeled with a two-dimensional planar finite element (FE) by minimizing the discrepancies of modal frequencies and mode shapes. In order to reduce the number of unknown variables, a bidimensional damage (element) function is proposed, resulting in a considerable improvement of the optimization performance. For damage identification, a reference FE model of the undamaged frame divided into a few damage functions is firstly obtained and then a rough identification is carried out to detect possible damage locations, which are subsequently refined with new damage functions to accurately identify the damage. From a design point of view, it would be useful to evaluate, in a simplified way, the remaining bending stiffness of cracked beam sections or segments. Hence, an RC damage model based on a static mechanism is proposed to estimate the remnant stiffness of a cracked RC beam segment. The damage model is based on the assumption that the damage effect spreads over a region and the stiffness in the segment changes linearly. Furthermore, the stiffness reduction evaluated using this damage model is compared with the FE updating result. It is shown that the proposed bidimensional damage function is useful in producing a well-conditioned optimization problem and the aforementioned damage model can be used for an approximate stiffness estimation of a cracked beam segment.
No abstract is provided for this article.
Vibration testing is a well-known practice for damage identification of civil engineering structures. The real modal parameters of a structure can be determined from the data obtained by tests using system identification methods. By comparing these measured modal parameters with the modal parameters of a numerical model of the same structure in undamaged condition, damage detection, localization, and quantification is possible. This paper presents a real-life application of this technique to assess the structural health of the 50-year old bridge of Tilff, a prestressed three-cell box-girder concrete bridge with variable height. A complete ambient vibration survey comprising both vertical accelerations and axial strains has been carried out. The in situ use of optical fiber strain sensors for the direct measurement of modal strains is an original contribution of this work. It is a big step forward in the exploration of modal curvatures for damage identification because the accuracy in calculating the modal curvatures is substantially improved by directly measuring modal strains rather than deriving the modal curvatures from acceleration measurements. From the ambient vibrations, natural frequencies, damping factors, modal displacements and modal curvatures are extracted by the stochastic subspace identification method. These modal param- eters are used for damage identification which is performed by the updating of a finite element model of the intact structure. The obtained results are then compared to the inspections performed on the bridge.
A method for vibration-based damage localization and quantification, based on quasi-static flexibility, is presented. The experimentally determined flexibility matrix is combined with a virtual load that causes nonzero stresses in a small part of the structure, where a possible local stiffness change is investigated. It is shown that, if the strain–stress relationship for the load is proportional, the ratio of some combination of deformations before and after a stiffness change has occurred, equals the inverse local stiffness ratio. The method is therefore called local flexibility (LF) method. Since the quasi-static flexibility matrix can be composed directly from modal parameters, the LF method allows to determine local stiffness variations directly from measured modal parameters, even if they are determined from output-only data. Although the LF method is in principle generally applicable, the emphasis in this paper is on beam structures. The method is validated with simulation examples of damaged isostatic and hyperstatic beams, and experiments involving a reinforced concrete free–free beam and a three-span prestressed concrete bridge, that are both subjected to a progressive damage test.
No abstract is provided for this article.
No abstract is provided for this article.
No abstract is provided for this article.
Long multi-span bridges represent a broad section of the civil roadway infrastructure. Despite being so common, their condition-based maintenance through vibration-based Structural Health Monitoring (SHM) has been scarcely investigated in the literature. The dynamic identification of such structures through Operational Modal Analysis (OMA) is especially challenging due to of their quasi-periodic nature and the common existence of weak inter-span coupling. Even when designed following an isostatic scheme, there always exists a certain degree of coupling between spans due to the continuity of the deck, the pavement and imperfect expansion joints. Hence, the modal poles of the spans typically appear as dense clusters with closely spaced frequencies and mode shapes with similar wavelengths, which significantly hinders the identification of physical poles through stabilization diagrams. In this light, this paper proposes a model-based machine learning approach to conduct and interpret the OMA results of partially continuous multi-span bridges. The proposed method is a hierarchical clustering approach that leverages on the analytical solution of the vertical free vibration response of multi-span girders with weak inter-span rotational coupling, allowing the estimation of the modal features of any bridge configuration ranging from simply supported to perfectly continuous conditions. Detailed parametric analyses and discussions are presented to appraise the correlation between the inter-span rotational coupling and the clustering of the modal poles of multi-span bridges, as well as the influence of damage conditions of varying severity and extension. On this basis, a model-based cut-off distance threshold for hierarchical clustering of stable poles is proposed to assist the automation of the global OMA of multi-span bridges. The developed formulation is tested in a real-world in-operation seven-spans reinforced concrete girder bridge, the Trigno V Bridge in Italy.
