Search

Monitoring information can facilitate the condition assessment of railway infrastructure, via delivery of data that is informative on condition. A primary instance of such data is found in Axle Box Accelerations (ABAs), which track the dynamic vehicle/track interaction. Such sensors have been installed on specialized monitoring trains, as well as on in-service On-Board Monitoring (OBM) vehicles across Europe, enabling a continuous assessment of railway track condition. However, ABA measurements come with uncertainties that stem from noise corrupt data and the non-linear rail–wheel contact dynamics, as well as variations in environmental and operational conditions. These uncertainties pose a challenge for the condition assessment of rail welds through existing assessment tools. In this work, we use expert feedback as a complementary information source, which allows the narrowing down of these uncertainties, and, ultimately, refines assessment. Over the past year, with the support of the Swiss Federal Railways (SBB), we have assembled a database of expert evaluations on the condition of rail weld samples that have been diagnosed as critical via ABA monitoring. In this work, we fuse features derived from the ABA data with expert feedback, in order to refine defection of faulty (defect) welds. Three models are employed to this end; Binary Classification and Random Forest (RF) models, as well as a Bayesian Logistic Regression (BLR) scheme. The RF and BLR models proved superior to the Binary Classification model, while the BLR model further delivered a probability of prediction, quantifying the confidence we might attribute to the assigned labels. We explain that the classification task necessarily suffers high uncertainty, which is a result of faulty ground truth labels, and explain the value of continuously tracking the weld condition., Sensors, 23 (5), ISSN:1424-8220

Data from guided wave propagation in structures, produced by piezoelectric elements, can offer valuable information regarding the possible existence of flaws. Numerical models can be used to complement the attained data for refining the potential for flaw characterization. Unfortunately, evaluation of these models remains computationally expensive, especially for small defects, due to the short wavelength required for detection and, the in turn fine discretization in time and space. This renders real–time simulation infeasible, rendering GW–approaches less attractive for inverse problem formulations, where the forward problem needs to be solved several times. We propose an accelerated computation method, which exploits the properties of guided waves interacting with defects, where an extra band of waves is created, whose phase is differentiated, depending on the location of the flaw (e.g. notch) within the medium. To expedite the actual simulation for the inverse problem, the system is parametrized in terms of the location of the flaw and, in an offline phase, is repeatedly solved to produce snapshots of the system’s response. The snapshots are used to create a physics–informed interpolation of the solution of the wave propagation problem for different flaw locations. The gained information is then used in an inverse setting for localising the defect using an evolution strategy as a means to stochastic, derivative-free numerical optimization. The method is demonstrated in simulations of a 2D slice of a thin plate., Structural Health Monitoring 2021: Enabling Next Generation SHM for Cyber-Physical Systems, ISBN:978-1-60595-687-9

A significant part of the existing building stock in regions of low to moderate seismic hazard has been designed without modern seismic considerations and is, in the meantime, exceeding its design life span. The assessment of seismic performance poses an engineering challenge, due to unknown material properties, undocumented structural interventions and the scarcity of event-based information. Operational modal analysis has been applied in some cases to verify model assumptions beyond visual inspection. However, masonry buildings exhibit amplitude-dependent stiffness even at very low response amplitudes, raising questions about the validity of such methods. Planned demolitions provide engineers with the opportunity to leverage higher-amplitude vibrations generated during demolition activities to better understand the dynamic behaviour of existing buildings. This paper introduces a Bayesian model-updating framework, which aims at reducing uncertainty in seismic analysis, by fusing dynamic measurements with best-practice structural models. The proposed hybrid framework is applied to nine real masonry buildings, representative of existing residential buildings, as typically encountered in Switzerland, that have been monitored during controlled demolition. A vast reduction in prediction uncertainty is achieved through data-driven model updating, additionally exposing intra- and inter-typological differences in terms of seismic capacity and ductility. In addition, differences between updated model predictions and typical engineering assumptions and generic typological curves are discussed. Overall, this contribution demonstrates, applies and discusses the practical benefits of a straightforward methodology for fusing monitoring data into the seismic evaluation of existing masonry structures., Bulletin of Earthquake Engineering, 20, ISSN:1570-761X, ISSN:1573-1456

Lecture Notes in Civil Engineering, 200, ISSN:2366-2557, ISSN:2366-2565, Proceedings of the 1st Conference of the European Association on Quality Control of Bridges and Structures, ISBN:978-3-030-91876-7, ISBN:978-3-030-91879-8, ISBN:978-3-030-91877-4

Injecting parametric dependency and treating nonlinear phenomena pose the main challenges when seeking to construct an accurate Reduced Order Model (ROM) of an actual complex system. Also, real-life structures may often comprise multiple components demanding separate treatment. This paper derives a physics-based ROM, reflecting dependencies on system properties and characteristics of the induced excitation. This is achieved utilizing a projection strategy relying on Proper Orthogonal Decomposition. The framework is then coupled with the substructuring approach in [13]. This representation allows integrating localized domains experiencing nonlinearity or damage on the ROM and enables response learning both at a global and a local level. The pROM derived through this fusion can address each component's dynamics separately and apply individual reduction, leading to a modular formulation., Book of Abstracts

Structural systems are usually characterized by complex dynamic response under earthquake or damage events, which extends beyond the expected behavior during regular operational conditions, posing thus unique and challenging requirements for designing and utilizing digital twins. Such representations have become indispensable due to the ongoing virtualization, and their efficiency in addressing such phenomena is critical, especially in the context of structural health monitoring applications. Our work features parametric Reduced Order Models (ROMs) as efficient, low-order surrogates of parametrized structural, dynamical systems and addresses the challenges of adaptive performance and, subsequently, response prediction under state deterioration or stochastic excitation case studies. Specifically, our approach initiates by deriving a projection-based, reliable ROM, reproducing the high fidelity system dynamics, and capturing any parametric dependencies during an initial system modeling stage. To achieve this, we rely on Proper Orthogonal Decomposition (POD) to approximate the subspace of the response from a set of training evaluations and then project and propagate the respective reduced coordinates’ dynamics. However, extreme loads like those induced by earthquakes challenge the performance limits of such representations during operating conditions as POD restricts the dynamics in a single, linear approximation of the original manifold. In contrast, the actual excitation may lead to substantially different behavior than the instances observed during the training configurations. As a remedy to this issue, we propose adopting the initial ROM as a forward simulator and adapt the projection basis during operation through a Gaussian Processes Regressor (GPR) coupled to the ROM framework. Relying on monitoring output measurements from a sparse number of system channels, a suitable performance indicator signals the failure of the initial surrogate in the event of extreme operating conditions. In turn, the framework enriches the initial projection subspace by employing the GPR-based scheme, which relies upon data assimilation from the monitoring channels to approximate the deformed configuration of the system and assemble the respective ”damaged” mode. This fusion allows for an adaptive ROM as it enables online projection basis refinement during operation, whilst warning for potential irreversible, or even catastrophic, consequences. In addition, the confidence bounds of the refinement mode prediction enable an initial quantification of the uncertainty involved in the output ROM response estimation.

In the present work, we consider the problem of combining the output from multiple stochastic computer simulators to make inference on a quantity of interest, as a means of reducing the inherent model-form uncertainty in the absence of any measurements. In most real-world situations, judging an individual stochastic simulator to be the “best” for any given point in the input space is highly doubtful. Thus, making inference by relying on the so-deemed best simulator may not be adequate, especially when the sampled data is limited. To this end, we propose an ensemble learning method based on local Clustering and bootstrap aggregation (Bagging), which rather than treating the stochastic predictions of the simulators as competing individual information sources, treats those as part of an ensemble, thus diversifying the hypothesis space. We call the proposed method: unsupervised local cluster-weighted bootstrap aggregation. Variational Bayesian Gaussian mixture clustering is the first step in this ensemble learning approach for discriminating the outputs, and deriving the probability map (weights) of the clustered simulators output. Clustering is performed on the stochastic output corresponding to the binned input space. Performing the clustering independently and deriving the probability map for each local region of the binned input space is a novelty that guarantees an adaptive solution, whereby certain simulators are potentially more fitting than others in corresponding regions of the input space. The second step consists in a local cluster-weighted Bootstrap Aggregation, which serves the purpose of weighted combination of the clustered ensemble of outputs from the individual simulators. Based on simulations, we demonstrate how the input bin size, sample size, output dispersion and level of agreement amongst the simulators affect the performance of the proposed method. We compare the unsupervised local cluster-weighted bootstrap aggregation method to classical Bagging, Bayesian Model Averaging and Stacking of predictive distributions. Finally, we demonstrate the method by evaluating the fatigue damage equivalent load on a wind turbine blade, using 10 finite element based simulators. The results point to the need for practitioners to consider this as a useful method, when model-form uncertainty is of concern and when output from multiple stochastic simulators are available., Reliability Engineering & System Safety, 199, ISSN:0951-8320

Mechanical Systems and Signal Processing, 145, ISSN:0888-3270, ISSN:1096-1216

We consider in this paper the problem of aggregating the output from multiple computer simulators (models) and measurements to make inference on a stochastic quantity of interest. Our ensemble learning approach consists in a clustering-Bagging step, followed by a data assimilation step based on Bayesian model averaging or Bayesian model combination. Unsupervised clustering is the first step in the ensemble learning approach and serves the purpose of distinguishing the output from each simulator (or model), and deriving apriori probability map (weights) of the simulators. Clustering is performed on the stochastic output corresponding to the binned input space, where each bin is considered as a dimension. The second step consists in a weighted Bootstrap Aggregation (Bagging), which serves the purpose of weighted aggregation (combination) of the clustered ensemble of outputs from the individual simulators. Finally, when observational training datasets of the quantity of interest become available, the Bayesian Model Averaging (BMA) or Bayesian Mode Combination (BMC) is used to combine the prior aggregated simulations predictions with the assimilated data to update the posterior model probabilities (weights) of each of the proposal models. We demonstrate our approach on the extreme blade deflection of a wind turbine blade., Reliability and Optimization of Structural Systems: Proceedings of the 19th IFIP WG-7.5 conference on Reliability and Optimization of Structural Systems, ISBN:978-3-906916-56-9

This paper presents a novel model order reduction technique for 3D flexible multibody systems featuring nonlinear elastic behavior. We adopt the mean-axis floating frame approach in combination with an enhanced Rubin substructuring technique for the construction of the reduction basis. The standard Rubin reduction basis is augmented with the modal derivatives of both free-interface vibration modes and attachment modes to consider the bending–stretching coupling effects for each flexible body. The mean-axis frame generally yields relative displacements and rotations of smaller magnitude when compared to the one obtained by the nodal-fixed floating frame. This positively impacts the accuracy of the reduction basis. Also, when equipped with modal derivatives, the Rubin method better considers the geometric nonlinearities than the Craig–Bampton method, as it comprises vibration modes and modal derivatives featuring free motion of the interface. The nonlinear coupling between free-interface modes and attachment modes is also considered. Numerical tests confirm that the proposed method is more accurate than Craig–Bampton’s, a nodal fixed floating frame counterpart originally proposed in Wu and Tiso (Multibody Syst. Dyn. 36(4): 405–425, [2016]), and produces significant speed-ups. However, the offline cost is increased because the mean-axis formulation produces operators with decreased sparsity patterns., Multibody System Dynamics, 45 (1), ISSN:1573-272X, ISSN:1384-5640

This work addresses the problem of vibration-based damage detection on structures operating under significant levels of uncertainty originating from variable environmental and operational conditions. For this purpose, a Gaussian Process Regression Vector AR (GPR-VAR) model is postulated for the representation of the vibration response of a structure at multiple measurement locations as a function of the uncertain inputs. The GPR-VAR model is a hierarchical Bayes representation associating the vibration response, the model parameters and the uncertain inputs. The Bayesian framework further helps to quantify the confidence attributed to the decision conditioned on the quality of the training set. The workings of the method are demonstrated via a simulated wind turbine blade driven by turbulent wind excitation, with uncertainty being introduced by changing temperatures and wind speeds., Procedia Engineering, 199, X International Conference on Structural Dynamics, EURODYN 2017

This study proposes a computational framework for the online estimation of fatigue damage using operational vibration measurements from a limited number of sensors. To infer the stress response time histories required for fatigue prediction, the measured structural response is driven to a high fidelity finite element (FE) model, which is reconciled using appropriate model updating techniques that minimize the discrepancy between the experimental and analytical frequency response functions (FRFs). Fatigue is accordingly estimated via the Palmgren-Miner rule, while the available FE model allows for stress estimation at unmeasured spots. The method is successfully validated and assessed through an experimental study that pertains to a linear steel substructure supporting the entire body of a pre-beater assembly at a PPC power plant., Procedia Engineering, 199

For a number of applications, including real/time damage diagnostics as well as control, online methods, i.e., methods which may be implemented on-the-fly, are necessary. Within a system identification context, this implies adoption of filtering algorithms, typically of the Kalman or Bayesian class. For engineered structures, damage or deterioration may often manifest in relation to phenomena such as fracture, plasticity, impact, or friction. Despite the different nature of the previous phenomena, they are described by a common denominator: switching behavior upon occurrence of discrete events. Such events include for example, crack initiation, transitions between elastic and plastic response, or between stick and slide modes. Typically, the state-space equations of such models are non-differentiable at such events, rendering the corresponding systems non-smooth. Identification of non-smooth systems poses greater difficulties than smooth problems of similar computational complexity. Up to a certain extent, this may be attributed to the varying identifiability of such systems, which violates a basic requirement of online Bayesian Identification algorithms, thus affecting their convergence for non-smooth problems. Herein, a treatment to this problem is proposed by the authors, termed the Discontinuous D– modification, where unidentifiable parameters are acknowledged and temporarily excluded from the problem formulation. In this work, the D– modification is illustrated for the case of the Unscented Kalman Filter UKF, resulting in a method termed DUKF, proving superior performance to the conventional, and widely adopted, alternative., Frontiers in Built Environment, 3, ISSN:2297-3362

Structural health monitoring of large systems is a complex engineering task due to important practical issues. When dealing with large structures, damage diagnosis, localization, and prognosis necessitate a large number of sensors, which is a nontrivial task due to the lack of scalability of traditional sensing technologies. In order to address this challenge, the authors have recently proposed a novel sensing solution consisting of a low-cost soft elastomeric capacitor that transduces surface strains into measurable changes in capacitance. This paper demonstrates the potential of this technology for damage detection, localization, and prognosis when utilized in dense network configurations over large surfaces. A wind turbine blade is adopted as a case study, and numerical simulations demonstrate the effectiveness of a data-driven algorithm relying on distributed strain data in evidencing the presence and location of damage, and sequentially ranking its severity. Numerical results further show that the soft elastomeric capacitor may outperform traditional strain sensors in damage identification as it provides additive strain measurements without any preferential direction. Finally, simulation with reconstruction of measurements from missing or malfunctioning sensors using the concepts of virtual sensors and Kriging demonstrates the robustness of the proposed condition assessment methodology for sparser or malfunctioning grids., Shock and Vibration, 2016, ISSN:1070-9622, ISSN:1875-9203