Boris A. Zárate

Probabilistic Prognosis of Fatigue Crack Growth Using Acoustic Emission Data

This paper presents a structural health monitoring methodology that uses acoustic emission (AE) features to predict crack growth in structural elements subjected to fatigue. This allows for the prediction of the failure of the structural element at the current load level. The methodology uses Bayesian inference to account for different sources of uncertainty such as uncertainty in the data (AE signal), unknown fracture mechanics parameters, and model inadequacy. The methodology is divided into two main components: a model updating component that uses available data to build a joint probability distribution of the different unknown fracture mechanics parameters, and a prognosis component in which this multivariable probability distribution is sampled to predict the stress intensity factor range at a future number of cycles. The application of the methodology does not require knowledge of the load amplitude nor the initial crack length. The methodology is validated using experimental data from a compact test specimen under cyclic loading.

Reducing Epistemic Uncertainty Using a Model Updating Cognitive System

Developing numerical models of existing structural systems is challenging because of the uncertainty inherent on the development of the numerical model and the estimation of the structural parameters. This uncertainty is a combination of lack of knowledge (epistemic uncertainty) and inherent randomness on the system. This paper introduces a Model Updating Cognitive Systems (MUCogS) as a new paradigm for model updating of structural systems with incomplete data. MUCogS seeks to merge the computational power of computers with the analytical power of the analyst. In most cases, the posterior probability density function (PDF) within a Bayesian framework has one region of high probability. However, several regions of high probability can be obtained on the likelihood when data is incomplete. These areas can be considered by the analyst to enhance his/her knowledge about the structure. This paper discusses a methodology used to identify these regions of high probability without the need of calculating the complete likelihood using Modeling to Generate Alternatives (MGA).

Quantification of Fatigue Cracking in CT Specimens with Passive and Active Piezoelectric Sensing

Monitoring of fatigue cracks in steel bridges is of interest to bridge owners and agencies. Monitoring of fatigue cracks has been attempted with acoustic emission using either resonant or broadband sensors. One drawback of passive sensing is that the data is limited to that caused by growing cracks. In this work, passive emission was complemented with active sensing (piezoelectric wafer active sensors) for enhanced detection capabilities. Passive and active sensing methods were described for fatigue crack monitoring on specialized compact tension specimens. The characteristics of acoustic emission were obtained to understand the correlation of acoustic emission behavior and crack growth. Crack and noise induced signals were interpreted through Swansong II Filter and waveform-based approaches, which are appropriate for data interpretation of field tests. Upon detection of crack extension, active sensing was activated to measure the crack size. Model updating techniques were employed to minimize the difference between the numerical results and experimental data. The long term objective of this research is to develop an in-service prognostic system to monitor structural health and to assess the remaining fatigue life.

Effects of Cable Dynamics in the Modeling of Cable-Stayed Bridges Under Seismic Excitation

New longer and bigger cable-stayed bridges are planned for the future, increasing the necessity of improving the way bridge models are used, especially when the models are used for dynamic behavior. This paper studies how the dynamic behavior of the deck and towers of a particular structure are affected when the cables are modeled with three different cable models. A finite element model of the Bill Emerson Memorial Bridge is used to compare the different methodologies. Three cable models are used in this study including the equivalent Youngs modulus, the elastic catenary and the elastic isoparametric formulation. Results show that in the case of the Bill Emerson Memorial Bridge the tower behaves differently depending on the cable model used while the deck has a similar behavior independent on the cable model used.

Development and Implementation of a Cyberinfrastructure Framework for Research in Nondestructive Evaluation Using Acoustic Emission Data

AbstractThis paper presents the development and validation of a cyberinfrastructure architecture for research in nondestructive evaluation (NDE) using acoustic emission (AE) data. Existing cyberinfrastructures for civil engineering focus in the curation and preservation of data. In contrast, the proposed cyberinfrastructure is intended to serve as a tool to enable innovation by providing a platform to prototype analysis techniques and sharing data and analysis methods among a research team while removing the burden of memory and computational cost from the user. This is achieved by streamlining the access of large and complex experimental data sets, facilitating the selection of part of the experimental data depending on data features, distributing data analysis using a distributed computing strategy, and allowing the creation of new data features that can be used for subsequent analysis. The experimental data set potentially include data from AE sensors, strain gages, load cells, clip gages, and accelero...

Using P-Box and PiFE to Express Uncertainty in Model Updating

This paper proposes the use of probability bounds with the Pseudo-inverse Finite Element (PiFE) method for structural model updating. The technique estimates the probability bound of structural parameters based on dynamic or static features such as modal parameters or static displacements. Two methods are explored for the calculation of the probability bounds: (i) Naive method and (ii) all possible combinations. The capabilities of the technique are explored using a two degree of freedom structural system where the stiffness is considered uncertain. Results indicate that both the Naive and all possible combination techniques are applicable with PiFE and produce bounds that include the cumulative distribution function of the structural parameters. The probability bounds found with the all possible combinations method was narrower for this particular example.

Uncertainty quantification of acoustic emission filtering techniques

This paper compares six different filtering protocols used in Acoustic Emission (AE) monitoring of fatigue crack growth. The filtering protocols are combination of three different filtering techniques which are based on Swansong-like filters and load filters. The filters are compared deterministically and probabilistically. The deterministic comparison is based on the coefficient of determination of the resulting AE data, while the probabilistic comparison is based on the quantification of the uncertainty of the different filtering protocols. The uncertainty of the filtering protocols is quantified by calculating the entropy of the probability distribution of some AE and fracture mechanics parameters for the given filtering protocol. The methodology is useful in cases where several filtering protocols are available and there is no reason to choose one over the others. Acoustic Emission data from a compact tension specimen tested under cyclic load is used for the comparison.

BAYESIAN FINITE ELEMENT MODEL UPDATING USING STATIC AND DYNAMIC DATA

Finite element models of current structures often behave differently than the structure itself. Model updating techniques are used to enhance the capabilities of the numerical model such that it behaves like the real structure. Experimental data is used in model updating techniques to identify the parameters of the numerical model. In civil infrastructure these model updating techniques use either static or dynamic measurements, separately. This paper studies how a Bayesian updating framework behaves when both static and dynamic data are used to updated the model. Displacements at specific structure locations are obtained for static tests using a computer vision method. High density mode shapes and natural frequencies are obtained using a moving accelerometer structure. The static data and the modal characteristics are combined in a Bayesian modal updating technique that accounts for the incompleteness and uncertainty of the data as well as the possible nonuniqueness of the solution. Results show how the posterior probability density function changes when different type of information is included for updating.

Model updating and prognosis of acoustic emission data in compact test specimens under cyclic loading

Acoustic emission (AE) is generated when cracks develop and it is used as an indicator of the current state of damage in structural elements. Algorithms that use AE data to predict the state of a structural element are still in their research stages because the relationship between crack length and AE activity is not well understood. The process of trying to predict the future stage of a crack based on AE data is usually performed by an expert, and requires significant experience. This paper proposes a new strategy for the use of AE data for structural prognosis. A probabilistic model is used to predict AE data. An expert can analyze this data to draw conclusions about the health of the structural member. The goal is to aid the analyst by providing an estimation of the AE activity in the future. The methodology provides the cumulative signal strength at a future number of cycles, assuming the loading and boundary conditions hold. The methodology uses a relationship between the rate of change of the cumulative absolute energy of the AE with respect to the number of cycles and the stress intensity range. A third order polynomial equation that describes the stress intensity range as function of the AE data is proposed. The variables to be updated are treated as random and their joint probability distribution is computed using Bayesian inference. Markov Chain Monte Carlo (MCMC) is used to forecast the cumulative signal strength at some number of cycles in the future. The methodology is tested using a compact test specimen tested in structures lab at the University of South Carolina.

Bayesian Finite Element Model Updating for Crack Growth

This paper describes a probabilistic structural health monitoring framework to determine crack growth on structural members using model updating. The framework uses Bayesian inference to estimate crack lengths. On the proposed framework data from embedded piezoelectric wafer sensors (PWAS) and acoustic emission sensors is used for model updating. This paper presents preliminary results obtained using simulated data of a steel specimen. As a first step, the crack length is estimated using calculated displacements at the tip of the specimen. Results show that Bayesian inference can be used to estimate crack lengths on structural members.