Structural damage detection and localisation using multivariate regression models and two-sample control statistics

Shahidi, S. Golnaz; Nigro, Mallory B.; Pakzad, Shamim N.; Pan, Yingtian

doi:10.1080/15732479.2014.949277

Cited by 28 publications

(21 citation statements)

References 46 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…7.3b, c) indicates a significant increase in effort is necessary for higher model orders. Although, AR modeling is computationally less intensive [21], presents examples of application of SVR and ARX models that outperformed AR models in damage localization. This comparison was performed at a pre-specified model order and did not include the model order selection attempts [23], however it establishes an overall behavior of computation costs for these methods.…”

Section: Damage Detection Applicationmentioning

confidence: 98%

Are Today’s SHM Procedures Suitable for Tomorrow’s BIGDATA?

Matarazzo

Shahidi

Chang

et al. 2015

Structural Health Monitoring and Damage Detection, Volume 7

Self Cite

View full text Add to dashboard Cite

Large SHM datasets often result from special applications such as long-term monitoring, dense sensor arrays, or high sampling rates. Through the development of novel sensing techniques as well as advances in sensing devices and data acquisition technology, it is expected that such large volumes of measurement data become commonplace. In anticipation of datasets magnitudes larger than today's, it is important to evaluate current SHM processing methods at BIGDATA standards and identify potential limitations within computational procedures. This paper will focus on the processing of large SHM datasets and the computational sensitivity of common SHM procedures. Processing concerns encompass efficiency and scalability of SHM software, particularly the computational sensitivity of common system identification and damage detection algorithms with respect to a large amount of sensors and samples. IntroductionThe data acquisition process plays a key role in the monitoring of structural systems. Over the years, sensing networks have evolved from basic to complex. Today, structural response quantities are measured with high resolution in time and space by means of wired and wireless contact sensors as well as noncontact digital camera lenses. In order to infer about structural health condition, the measured signals are processed through various SHM algorithms (system identification, model calibration, and damage detection). Therefore, along with improvement in sensing technology, it is critical to evaluate the performance of SHM algorithms to study their scalability potential and overcome possible limitations in processing large SHM datasets. Toward this goal, this paper investigates performance of a subset of the current SHM procedures with growth of network size and data acquisition interval. This investigation is conducted in two parts: the preprocessing of measured data as the first step in any SHM procedure, and the computational cost of certain system identification and damage detection algorithms. Although it is expected that BIGDATA storage requirements will be far greater than limits of local hard disks and RAM, this aspect of the problem is outside of the scope of this paper. Exactly How Big Is "BIGDATA"?Today, a dataset earns the title "large" from the sizes of its dimensions and storage space. SHM datasets are typically multivariate times series with two dimensions: number of sensors and number of time samples. However, the mere storage requirements, say in megabytes, of a dataset is not enough to constitute the title "BIGDATA". Inherently, BIGDATA requires "BIGPROCESSING", a subsequent analysis of BIGDATA, of which computational complexity varies among SHM applications and techniques.

show abstract

Section: Damage Detection Applicationmentioning

confidence: 98%

Are Today’s SHM Procedures Suitable for Tomorrow’s BIGDATA?

Matarazzo

Shahidi

Chang

et al. 2015

Structural Health Monitoring and Damage Detection, Volume 7

Self Cite

View full text Add to dashboard Cite

show abstract

“…General regression models include AR and ARX models, as well as Single Variate Regression (SVR) and Collinear Regression (CR); special cases of the ARX model discussed in [17]. Several damage features are available using these models: (1) coefficient-based features [12,17,18]: regression coefficients themselves, angle coefficients, Mahalanobis distance, and spectral distance of regression coefficients. (2) residual-based damage features [12]: (normalized) variance, (normalized) standard deviation, and Ljung-Box statistics of the residual vectors.…”

Section: Damage Identification Toolsuite (Dit)mentioning

confidence: 99%

Data-Driven Structural Damage Identification Using DIT

Shahidi

Yao

Chamberlain

et al. 2015

Conference Proceedings of the Society for Experimental Mechanics Series

Self Cite

View full text Add to dashboard Cite

Vibration-based damage detection research aims to develop efficient algorithms to identify structural damage from monitoring data. One of the main categories of such algorithms is data-driven techniques which extract features from measured signals, and identify the damage by evaluating the significance of potential changes in these features. This paper presents application of several data-driven damage identification methodologies on a multivariate simulated data set. First, general regression models are applied to data collected through clusters of sensors and damage sensitive features are extracted. For systems with linear topology, it is shown that substructural regression modeling can also be performed on time-and frequency-domain transforms of the measured signals to estimate local stiffness of the structure as damage features. Subsequently, change detection techniques are utilized to statistically determine the significance of changes in the extracted features in order to distinguish between assignable changes as a result of damage and common changes due to environmental factors. Finally, a toolsuite is developed to facilitate application of the developed algorithms and improve the damage identification performance through incorporation of various combinations of regression models, damage features and statistical tests. IntroductionThe ultimate goal of vibration-based structural health monitoring (SHM) research is to develop efficient algorithms capable of detecting the time, location, and severity of any damage induced changes in structural components of monitored systems. Toward this goal, multitudes of methods have been proposed over recent decades that commonly establish a baseline for selected damage features and monitor their change from the baseline in the following unknown health conditions. Various features have been used for the purpose of damage identification; uncertain parameters of a finite element (FE) simulation of the structure that is calibrated with reference to modal quantities extracted using system identification (SID) algorithms [1,2], or features that are directly extracted from measured signals without using FE or SID procedures to train the data [3][4][5][6]. Each group of these damage identification methods has their own advantages and drawbacks. The first group is usually more laborious to implement and require certain a priori knowledge of structural properties, and location of damage; however, the calibrated model could be beneficial in design of repair scenarios or estimating the remaining life of the structure. The main advantage of the second group is their efficiency, and that they can be readily applied to measured signals without any prior information. Therefore, their application for a general automated damage detection platform is more promising. However, these data-driven methods are ineffective without statistical analyses to determine the change threshold for the extracted features. This paper presents application of multiple data-driven damage detection pr...

show abstract

“…The responses from sensor L5 were considered as the "output" or y values of the ARX model, with L4 constituting the "input" u values. These sensor designations are used so as to ease the cross-referencing of this work with that in Nigro and Pakzad (2014) and Shahidi et al (2015). The test lasted 2 s and each wireless sensor had 500 Hz sampling rates, yielding 1,000 total measurements apiece.…”

Section: Structural Vibration Datamentioning

confidence: 99%

“…Structural vibration data has popularly been used in the field for system identification (Juang and Pappa, 1985;James III et al, 1993;Andersen, 1997;Huang, 2001;Pakzad et al, 2011;Dorvash and Pakzad, 2012;Pakzad, 2013, 2014;Dorvash et al, 2013;Cara et al, 2014;Matarazzo and Pakzad, 2016c;Nagarajaiah and Chen, 2016), finite element model updating (Shahidi and Pakzad, 2014a,b;Yousefianmoghadam et al, 2016;Nozari et al, 2017;Song et al, 2017), and damage-sensitive feature extraction (Sohn et al, 2001;Gul and Catbas, 2009;Kullaa, 2009;Dorvash et al, 2015;Shahidi et al, 2015), with the ultimate goal of inferring information about the current condition of the monitored structure. Regarding regression models, He and De Roeck (1997) shows their utility for describing structural vibration responses, and Shahidi et al (2015) and Yao and Pakzad (2012) provide structural damage-sensitive features created using the parameters of regression models.…”

Section: Introductionmentioning

confidence: 99%

Parameter Estimation of Autoregressive-Exogenous and Autoregressive Models Subject to Missing Data Using Expectation Maximization

Horner

Pakzad

Gulgec

2019

Front. Built Environ.

Self Cite

View full text Add to dashboard Cite

Missing observations may present several problems for statistical analyses on datasets if they are not accounted for. This paper concerns a model-based missing data analysis procedure to estimate the parameters of regression models fit to datasets with missing observations. Both autoregressive-exogenous (ARX) and autoregressive (AR) models are considered. These models are both used to simulate datasets, and are fit to existing structural vibration data, after which observations are removed. A missing data analysis is performed using maximum-likelihood estimation, the expectation maximization (EM) algorithm, and the Kalman filter to fill in missing observations and regression parameters, and compare them to estimates for the complete datasets. Regression parameters from these fits to structural vibration data can thereby be used as damage-sensitive features. Favorable conditions for accurate parameter estimation are found to include lower percentages of missing data, parameters of similar magnitude with one another, and selected model orders similar to those true to the dataset. Favorable conditions for dataset reconstruction are found to include random and periodic missing data patterns, lower percentages of missing data, and proper model order selection. The algorithm is particularly robust to varied noise levels.

show abstract

Structural damage detection and localisation using multivariate regression models and two-sample control statistics

Cited by 28 publications

References 46 publications

Are Today’s SHM Procedures Suitable for Tomorrow’s BIGDATA?

Are Today’s SHM Procedures Suitable for Tomorrow’s BIGDATA?

Data-Driven Structural Damage Identification Using DIT

Parameter Estimation of Autoregressive-Exogenous and Autoregressive Models Subject to Missing Data Using Expectation Maximization

Contact Info

Product

Resources

About