Norman F. Hunter

Structural Health Monitoring Using Statistical Pattern Recognition Techniques

All members of the Los Alamos Structural Health Monitoring Team contributed to this study reported herein. The team members include George Papcum and Michael L. Fugate from the CIC3 Group, and Scott Doebling from the Engineering Analysis Group. Funding for this investigation came primarily through Los Alamos National Laboratory Director’s Funded Postdoctoral Fellows Program. The authors also thank Gregg Johnson and Mike Todd of NRL for providing the experimental data and allowing the publication of the test results.

Applying the LANL Statistical Pattern Recognition Paradigm for Structural Health Monitoring to Data from a Surface-Effect Fast Patrol Boat

This report summarizes the analysis of fiber-optic strain gauge data obtained from a surface-effect fast patrol boat being studied by the staff at the Norwegian Defense Research Establishment (NDRE) in Norway and the Naval Research Laboratory (NRL) in Washington D.C. Data from two different structural conditions were provided to the staff at Los Alamos National Laboratory. The problem was then approached from a statistical pattern recognition paradigm. This paradigm can be described as a four-part process: (1) operational evaluation, (2) data acquisition & cleansing, (3) feature extraction and data reduction, and (4) statistical model development for feature discrimination. Given that the first two portions of this paradigm were mostly completed by the NDRE and NRL staff, this study focused on data normalization, feature extraction, and statistical modeling for feature discrimination. The feature extraction process began by looking at relatively simple statistics of the signals and progressed to using the residual errors from auto-regressive (AR) models fit to the measured data as the damage-sensitive features. Data normalization proved to be the most challenging portion of this investigation. A novel approach to data normalization, where the residual errors in the AR model are considered to be an unmeasured input and an auto-regressive model with exogenous inputs (ARX) is then fit to portions of the data exhibiting similar waveforms, was successfully applied to this problem. With this normalization procedure, a clear distinction between the two different structural conditions was obtained. A false-positive study was also run, and the procedure developed herein did not yield any false-positive indications of damage. Finally, the results must be qualified by the fact that this procedure has only been applied to very limited data samples. A more complete analysis of additional data taken under various operational and environmental conditions as well as other structural conditions is necessary before one can definitively state that the procedure is robust enough to be used in practice.

State analysis of nonlinear systems using local canonical variate analysis

Time series analysis is concerned with the evolution of univariate or multivariate time series and especially with the functional form underlying the evolution of the time series values. Key issues include the determination of state rank, Lyapunov exponents, attractor form, and prediction of future values. Numerous papers have proposed procedures for prediction of future time series values based on past response values. We propose and demonstrate a method for deriving an approximate state space model of a nonlinear system from time series data. The data may be multivariate, and may include both input and response values. The method, which we call local canonical variate analysis, estimates state rank, describes the state evolution, and predicts future response values. A detailed background of local canonical variate analysis is provided and the procedure is applied to several time series generated by nonlinear oscillatory systems.

Statistical validation of a plate finite element model for damage detection

It is common practice in applied mechanics to develop finite element models for mechanical system behavior. Most structural integrity monitoring techniques, proposed to date, rely on an accurate model of the structure at hand. In many situations the structure being monitored is already built; in those cases, it is good engineering practice to ensure that the finite element model matches the behavior of the physical structure. However, no general-purpose technique exists or formally, statistically judging the quality of the finite element model. This paper applies a formal statistical procedure for the validation of finite element models of structural systems, when data taken during operation of the system are available. The statistical validation procedure is based on the bootstrap, and it seeks to build a tool for assessing whether or not a finite element model is an acceptable representation of the structure. The approach uses experimental data to construct confidence bounds that permit the assessment of the model. The case of a finite element model of an aluminum plate is presented.