Bridge Structural Identification for Condition Assessment

The overall project goal is to develop a methodology that makes structural identification (ST-ID) a useful and practical tool for bridge condition assessment. The six-step methodology is implemented by the investigators & collaborators as an integrative effort:

Structural Identification an Integrative Process
  1. A-priori Model Development
  2. Experimental Design
  3. Full-Scale Testing
  4. Data Processing -- tools
  5. Model Calibration and Parameter Estimation
  6. Utilization
[ Back to Research or SCA Home ]

ST-ID: AN INTEGRATIVE PROCESS

(ref: "Structural Identification:  A Tool For Bridge Reliability Evaluation")

A six-step approach proposed for St-Id is listed above.  The quality of each step is improved by integrating experimental and analytical components.  Any step may be repeated to ensure that the geometric model converges to a mathematically optimal and a physically realistic solution.

A-priori models (Step 1) are developed at the beginning of the process to represent the best knowledge about the structure/connections/foundation/soil systems construction.  The finite element (FE) model must be capable of simulating the system's true behavior while being adaptable to the optimization procedure used in parameter estimation.

[ top ]

Experiment design (Step 2) requires selecting the inputs (loads, excitation, and temperature) and sensors (used to record input and response measurements) with consideration of hardware , software, and information technology constraints.  The load and sensor locations and data acquisition systems are selected to obtain a set of data meeting optimality criteria considering both required bounds of confidence for parameter estimation and test feasibility. Sensors have different installation, durability/ruggedness, signal conditioning, rate, range, linearity, sensitivity, resolution, precision, and accuracy attributes. In addition, the information technology requirements of different experiment designs are a significant issue and may vary greatly. Therefore, matching each sensor to the measurement that is needed, and designing the corresponding information technology for an optimal experiment is an art that requires extensive experience.  Thorough analytical evaluations are needed for each design scenario to determine if the finite element model and parameter estimation technique can support the experimental limitations such as expected noise level on the measurements.  Error sensitivity analyses are required to determine the required confidence bounds for the structural parameter estimates.  Sanayei and Saletnik (1996) developed a heuristic method for design of NDTs to reduce the error in the geometric parameter estimates caused by noisy measurements.

[ top ]

Full-scale tests (Step 3) include both instrumented monitoring (IM) for static tests and multi-reference impact tests (MRIT) for modal tests.  The tests are performed with different sets of reference sensors, impact loads, forced excitations, and static loads.  IM yields results usable for geometric parameter estimation.  MRIT results are used to verify global system behavior, localized mechanisms, and critical mechanisms of a structure.  The results from both static and modal tests can be used to verify the different testing methods.  MRIT and IM are both emerging arts in the context of field testing of civil engineered constructed facilities (Raghavendrachar and Aktan, 1992; Shelley et al., 1995).

[ top ]

Data processing (Step 4) of the experimental data is a critical step for error mitigation and quality assurance.  As a result, the data obtained from the experiment must be processed for use in the parameter estimation module such that the quality of the data is maintained.  Issues that need to be considered are: (1) Data must be validated to ensure that it originated from a reliable sensor; (2) Channels should be separated so that parameter estimation and verification can be performed in a clear checks and balances system; and (3) To ensure consistent conclusions, the results of different tests must be comparable.  Quality assessment of measurements is an extremely important factor in the data processing step.  Confidence factors for each sensor or data channel can be calculated using probabilistic methods, physical behavior, structural principles, and engineering judgment.  Matlab-based data processing tools are used to prepare the data for parameter estimation.

[ top ]

Model calibration and parameter estimation (Step 5) involve identifying critical geometric parameters using the processed static (Sanayei et al., 1997) and modal (Sanayei et al., 1998a) data from Step 4.  The calibration procedures refine the global and local characteristics of geometric FE models.  The refined model must be unique and represent a realistic physical structure.  Model reduction and expansion techniques can facilitate the estimation procedure.  Transformations that can convert a global model to a local model and vise versa need to be considered.  Finally, the calibrated model must be confirmed and validated.

[ top ]

Utilization (Step 6) of the calibrated FE models includes obtaining interpretations in a form that are useful to other researchers and practitioners for bridge management.  Once established for a specific bridge, calibrated models will consider design and/or construction defects and can be used as a baseline in the future for damage diagnoses and deterioration monitoring.  Here, a motivation is provided to use St-Id for the development of reliability indices descriptive of structural performance.  Additionally, when performed intermittently relative to a baseline, the updated geometric models should be able to identify, locate, and quantify damage making it possible to evaluate the impact of damage on global health.

There are a number of challenging issues within each step of the above-mentioned process that need to be considered.  These challenges include (1) the design and implementation of the experiment, (2) errors in the mathematical model used for parameter estimation, and (3) errors within parameter estimation.  Sanayei et al. (1998b) address these challenges in the context of a literature review.

[ top ]