hoeslywhyte

Optimal Sensor Placement and Damage Detection for Structural Health Monitoring: 

Optimal Sensor Placement and Damage Detection for Structural Health Monitoring REUJAT 2004 The University of Tokyo Bridge & Structure Laboratory Nathan Hoesly Catherine Whyte

Benefits of Structural Health Monitoring: 

Benefits of Structural Health Monitoring Continuous monitoring instead of periodic manual inspections. Measurement data can be used to detect damage that is not visible. Anticipated lower costs than visual inspection of structures.

Research Objectives: 

Research Objectives Use two types of models to study sensor placement. Simply supported beam model. Continuous beam bridge model. Implement three existing sensor placement techniques to identify damage induced in each model. Examine the importance of minimizing the number of sensors and finding the most effective placement.

Bill Emerson Memorial Bridge: 

Bill Emerson Memorial Bridge Pier 1 Pier 2 Bridge about 200 km south of St. Louis, and the subject of a collaborative project between WUSTL and Univ. of Tokyo. 32 element simply supported beam model. 30 element continuous bridge model (58 elements is too many for probabilistic method). Assembled using Finite Element Toolbox created by Juan Caicedo at WUSTL.

Sensor Placement Analysis (I): 

Sensor Placement Analysis (I) Effective Independence Method (Kammer) Fisher Information Matrix [[ψ]T[ψ]] Effective Independence Distribution Vector: The DOFs with the highest EfI values represent the best sensor locations.

Sensor Placement Analysis (II): 

Sensor Placement Analysis (II) Eigenvector Sensitivity Method (Shi) Fisher Information Matrix as distribution of strain energy [[ψ]T [K] [ψ]] The sensitivity matrix for one element is defined as     Then the contribution vector is defined as

Sensor Placement Analysis (III): 

Sensor Placement Analysis (III) Damage Measurability Method (Xia) Sensitivity of the damage to the noise in measurement is defined as   Then the damage measurability is defined as where {F} is the contribution of measurement points to a Fisher Information Matrix.

Eigenvector Sensitivity Method: 

Eigenvector Sensitivity Method 11 sensors 7 sensors 11 sensors 7 sensors Beam Model Bridge Model

Effective Independence Method: 

Effective Independence Method 11 sensors 7 sensors 11 sensors 7 sensors Beam Model Bridge Model

Damage Measurability Method: 

Damage Measurability Method 11 sensors 7 sensors 11 sensors 7 sensors Beam Model Bridge Model

Damage Detection: 

Damage Detection Damage toolbox (Damtool) created by Dr. Jerome Lynch at Stanford University used to determine whether the sensors identified the location of the damage correctly. Continuous damage monitoring can only use ambient excitation sources in most types of civil structures. The first four modes are used to perform damage detection. Each damage case takes about 20 minutes to analyze.

Damage Detection Analysis: 

Damage Detection Analysis Bayesian Probabilistic Method (Hoon) Bayes’ Theorem Error Function  

Numerical Experiment: Test Plan: 

Numerical Experiment: Test Plan Accelerometers record only vertical displacements. Each element has a length of 1 meter. 7 and 11 accelerometers, 1% and 2% noise, 7 sets. Random noise is added when mode shapes are computed. Each beam element defined by the Finite Element Model is damaged individually. All damage cases are defined as 10% of stiffness reduction of the selected element.

Damtool Procedure: 

Damtool Procedure

Damage Detection Tree: 

Damage Detection Tree

Damage Detection Tree: 

Damage Detection Tree

Damtool Report: 

Damtool Report

Damtool Report: 

Damtool Report

Beam Model: Eigenvector Sensitivity Method Results: 

Beam Model: Eigenvector Sensitivity Method Results

Bridge Model: Eigenvector Sensitivity Method Results: 

Bridge Model: Eigenvector Sensitivity Method Results

Eigenvector Sensitivity Method Conclusions: 

Eigenvector Sensitivity Method Conclusions Beam Model Most consistent method. Not as affected by noise as other two methods. Bridge Model Still stable. Problems identifying damage near the supports and center of the span.

Effective Independence Method Conclusions: 

Effective Independence Method Conclusions Beam Model Highly influenced by noise and number of sensors. Bridge Model Sometimes better results achieved with 7 sensors than 11 sensors or 2% noise than 1% noise.

Damage Measurability Method Conclusions: 

Damage Measurability Method Conclusions Beam Model Easily affected by noise. Damtool identifies many damage possibilities with the same probability so there is very little certainty in which is correct. Increased number of sensors did not necessarily improve the quality of damage detection. Bridge Model Performed similarly to eigenvector sensitivity method but fewer damage cases were identified in the center of the span.

General Conclusions: 

General Conclusions Difficult to locate damage in the regions near the supports and the center of the span. According to our study, the Eigenvector Sensitivity Method seems to perform the best for sensor placement in these structures. Further studies are needed to implement this probabilistic method in more complex structures such as cable-stayed bridges.

Many Thanks to:: 

Many Thanks to: Fujino Sensei (University of Tokyo) National Science Foundation Carlos Riveros (University of Tokyo) Dr. Shirley Dyke (WUSTL) Dr. Makola Abdullah (FAMU) Diego Giraldo (WUSTL) Juan Caicedo (WUSTL) Dr. Jerome Lynch (Stanford University) Terri Norton (FAMU)