This paper develops a fault diagnosis methodology for civil engineering structures based on the bond graph approach. The bond graph theory provides a modeling framework that includes parametric models of the physical system and the sensors. Structural faults are modeled as abrupt or gradual damage in structural components. Sensor faults are modeled as biases or drifts from true measurements. Fault detection uses a statistical method to identify significant deviations of measurements from nominal behavior of the structure. Fault isolation is carried out by comparing predicted effects of hypothesized faults with observed behavior of the structure. Numerical illustrations of fault diagnosis of a frame struc- ture driven by time-varying loads are provided. (Rosenberg and Karnopp 1983). It is based on modeling energy flow between system components and inherently enforces continuity of power and conservation of energy. Bond graphs provide a systematic framework for building consistent and well constrained models of dynamic physi- cal systems across multiple domains that include electrical, mechanical, hydraulic and thermal systems. The topologi- cal structure of BG models causality constraints that pro- vide the mechanisms for effective and efficient fault diag- nosis based on cause and effect analysis. Our diagnosis algorithm is based on the TRANSCEND diagnosis framework (Mosterman & Biswas 1999). The diagnosis models, temporal causal graphs (TCGs), are de- rived systematically from BG models and provide the tem- poral and causal relations between deviant observations made on the system during its operation and hypothesized faults. Residuals are computed as deviations in measure- ments from nominal behavior. In an ideal or undamaged system the residuals should be zero. Nonzero residual val- ues imply faults in the structural components or in the sen- sors. An abrupt fault is a sudden change in a system pa- rameter (e.g., a sudden reduction in the stiffness or the damping of a structural member). Incipient faults result from slow variation in any of the system parameters over time that causes the system behavior to drift from its steady state (e.g., degradation or corrosion of steel bars over a long period of time). We assume faults are persistent. This paper presents a methodology for applying bond graph theory to damage diagnosis for civil structures. The TRANSCEND algorithms are applied to a new domain, namely building structures. The paper focuses on detection and isolation of faults in structural components and sen- sors. For simplicity, we make the single fault assumption, but the methodology can be extended to multiple faults (Daigle et al. 2006).
[1]
Sankaran Mahadevan,et al.
Bayesian wavelet packet denoising for structural system identification
,
2007
.
[2]
Hojjat Adeli,et al.
Pseudospectra, MUSIC, and dynamic wavelet neural network for damage detection of highrise buildings
,
2007
.
[3]
D. L. Livesey.
Introduction to Physics
,
1993
.
[4]
Pieter J. Mosterman,et al.
Diagnosis of continuous valued systems in transient operating regions
,
1999,
IEEE Trans. Syst. Man Cybern. Part A.
[5]
Gautam Biswas,et al.
Distributed diagnosis of coupled mobile robots
,
2006,
Proceedings 2006 IEEE International Conference on Robotics and Automation, 2006. ICRA 2006..
[6]
Pieter J. Mosterman,et al.
A Method for Efficient Simulation of Hybrid Bond Graphs
,
2006
.
[7]
Harley Earl Howe,et al.
Introduction to physics
,
1948
.
[8]
Achintya Haldar,et al.
Health Assessment at Local Level with Unknown Input Excitation
,
2005
.
[9]
Achintya Haldar,et al.
Element-level system identification with unknown input
,
1994
.
[10]
B. Peeters,et al.
Stochastic System Identification for Operational Modal Analysis: A Review
,
2001
.
[11]
R. Clough,et al.
Dynamics Of Structures
,
1975
.