Structural Health Monitoring (SHM) at its core is a process to identify damage, defined as either
material or geometric change, of a system that negatively affects the systems performance. SHM
seeks to address four questions:
(Level 1) Detection: Is damage present?
(Level 2) Localization: What is the probable location of damage?
(Level 3) Assessment: What is the severity of the damage?
(Level 4) Prognosis: What is the remaining service life of the damaged system?
SHM can be used to monitor structures affected by external stimuli, long-term movement, material
degradation, or demolition. It can also assess integrity after natural disasters, lessen need for
maintenance and repair construction, provide information to improve future designs, and encourage
the move towards performance-based design philosophy instead of time-based design. Multiple
damage-detection methods are currently in practice including visual detection, acoustic emission,
ultrasonic methods, magnet particle inspection, radiography, eddy-current methods and thermal field
methods. These methods, though, require prior knowledge of the damage location and that the
inspected area is readily accessible. Therefore, research in the field of vibration-based inspection has
significantly increased in the past two decades as a way to detect damage.
Damage can alter stiffness, mass, and energy dissipation, which result in a change in the dynamic
characteristics of a system including natural frequencies, mode shapes, modal curvatures, and
flexibility. SHM methods identify one or more of these features using data obtained from sensors
attached to the system being investigated. Then using prior data from the undamaged state or from
computer modeling, irregular behavior is identified and statistical approaches are used to detect damage.
Each identified feature has its own advantages and disadvantages such as sensitivity to noise and
environmental effects, ease of calculation, and ability to answer the higher leveled questions mentioned
The majority of methods for structural health monitoring have observed the whole or global system.
Recently there has been a shift to methods focusing on smaller substructures of the system due to
several reasons. Classical global methods have been tested on small structural systems comprised of
only a few unknowns and degrees-of-freedom. In contrast, real-world structures are becoming larger
and more advanced. Due to ill-conditioned behavior in many global monitoring approaches, the large
increase in DOFs can create problems in convergence. In addition, local damage can introduce
insignificant changes in global modal data. A division of the whole structural system into smaller
substructures lessens the amount of unknowns and can treat local areas as isolated structures, therefore
allowing for more accurate results, improved identification of local damage, and easier convergence
to a solution. Other advantages include the need for smaller amounts of sensors, reductions
in modeling errors due to the smaller size and reduced complexity, and minimizing the need for
expensive computation otherwise needed for processing large amounts of data. Lastly, the use
of substructures can allow for output-only identification if the excitation is random or located only
outside the substructure; this is advantageous in the common case where ambient excitation is used
With both global and substructural methods, many problems arise due to variability in the field,
test procedures, data manipulation, as well as assumptions made and uncertainties of many aspects
involved. As opposed to computer models and laboratory testing, field systems are affected by environmental
and operational factors that impact SHM results such as temperature, humidity, changing
boundary conditions, ambient loading conditions, and mass loading effects. Each of these factors
can significantly change the response of a system potentially leading to a false-positive identification
of damage. Methods should seek to understand these factors so to avoid this problem.
Generating input forces can also produce challenges for monitoring. Ideally, testing of a system
should not take a bridge out of service. To achieve this, ambient excitation, though impossible to
measure, should be used and the chosen SHM methods should be able to use output-only data.
Ambient excitation cannot be dictated, so certain vibration modes may not be excited and as a result
not identifiable from the data. Furthermore, almost all SHM methods developed assumed linear
behavior, which is not always the case. In nonlinear behavior, modal parameters are sensitive to
excitation amplitude, which may again falsely detect damage. Related to linearity of a system,
numerical models represent linear behavior that may not be accurate. Additional challenges with
models arise with material properties and connections that are hard to verify and heterogeneity of
systems which are hard to replicate.
This paper proposes a substructural monitoring method able to detect the presence of damage using
the Extended Kalman Filter and State-Space analysis. The novelty of this method is in the simulation
of the interface degrees of freedom with modal coordinates for the global structure, such that internal
forces at the interface need not be calculated and measurements of the motion at the interfaces need
not be collected. This method’s objectives are to
1. Minimize the need for interface measurements, but also leverage interface measurements if available
by modeling internal forces as ”ground motions” at the interfaces
2. Avoid the requirement of measuring external forcing by regarding unknown forcing as noise and
estimating its value for each time step
3. Identify directly the stiffness of the structure as the damage-relevant feature for SHM
4. Operate efficiently and in near real-time
This thesis will discuss other methods and research in the field of substructural health monitoring,
provide a walk-through of the theory behind the proposed substructural monitoring method, discuss
application of the method on numerical models and its results, and then conclude with a summary,
setbacks, and potentials of this method.||en_US