| Before about the middle of this century, train inspectors known as carknockers were a common sight at railroad yards and stations [18]. The carknockers looked for structural flaws in the wheels of the cars by walking along the length of the train and banging each wheel with a hammer. Their trained ears could detect damage in the wheel from the resulting sound. This thesis presents a damage detection filter which performs the same function as the carknocker's ear: the filter detects the onset and location of damage in a structure, in real-time.; A damage model based on a structural connectivity matrix is proposed, and justified using the stiffness (or displacement) method from structural mechanics. This connectivity matrix makes it possible to treat the forces resulting from damage as inputs to the system, which allows the filter to detect any form of damage: partial or complete, gradual or sudden, and linear or nonlinear.; Damage detection filters are based on model-based state-space observers (or estimators). Design of the filters consists of finding feedback gains and a filter output gain which make the filter error transfer matrix block-diagonal. Necessary and sufficient conditions are given for the existence of these gains, and a parameterization for the feedback gains is derived. It is shown that a certain number of the filter poles can be placed in desired locations in the complex plane by the proper choice of gains.; The method is demonstrated on an 8-degree-of-freedom analytical test structure with general (i.e. non-proportional) damping, and simulated time responses are presented. A damage detection filter is also designed for a 42 degree-of-freedom finite element model of a test structure, and it is shown that the filter successfully indicates damage in a subset of truss members when one or more of these members is weakened in the model. Finally, the same filter is used to demonstrate damage detection on the real test structure. |
