Methods for real-time actuator-sensor-failure and structural damage detection
Doctor of Philosophy thesis
This research focuses on isolation and quantification aspects of structural damage detection using global damage detection methods. It includes two parts: (1) isolation and detection of failure of the actuator/sensor/structural elements in real-time using interaction matrix formulation and parity based method, and (2) isolation and quantification of the structural damage using a novel physical parameter realization method. The failure of actuator/sensor/structural element considered in this thesis can be any type of deviation from the normal operating conditions. First a novel algorithm is developed to isolate and detect the time instants of actuator failure based on the interaction matrix formulation, leading to one error function for each actuator. The existence of interaction matrix is extended to eliminate the state and the unexamined inputs from the error function. For measurement noise free case, the error function will have non-zero values when the examined actuator fails to follow the commanded input, regardless of the status of other actuators. Further the coefficients of the error function are directly calculated from the healthy input data, from the examined actuator, and all measured outputs. Thus the need to know the state-space model of the system is bypassed and it also avoids the errors incurred in the system identification step. Combining the interaction matrix formulation with inverse model, novel sensor failure detection algorithm is developed. The error function, one for each uncertain sensor, can isolate and detect the real-time failure of the examined uncertain sensor. Parity based method is extended for structural damage detection. By utilizing the geometric properties and frequency information incurred by the damaged elements, a robust and sensitive structural damage detection filter is developed. The effectiveness of the developed algorithms is demonstrated using numerical simulations and verified using experimental results. A novel off-line physical parameter realization method is developed in this thesis. By assuming that the geometric connections of the stiffness and damping matrices are known, the stiffness and damping matrices can be fully realized with limited number of inputs and outputs. Then the developed algorithm is applied for structural damage detection. Using numerical simulations it is shown that the method can detect, isolate, and quantify the severity of average in the stiffness and damping of structural elements.