Linear Regression Based Damage Detection Algorithm Using Data from a Densely Clustered Sensing System

For the purpose of identify the early appearance of damage on under-service structures, cost-effective and precise ways of damage detecting are needed. In this thesis, a structural behavior based damage detection method is introduced. This method utilizes dynamic responses collected by a densely clustered sensor network on the structure and processes them with linear regression based algorithms. The regression coefficients in the linear regression function are used to calculate damage indicators in the algorithm. The coefficient of determination and the variable inflation factor are introduced to verify the reliability of each damage indicator to be used to identify damages. Also control indicators are calculated to reveal the level of noise in the tests and set up a base line for useful damage indicators.;Test specimen used in the research is a scaled two-bay steel frame at Lehigh University's ATLSS Center. A finite element model of the frame is also developed based on this frame. Tests are conducted on both the finite element model and the specimen. Results are analyzed and compared to check the performance of the algorithm. In the tests, the frame is subjected to a single horizontal input which is brought by an actuator. A sensor network with 21 accelerometers is implemented on the frame to collect acceleration responses that are perpendicular to the surface of the frame from 21 locations on the frame.;The effectiveness of three different linear regression based algorithm is discussed in this thesis. The single-variable linear regression based algorithm is first evaluated using both simulation data and test data. Then an advanced algorithm, the two-variable linear regression based algorithm, is introduced and assessed. The same data is again used and damage prediction results from this algorithm are compared with ones from the single-variable linear regression based algorithm. At last, based on the performance of the previous to algorithms, an iterative linear regression based algorithm is introduced and its performance on analyzing data from random input tests is presented showing the effectiveness of this method.