Damage detection with interval analysis for uncertainties quantification

Infrastructure damage detection is widely adopted to prevent structural collapse and cut down the maintenance for owners with timely repair, but most damage detection methods only obtain marginal performance for in-situ structures due to uncertainties. A non-probabilistic damage detection method for uncertainty quantification is presented in this study. The diagnosis elements are extracted from the coefficient matrix of the vector auto-regressive (VAR) model which is identified from the measured acceleration, and the Mahalanobis distance (MD) of these diagnosis elements between pristine and unknown condition is employed as damage feature. The interval of MD due to physical variability is computed by optimal interval analysis with the differential evolution algorithm. A modified receiver operating characteristic (ROC) curve is developed and the area under ROC is utilized to localize damage. The overlap rate (OR) of MD interval are defined to quantify damage severity. The numerical simulation results demonstrate that the interval analysis method can successfully detect damage when the physical variability is considering