Enhanced sparse regularization for structural damage detection based on statistical moment sensitivity of structural responses
Haifeng Bu, Dansheng Wang
Abstract
A novel structural damage identification technique based on enhanced sparse (ℓ1/2) regularization is proposed, which is integrated with the fourth-order statistical moment of dynamic response and its sensitivity matrix. ℓ2, ℓ1, and ℓ1/2 regularization schemes are derived and compared for structural damage identification, respectively. It is found that ℓ2 regularization needs more iterations for convergence and cannot obtain sparse damage solution well. ℓ1 regularization is better than ℓ2 on efficiency and accuracy, but it is more sensitive to noise than ℓ2 and ℓ1/2 regularizations. ℓ1/2 regularization has the sparsest solution among three regularization methods for structural damage identification and also needs much less iterations than ℓ2 and ℓ1 regularizations. Two numerical examples of a simply supported beam and a shear-type frame are implemented to demonstrate that ℓ1/2 performs the best on accuracy, efficiency, and steady when compared with ℓ2 and ℓ1 regularizations. Experimental results for a three-story frame also verify the effectiveness and efficiency of the proposed damage identification technique.