Subspace-based damage localization using Artificial Neural Network

In this paper, an Artificial Neural Network (ANN) based approach using a new non-parametric residual, as input, is presented for damage diagnosis. The residual is associated with Observability null-space of the system and is generated by using parity matrices, obtained from covariance driven output-only Subspace Identification (SubID). The proposed residual is compared with existing subspace based damaged indicators by using a simple numerical example. For damage localization a modal based approach is adopted, where a Finite Element (FE) model is employed to simulate the temporal response of a structure under different excitation conditions and damage scenarios. Training of ANN is established using residuals generated from these simulated responses. This trained ANN is in turn used to locate, in semi-real time, the predefined damage types. The effectiveness of this algorithm to identify damage is studied experimentally by localizing single edge cracks in a thin aluminum plate.