An unsupervised statistical damage detection method for structural health monitoring(applied to detection of delamination of a composite beam)

Abstract

The present paper proposes a new damage diagnosis method for structural healthmonitoring that does not require data on damaged-state structures. Structural healthmonitoring is an essential technology for aged civil structures and advanced compositestructures. For damage diagnostic methods, most current structural health monitoringsystems adopt parametric methods based on modeling, or non-parametric methods such asartificial neural networks. The conventional methods require FEM modeling of structure ordata for training the damaged-state structure. These processes require judgment by ahuman, resulting in high cost. The present paper proposes a new automatic damagediagnostic method for structural health monitoring that does not require these processes byusing a system identification and statistical similarity test of the identified systems using anF-test.As an example of damage diagnosis using the new method, the present study describesdelamination detection of a CFRP beam. System identification among the strain datameasured on the surface of a composite beam is used for damage diagnosis. The resultsshow that the new statistical damage diagnostic method successfully diagnosesdamage without the use of modeling and without learning data for damagedstructures.