Hybrid computational strategy for structural damage detection with short-term monitoring data

• Method to use measured data from multiple setup for damage detection.
• Local damages are evaluated from Pattern Search instead of inverse analysis.
• Multiple CPUs and GPUs are used in the parallel computation of force and model updating.

The scenario of damage detection of a structure without permanent structural health monitoring system is explored. Dynamic responses are collected from different measurement setups and excitations in different regions of a structure. All the measured data are analyzed together in the proposed strategy with consideration of the different excitations. Local damages in terms of physical structural parameters are evaluated from the direct analysis using the Pattern Search method with a hybrid parallel computing strategy using CPU and GPU instead of the usual inverse analysis. The objective function is the difference between the measured and calculated responses from an updated finite element model of the structure. The proposed approach is illustrated with a plane frame structure with different tests and measurements in three regions. The effect of the size of the Generating Matrix of the Pattern Search method is also investigated and discussed. There is noted and improved computation efficiency when a large Generating Matrix with 2000 columns is used. The simulations illustrate the feasibility of using a few sensors for the damage detection of a large-scale structure with short-term monitoring data.