Quasi real-time FEM calibration by 3D displacement measurements of large shaking table tests using HPC resources

Abstract

The present paper concerns the implementation of a methodology aiming to achieve a quasi real-time calibration of Finite Element Models (FEMs) of large structural mock-ups during shaking table tests. The damage achieved after each test step is commonly evaluated by visual inspection, since running large numerical analyses or processing the experimental data for damage estimation with common computing resources is very time-consuming and in case of a prolonged stand-by the shaking tables oil temperature stability can be put at risk. On the other hand, the specimen damage level can be monitored through the modal parameters estimated using Finite Element Analyses (FEAs), but FEMs require to be calibrated to provide accountable results.In the proposed approach the FEM calibration was carried out using 3D motion data of a large number of passive markers. They were acquired and processed by a dedicated Displacement Data Processing (DDP) procedure combining the Savitzky-Golay filtering for data noise reduction and the convolution derivation approach for the extraction of the motion parameters. To obtain results in reasonable time between test steps (a few minutes) the methodology exploits the hardware and software resources available on the ENEA High-Performance Computing (HPC) system. Also, the proposed approach allowed to integrate within a single web interface the possibility to share the seismic experiments in real time, while providing updated FEM calibration at each step of the test sequence during the experimental session.