Uploaded February 2021 | Updated September 2026, 7 hours ago
In this video, the idea of using Artificial Neural Networks (ANNs) to assess the fidelity of Real-Time Hybrid Simulation/Substructuring (RTHS) is proposed. It is investigated, whether ANNs could be a meaningful tool to predict the test fidelity---when no reference solution is avaialble---solely on data that can be measured during the test. An ANN was trained on data from 280 simulated RTHS tests and applied to data from real RTHS tests. The results revealed that the training process is successful and a relation between the selected input features (error indicators from literature, dynamical properties of the investigated system) and the target (test fidelity) found.
The application of the trained ANN to a different dynamical system showed that more data have to be included in the training process such that the fidelity of never-seen dynamical systems can be predicted robustly. Therefore, a next step could be to gather data from historical RTHS tests and use all available tests for the training process.
The work was presented at the conference IMAC-XXXIX.
In this video, the idea of using Artificial Neural Networks (ANNs) to assess the fidelity of Real-Time Hybrid Simulation/Substructuring (RTHS) is proposed. It is investigated, whether ANNs could be a meaningful tool to predict the test fidelity---when no reference solution is avaialble---solely on data that can be measured during the test. An ANN was trained on data from 280 simulated RTHS tests and applied to data from real RTHS tests. The results revealed that the training process is successful and a relation between the selected input features (error indicators from literature, dynamical properties of the investigated system) and the target (test fidelity) found.
The application of the trained ANN to a different dynamical system showed that more data have to be included in the training process such that the fidelity of never-seen dynamical systems can be predicted robustly. Therefore, a next step could be to gather data from historical RTHS tests and use all available tests for the training process.
The work was presented at the conference IMAC-XXXIX.
