 |
International Society of Science and Applied Technologies |
| An Efficient Inverse Solution for Mimo Dynamic Systems by a Moving Least Squares Meta-Model | ||||
| Author | Gordon J. Savage
|
|||
| Co-Author(s) | Young Kap Son
|
|||
| Abstract | In the inverse problem, the component parameters and excitations (e.g. inputs), are to be determined given corresponding system query responses (e.g. outputs). Most model-based solutions to the inverse problem involve optimization using the so-called forward model - typically a mechanistic model in some form. Recently, databased approaches, or model-free methods, have been invoked whereby feature extraction methods such as Support Vector Machines (SVM) and artificial neural networks (ANN) are used. Alternatively, in this paper, we develop an inverse solution for dynamic systems through simple least-squares methodology. More importantly, this paper develops a Moving Least Squares (MLS) approach that permits larger design space and greatly improved accuracy. Once the training has been completed (in the normal fashion) the input and output training data are interchanged to provide a new causal relationship. Now, the inputs can be found for given query outputs. To do this, the pertinent closest responses to the query data are found by using a novel distance metric created primarily from the integral of squared error. Singularvalue decomposition (SVD) makes any matrix inversion tractable. A nonlinear dynamic mechanical system is invoked to demonstrate both parameter design and excitation allocation. The new inverse MLS meta-model is compared to the original least squares models using data from mechanistic models for fidelity. The results show significant error mitigation with acceptable computation effort.
|
|||
| Keywords | Inverse problem, Dynamic systems, Moving least-squares meta-model, Distance metric, Singular value decomposition. | |||
| Article #: RQD2025-282 | ||||
Proceedings of 30th ISSAT International Conference on Reliability & Quality in Design |