A collaborative machine learning-optimization algorithm to improve the finite element model updating of civil engineering structures
Finite element model updating has become a key tool to improve the numerical modelling of existing civil engineering structures, by adjusting the numerical response to the observed experimental behaviour of the structure. At present, model updating is mostly conducted using the maximum likelihood me...
| Autores: | , , , |
|---|---|
| Tipo de recurso: | artículo |
| Estado: | Versión aceptada para publicación |
| Fecha de publicación: | 2020 |
| País: | España |
| Institución: | Universidad de Sevilla (US) |
| Repositorio: | idUS. Depósito de Investigación de la Universidad de Sevilla |
| OAI Identifier: | oai:idus.us.es:11441/180728 |
| Acceso en línea: | https://hdl.handle.net/11441/180728 https://doi.org/10.1016/j.engstruct.2020.111327 |
| Access Level: | acceso abierto |
| Palabra clave: | Multi-objective harmony search optimization Machine learning Collaborative algorithm Best Pareto solution Finite element model updating Maximum likelihood method |
| Sumario: | Finite element model updating has become a key tool to improve the numerical modelling of existing civil engineering structures, by adjusting the numerical response to the observed experimental behaviour of the structure. At present, model updating is mostly conducted using the maximum likelihood method. Following this approach, the updating problem can be transformed into a multi-objective optimization problem. Due to the complex nonlinear behaviour of the resulting objective functions, metaheuristic optimization algorithms are normally employed to solve such optimization problem. However, and although this is nowadays a well-established technique, there are still two main drawbacks that need to be addressed for practical engineering applications, namely: (i) the high simulation time required to compute the problem; and (ii) the uncertainty associated with the selection of the best updated model among all the Pareto optimal solutions. In order to overcome these limitations, a new collaborative algorithm is proposed herein, which takes advantage of the collaborative coupling among two optimization algorithms (harmony search and active-set algorithms), a machine learning technique (artificial neural networks) and a statistical tool (principal component analysis). The implementation details of our proposal are discussed in detail throughout the paper and its performance is illustrated with a case study addressing the model updating of a real steel footbridge. Two are the main advantages of the newly proposed algorithm: (i) it leads to a clear reduction of the simulation time; and (ii) it further permits a robust selection of the best updated model. |
|---|