Using support vector regression for metal foam control
Profile monitoring focuses on the process or product variables that are characterized by a functional relationship of this variable with respect to time or space. The objective of this research is to develop a methodology based on Support Vector Regression, SVR, for no linear profiles monitoring and...
| Autores Principales: | Sanabria-Castro, Alexis, Meneses-Guzmán, Marcela, Chiné-Polito, Bruno |
|---|---|
| Formato: | Artículo |
| Idioma: | Español |
| Publicado: |
Editorial Tecnológica de Costa Rica (entidad editora)
2022
|
| Materias: | |
| Acceso en línea: |
https://revistas.tec.ac.cr/index.php/tec_marcha/article/view/5891 https://hdl.handle.net/2238/14117 |
| Sumario: |
Profile monitoring focuses on the process or product variables that are characterized by a functional relationship of this variable with respect to time or space. The objective of this research is to develop a methodology based on Support Vector Regression, SVR, for no linear profiles monitoring and implement it to the density profiles of a cellular material, aluminum metal foam. The shape of a profile is associated to certain mechanic characteristics of a product which means that a significant change in the shape would be detected as an out-of-control observation by a monitoring method designed for this purpose; if this happened, it can be concluded that the mechanic properties would be different from those required. The methodology considers the estimation of percentile curves that will be the basis to define the control chart limits, and the estimation of the parameter Cost and Sigma of a SVR model with a Gaussian Kernel with the aid of cross-validation. The performance of the established control chart is evaluated supported by the boostrapping technique. The proposed method is practical and of simple interpretation. According to the results, if the shape of the density profile were to change beyond the shape indicated by the natural variability of the profile, the implemented method would detect it as out-of-control with a type I error of 0.341% (ARLreal = 293). |
|---|