Predicción de propiedades fisicoquímicas de coques obtenidos en procesos de conversión de fondos de vacío usando espectroscopia raman y quimiometría
En el presente trabajo se combinaron los beneficios de la quimiometría con la facilidad y rapidez de la técnica espectroscópica Raman, para el desarrollo de cuatro modelos que predicen propiedades fisicoquímicas en coque de petróleo de manera confiable y en cortos periodos de tiempo. Mediante el aná...
- Autores:
-
Moscoso Almeida, Laura Sofia
- Tipo de recurso:
- http://purl.org/coar/version/c_b1a7d7d4d402bcce
- Fecha de publicación:
- 2016
- Institución:
- Universidad Industrial de Santander
- Repositorio:
- Repositorio UIS
- Idioma:
- spa
- OAI Identifier:
- oai:noesis.uis.edu.co:20.500.14071/34974
- Palabra clave:
- Coque De Petróleo
Análisis Multivariable
Quimiometría
Espectroscopia Raman
Fondos De
In this work
the benefits of chemometrics were combined with the ease of use and speed of Raman spectroscopy for the development of four models able to predict physical-chemical properties in a fast and reliable way for petroleum coke samples. By employing principal component analysis (PCA) and partial least square regression (PLS) the correlation between spectral data and the experimental values for heat of combustion and proximate analysis was found from 56 samples of petroleum coke obtained from hydrothermal and thermal cracking conversion processes of vacuum residua. The properties fixed carbon (CF)
volatile matter (MV)
heat of combustion (PCal) and ashes (CZ) for each sample were determined in previous works according to the standards established by the American Society for Testing and Materials (ASTM). The statistical parameters obtained in the four final PLS models showed adequate adjustment of data to the models; all correlation coefficients (R2) among reference and predicted values were above 0.95
mean square errors of validation were low (below 1.00)
cross validation for all PLS models showed satisfying results in the prediction of the properties CF
MV
PCal and CZ with low error percentages (below 2.00% in average). Likewise
the reported global errors for the CF
MV and PCal prediction models are less than 5.00% and confirm the reliability of these models. In spite of having obtained appropriate statistical parameters for the CZ prediction model
it was regarded as a non-reliable prediction model
since it showed a high global error taking into account that the range in which this property™s experimental values were measured was 3.33 ± 1.88.
- Rights
- License
- Attribution-NonCommercial 4.0 International (CC BY-NC 4.0)