SS-DAC: A systematic framework for selecting the best modeling approach and pre-processing for spectroscopic data

Selecting the best combination of pre-processing and modeling methodologies is a critical activity while building soft sensors from spectroscopic Process Analytical Technology data. To help practitioners in this task, a framework for Soft Sensor Development, Assessment and Comparison (SS-DAC) is pro...

Full description

Saved in:
Bibliographic Details
Published in:Computers & chemical engineering Vol. 128; pp. 437 - 449
Main Authors: Rato, Tiago J., Reis, Marco S.
Format: Journal Article
Language:English
Published: Elsevier Ltd 02.09.2019
Subjects:
Industrial data science
Model selection
Process Analytical Technology
Soft sensor
Spectral pre-processing
Spectral pre-processing
Soft sensor
Industrial data science
Process Analytical Technology
Model selection
ISSN:0098-1354, 1873-4375
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Selecting the best combination of pre-processing and modeling methodologies is a critical activity while building soft sensors from spectroscopic Process Analytical Technology data. To help practitioners in this task, a framework for Soft Sensor Development, Assessment and Comparison (SS-DAC) is proposed. By applying SS-DAC, the following goals are achieved: the models’ parameters are set to optimal performance; the models’ predictive performance is statistically compared against each other; the statistical analyses are summarized through graphical displays. SS-DAC is designed to accommodate different pre-processing and modeling methodologies, and it can be applied to any type of linear and nonlinear models. To exemplify the application of SS-DAC, 13 pre-processing and three modeling methodologies were compared on three real case studies over a total of 13 response properties. For all examples, SS-DAC provided useful insights on the best subset of models and highlighted trends on the impacts of the different pre-processing and modeling methodologies.
ISSN:0098-1354
1873-4375
DOI:10.1016/j.compchemeng.2019.05.036