ENBIS-11 in Coimbra

4 – 8 September 2011 Abstract submission: 1 January – 25 June 2011

Using Data Mining Modelling for Process Improvement on Polyester Foams Type – a Case Study

6 September 2011, 16:05 – 16:25

Abstract

Submitted by
Nuno Antonio
Authors
Silvio Costa (Flex 2000) Nuno António (StatSoft)
Affiliation
Flex2000/StatSoft
Abstract
The slabstock flexible foam industry Manufacturing Processes can be influenced by a large number of factors related to raw materials quality, physical attributes, formulations, engineering decisions. Data Mining algorithms can play an important role on process development because they show more flexibility to fit even very non-linear data while keeping a satisfactory predictive capacity.
This project included an holistic approach to the process improvement of the polyester flexible foam which included several steps:
- Merging data, with many different formats, stored on different sources (certificates of analysis, environmental conditions, machine settings, chemical formulations);
- Performing variable screening as well as checking for variable redundancies in order to prevent collinearity problems;
- Fitting data mining models to predict the probability of obtaining a non-conformity using several data mining algorithms;
- Evaluating the models quality through V-fold cross-validation and residual analysis;
- Using a multi-criteria approach to select the best model in order to balance predictive power with the usage of actionable variables;
- Deploying the final model on the real production environment as a management tool for the Production department of Flex2000.
This project is a joint work of Flex2000 Quality department and StatSoft Ibérica analytical team.

Return to programme