291899 Prediction of Onset Temperature for Thermal Decomposition of Organic Peroxides Using Quantitative Structure-Properties Relationship

Monday, October 29, 2012
Hall B (Convention Center )
Taufik Ridha, Chemical Engineering, Texas A&M University, College Station, TX, Yi Liu, Mary Kay O'Connor Process Safety Center, Texas A&M University, College Station, TX, Xiaodan Gao, Chemical Engineering, Mary Kay O’Connor Process Safety Center,Texas A&M University, College Station, TX and Dr. M Sam Mannan, Artie McFerrin Department of Chemical Engineering, Mary Kay O'Connor Process Safety Center, College Station, TX

Thermal decomposition of organic peroxides, due to the unstable Oxygen-Oxygen bond, may lead to a runaway reaction. Understanding the reactivity hazards of organic peroxides is crucial to develop an inherently safer design (ISD). Previously, Quantitative Structure Properties Relationship (QSPR) approach was used to develop a prediction model for the onset temperature (Yuan et al.  Ind. Eng. Chem. Res.2011, 50, 1515). Although the model was cross validated on the training data, no validation on external test data was conducted. Also, the large number of descriptors is comparable to the number of training data, which might cause over fitting. The lack of enough typical training data may also lead to poor representation.

In this work, T-test was performed on additional gathered data to check the effectiveness and six of them were selected as a test data set afterwards. A poor prediction of the previous model on Peroxycarbonate group was found, and the residual error was as large as 18%. After updating the training data set by adding one representative data related to the peroxycarbonate group and adopting a different set of descriptors with different formulas, two new models were found. One model, with a reduced set of seven descriptors in a linear formula, has better statistical values and predictions than the previous model with a large set of 13 descriptors. The other model, with the same set of descriptors but a quadratic formula, showed better predictions in both the training and test sets. The maximum residual error is reduced to 6%, and the average absolute relative deviation decreased from 4% to 3%. This study improved the QSPR methodology to develop new practical models for predicting onset temperature of organic peroxides that may assist in the development of inherently safer process.

Extended Abstract: File Not Uploaded
See more of this Session: Student Poster Session: Environmental
See more of this Group/Topical: Student Poster Sessions