Estimation of flash point and autoignition temperature of organic sulfur chemicals Mehdi Bagheri a , Tohid Nejad Ghaffar Borhani b , Gholamreza Zahedi b,⇑ a Young Researchers Club, Islamic Azad University, Science and Research Branch, Tehran, Iran b Process Systems Engineering Center (PROSPECT), Faculty of Chemical Engineering, Universiti Teknologi Malaysia, 81310 Skudai, Johor, Malaysia article info Article history: Received 22 September 2011 Received in revised form 22 January 2012 Accepted 22 January 2012 Available online 28 February 2012 Keywords: Organic sulfur chemicals Multivariate molecular modeling Artificial neural network Flash point Autoignition temperature Particle swarm optimization abstract The combustible nature of organic sulfur containing chemicals demands an accurate hazardous knowl- edge for their safe handling and application in industries and researches. In this work, a quantitative structure–property relationship (QSPR) study was performed to thoroughly investigate such crucial haz- ardous properties i.e., flash point (FP) and autoignition temperature (AIT) of the organic sulfur chemicals which are comprising a wide range of mercaptans, sulfides/thiophenes, polyfunctional C,H,O,S material clas- ses. Based on multivariate linear regression (MLR) the multivariate model was gained using a robust bin- ary particle swarm optimization (PSO) for the feature selection step, the three molecular descriptors were realized as the most responsible descriptors for the flammability behaviors of such chemicals. Next, a three-layer feed-forward neural network model (ANN model) was utilized. The implemented multivari- ate linear regression and three-layer feed-forward neural network models were practically able to predict the flammability characteristics of a diverse range organic sulfur containing chemicals with high accu- racy. The results for PSO-MLR model illustrated that the squared correlation coefficient (R 2 ) between pre- dicted and experimental values were 0.9286 and 0.9259 for FP and AIT, respectively. The results for ANN model showed that the squared correlation coefficients (R 2 ) were 0.9858 and 0.9889 for FP and AIT, respectively. The ANN model of FP and AIT is more accurate than the multivariate model, and the PSO-MLR model is more simple and touchable. Ó 2012 Elsevier Ltd. All rights reserved. 1. Introduction The class of organic sulfur containing is used in a phenomenal range of important applications covering the production of phar- maceuticals, petrochemicals, agrochemical supplements, polymer- ization modifiers, and gas odorants to many other mining industrial chemicals [1–3]. Due to the combustible and flammable nature of the organic sulfur containing chemicals the safety knowl- edge have the crucial importance in several aspects [4]. There are several essential parameters used to classify chemicals according to their degree of flammability along with flash point (FP) and autoignition temperature (AIT), which are well accepted as the most important parameters to evaluate the ability of flammable chemicals to form an explosive atmosphere. As a result, this flam- mability knowledge can greatly benefit for safe handling, process- ing, transportation, and storage of the organic sulfur chemicals [5]. The American Society for Testing and Materials (ASTM) defines flash point as the lowest temperature at which the application of an ignition source causes the vapors of a sample specimen to ignite under specified testing conditions and 101.3 kPa pressure [6]. The FP data is widely used to evaluate the fire and explosion hazards of liquids and has great practical significance in the handling and transporting of such chemicals in bulk quantities. The AIT is defined as the lowest temperature at which the sub- stance will produce hot-flame ignition in air at atmospheric pres- sure without using any external ignition source such as a spark or flame [6]. The AIT is a criteria of the temperature at which a material will spontaneously burst into flames when exposed to the atmosphere and is also crucial for the performance of internal combustion engines through the phenomenon of engine knock [7]. Experimental flammability data is the main source of the safety information used in hazardous evaluations. However, the experimental measurements of FP and AIT data is expensive, te- dious, and time consuming, and for toxic, volatile, explosive, and radioactive chemicals, the measurement is more difficult and even impossible [8–11]. In addition, as an example of experimental uncertainties, the measurement of AIT is very dependent on the apparatus and the test methods employed, and the measured AIT values reported by different authors are often quite different (i.e. sometimes vary by hundreds of degrees) [10]. Similarly, in the case of flash point depending on the design of test apparatus the exper- imental FP data is often inconsistent to some extent [12]. Therefore, in order to support and expand the FP and AIT dataset using in industry, the development of theoretical prediction models which are desirably convenient and reliable for prediction is required [11–14]. By having a reliable model, determination of properties 0196-8904/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved. doi:10.1016/j.enconman.2012.01.014 ⇑ Corresponding author. Tel.: +60 75502371; fax: +60 75566177. E-mail address: grzahedi@cheme.utm.my (G. Zahedi). Energy Conversion and Management 58 (2012) 185–196 Contents lists available at SciVerse ScienceDirect Energy Conversion and Management journal homepage: www.elsevier.com/locate/enconman