Author's personal copy Thermochimica Acta 527 (2012) 59–66 Contents lists available at SciVerse ScienceDirect Thermochimica Acta jo ur n al homepage: www.elsevier.com/locate/tca Densities of ammonium and phosphonium based deep eutectic solvents: Prediction using artiﬁcial intelligence and group contribution techniques K. Shahbaz a , S. Baroutian b , F.S. Mjalli c,∗ , M.A. Hashim a , I.M. AlNashef d a Department of Chemical Engineering, University of Malaya, 50603 Kuala Lumpur, Malaysia b SCION, Te Papa Tipu Innovation Park, 49 Sala Street, Private Bag 3020, Rotorua 3046, New Zealand c Petroleum and Chemical Engineering Department, Sultan Qaboos University, 123 Muscat, Oman d Department of Chemical Engineering, King Saud University, 11421 Riyadh, Saudi Arabia a r t i c l e i n f o Article history: Received 18 August 2011 Received in revised form 4 October 2011 Accepted 5 October 2011 Available online 12 October 2011 Keywords: Density Deep eutectic solvent Artiﬁcial neural network Estimation a b s t r a c t As applications of deep eutectic solvents are growing fast as green alternatives, prediction of physical properties data for such systems becomes a necessity for engineering application designs and new process developments. In this study, densities of three classes of deep eutectic solvents, based on a phosphonium and two ammonium salts, were measured. Two predictive models based on artiﬁcial intelligence and group contribution methods were proposed for accurate estimation and evaluation of deep eutectic sol- vent densities. A feed forward back propagation neural network with 9 hidden neurons was successfully developed and trained with the measured density data. The group contribution method applied the modiﬁed Lydersen–Joback–Reid, Lee–Kesler and the modiﬁed Rackett equations. The comparison of the predicted densities with those obtained by measurement conﬁrmed the reliability of the neural network and the group contribution method with average absolute errors of 0.14 and 2.03%, respectively. Compar- ison of the model performances indicated a better predictability of the developed neural network over the group contribution method. © 2011 Elsevier B.V. All rights reserved. 1. Introduction Deep eutectic solvents (DESs) are gaining increasing attention in research and industry because of their potential as environmentally benign solvents and advantages over traditional ionic liquids such as non-toxicity, non-reactivity with water and most importantly being biodegradable [1]. In addition, they are prepared easily in high purity at low cost. One of the interesting features of using DESs is their potential as tunable solvents that can be customized to a particular type of chemistry [2]. DESs are created by mixing two or more components and they have low melting point compared to their constituting compounds. In other words, DESs are created from mixtures of an organic halide salt and an organic compound which is a hydrogen bond donor (HBD) able to form a hydrogen bond with the halide ion [3]. In spite of the importance of DESs and their interesting advan- tages, accurate values for many of their fundamental physical and chemical properties are either rare or even absent. The density is one of the important physical properties for DESs. DES den- sity data play a key role in engineering application designs of this environmental friendly solvent. However, it is not practical to ∗ Corresponding author. Tel.: +968 24142558; fax: +968 24141354. E-mail address: farouqsm@yahoo.com (F.S. Mjalli). experimentally evaluate density in many cases. Therefore a great need arises for developing estimation methods of acceptable accu- racy to fulﬁll this requirement [4]. Artiﬁcial neural networks (ANNs) is an estimation method which has been used widely for property predictions. ANNs are inspired and motivated by the structure and functional character- istics of human neurons and biological neural networks. Similar to the human nervous system, ANNs consists of an interrelated set of artiﬁcial neurons, and it processes information using a connection- ist approach to computation. ANNs are computing systems, which can be trained to learn a complex relationship between two or more variables or data sets [5]. The ANNs are good for tasks involv- ing incomplete data sets, fuzzy or incomplete information, and for highly complex and ill-deﬁned problems, where humans usually decide on an intuitional basis [6]. Among the available artiﬁcial neu- ral networks, the feed-forward neural network is one of the most important historical developments in neurocomputing [5]. Feed- forward neural network performs a weighted sum of its inputs and calculates an output using certain mathematical activation function embedded in its neurons. ANNs have been used in number of ﬁelds of engineering, environmental sciences, mathematics and economics [6–9]. The application of ANNs in prediction physical and chemical properties of materials is increasing rapidly. Many studies have been carried out to predict densities of various materials using ANNs [10–18]. It 0040-6031/$ – see front matter © 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.tca.2011.10.010