Prediction of the thermal decomposition of organic peroxides by validated QSPR models
Introduction
Organic peroxides are reactive compounds, containing the −O−O− bond [1], [2], that can be formed naturally by auto-oxidation with oxygen in certain solvents, such as diethylether. They can lead to highly explosive peroxidic residues, requiring specific safety precautions, like addition of oxidation inhibitors to prevent the formation of undesirable organic peroxides [3], [4], [5], [6]. Organic peroxides are more or less stable due to a relatively low −O−O− bond energy (20–50 kcal/mol) [7]. Since they generate instable radicals during their decomposition, organic peroxides are commonly used as catalyst and as radical polymerization initiators. The decomposition of organic peroxides can nevertheless be dangerous and can lead to serious effects [8], [9], [10]. To reduce the risk of incidents and of accidents, their hazards are intensively studied. In order to avoid the accidents by the customers and users, the organic peroxide producers provide information concerning the properties of commercial organic peroxides and give recommendations for safe handling and use of organic peroxides [11], [12]. General documents also describe the characteristics of organic peroxides and the safety rules to apply to safely handle them at laboratory scale [3].
Concerning the regulations, organic peroxides belong to a dedicated division (Division 5.2), as described in the UN Dangerous Goods Transportation Recommendations [13], or in GHS (Globally Harmonized System of classification and labelling of chemicals) [14] with 7 different classes (types A to G) related to their hazardous potential and leading to different amounts authorized for transport.
The decomposition of organic peroxides can be triggered and accelerated by heat, mechanical shock or friction and by various contaminants [15], [16]. To produce, transport and provide safely the numerous organic peroxides, the industry generally commercializes them in low concentration diluted in variable solvents.
In order to improve and homogenize the knowledge of the marketed chemicals, the European Union regulation REACH (Registration, Evaluation, Authorization and Restriction of Chemicals) [17], [18] requires the evaluation of physico-chemical, toxicological and eco-toxicological properties for all chemicals produced or imported by more than one ton by year in Europe. To help industry to meet the requirements of REACH regulation as far as the chemical safety assessment is concerned, ECHA technical guidances have been published [19], [20] and a general testing strategy for physico-chemical properties was proposed to consider the order of testing.
As far as organic peroxides are concerned, even if they belong to a dedicated regulated division or class, the knowledge of their explosive properties, as defined by the UN recommended tests [21], is of great importance. As entry data, their thermal stability (represented by the energy and temperature of decomposition) is a key property that is considered as a pre-selection criterion to identify substances that could undergo explosive reactions. So, it is used in the complex procedure of classification of explosives and organic peroxides [21]. Indeed, the UN regulation indicates that there is no need to perform this complex procedure when the decomposition heat (corresponding to the amount of energy released during the decomposition) is lower than 500 J/g. Measured by calorimetric analyses, notably by differential scanning calorimetry (DSC), the decomposition heat is estimated with measurement uncertainties of about 5–10% [22], [23] and less than 5 °C for the onset temperature [23].
For safety reasons but also for technical reasons, experimental tests can be difficult to implement for unstable substances like organic peroxides. As a consequence, the development of methods used for the prediction of data can be of great help at the research and development step and can help to accelerate and fulfil the next registration deadlines. As a simple prediction tool of reactivity hazards, the CHETAH software based on Benson's group contribution method was developed by ASTM [24]. Considering only six organic peroxides, Mohan et al. [25] demonstrated some correlations between CHETAH criteria (oxygen balance, the maximum decomposition heat, the difference between heat of combustion and decomposition heat) and explosive properties. In one recent study, Sato et al. [26] showed that there is a mutual correlation between CHETAH criteria and the explosibility of self-reactive substances except for organic peroxides and azo compounds. Nevertheless, it has to be noticed that the CHETAH software provides the maximum decomposition heat (considering that the available oxygen first oxidizes hydrogen to water and then carbon to carbon dioxide) and not the actual experimental decomposition heat.
Considering the REACH regulatory framework, Lewis et al. [27] advocated the use of powerful computer-aided ab initio techniques to generate predictions of key properties of broad classes of chemicals, without resorting to costly experimentation and potentially hazardous testing. Among these alternative methods to experimental testing, Quantitative Structure–Activity/Property Relationships (QSAR/QSPR) were clearly recommended in REACH and in technical guidances [19], [20] to obtain information data. Indeed, they represent powerful tools of prediction [28] used more and more for physico-chemical applications [29], [30]. Their applicability in an industrial context was also recently demonstrated by Patlewicz et al. [31]. To support the development and use of QSPRs, OECD drawn up 5 principles for the validation of QSAR/QSPR models for regulatory purpose [32].
Some recent reviews [33], [34] list the existing predictive models developed for the relevant properties of chemicals in the context of REACH. In particular, Dearden et al. [34] focused on the validation of models dedicated to physico-chemical properties according to OECD principles to favor the use of predicted property values in submissions to the European Chemicals Agency (ECHA).
Considering the prediction of thermal stability, some models exist for different families of compounds like nitroaromatic compounds [35], [36], [37], [38], [39], nitramines [40], [41], [42], ionic liquids [43] or polymers [44]. Nevertheless, to our knowledge, only one reference exists [45] to predict the reactivity hazards of organic peroxides using the QSPR approach. In this study, a limited database of 16 organic peroxides was used to derive models with no validation set, neither definition of applicability domain. Therefore, the robustness and the predictivity of these models must be at least validated and may be improved.
Consequently, the aim of this paper was to develop new robust QSPR models respecting OECD principles dedicated to the prediction of the heat and temperature of decomposition. To achieve this goal and considering that no large database exists in literature for organic peroxides (only 16 and 9 DSC data from the works of Lu et al. [45] and Ando et al. [22] respectively), an experimental database of 38 organic peroxides was built for these properties obtained in homogenous experimental conditions using DSC. The amount and quality of experimental data allowed to performing both internal and external validation methods to ensure the good performances of the multi-linear regression (MLR) models developed and then reach more robust and accurate models than the ones proposed by Lu et al.
Besides, the approach combined QSPR methodology with quantum chemical descriptors obtained with density functional theory (DFT) calculations. Indeed, a better chemical interpretation of the developed models can be expected using this type of descriptors as already demonstrated in previous works for nitro compounds [46], [47], [48]. To our knowledge, this work leads to the first completely validated QSPR models (including the definition of applicability domains) dedicated to the prediction of the thermal stability of organic peroxides.
Section snippets
Construction of the database
As mentioned above, the first work carried out on the prediction of thermal stability of organic peroxides [45] was based on only 16 samples. The concentration of the peroxides used ranged between 34% and 98% wt [45], with no indication of the dilution solvent or of the possible contaminants or co-products present in the samples.
In this study, to limit or even avoid any effect of the poor purity of organic peroxide samples, the origin of the organic peroxide samples was taken into account with
Partitioning of the dataset
The considered dataset of 38 organic peroxides was divided into a training set, containing two thirds of the molecules of the dataset and a validation set constituted by the remaining molecules. This partition enabled both sets to be of sufficient size with similar distributions to allow a robust development and an external validation of models. The partitioning of the dataset was also visually inspected to ensure that the validation set covered at best the chemical diversity of the domain of
DSC results
The reactivity data of organic peroxides are shown in Table 2, Table 3. The ranges of detected onset temperature (53–180 °C) and decomposition heat (441–2622 J/g) cover most of the commercial grade of organic peroxides. One example of DSC curve is given in Fig. 1. The concentration effect on the decomposition heat measured by unit mass of tested sample is linear in the range of the tested concentrations for the three tested organic peroxides (Fig. 2). That confirmed that the solvent did not
Conclusion
In this study, the largest experimental database including thermal properties for 38 organic peroxides obtained from homogeneous measurements was built to develop and validate the first QSPR models for the prediction of the heat and the onset temperature of decomposition. Considering more than 300 descriptors including quantum chemical ones calculated with the density functional theory, two MLR models presenting high performances were constructed according to all OECD principles for the
Acknowledgements
This work was performed using HPC resources from GENCI-CCRT (Grant 2013- t2013086639). The authors thank the ANR (PREDIMOL research project (ANR-10-CDII-007) including ARKEMA, Chimie ParisTech, CNRS, IFPEN, INERIS, Materials Design and University Paris Sud 11) for financing, AkzoNobel for providing free organic peroxides samples and ARKEMA Functional Additives Business Unit for providing most of Luperox® samples.
References (75)
Peroxides and peroxide-forming compounds
Chem. Health Safe.
(2001)- et al.
Homolytic bond-dissociation in peroxides, peroxyacids, peroxyesters and related radicals: ab-initio MO calculations
Tetrahedron
(1994) - et al.
Calorimetric studies and lessons on fires and explosions of a chemical plant producing CHP and DCPO
J. Hazard. Mater.
(2012) - et al.
Analysis of differential scanning calorimetric data for reactive chemicals
J. Hazard. Mater.
(1991) - et al.
Hazard evaluation of organic peroxides
J. Hazard. Mater.
(1982) - et al.
Integrating process safety with molecular modeling-based risk assessment of chemicals within the REACH regulatory framework: benefits and future challenges
J. Hazard. Mater.
(2007) - et al.
On the prediction of thermal stability of nitroaromatic compounds using quantum chemical calculations
J. Hazard. Mater.
(2009) - et al.
On the development of QSPR models for regulatory frameworks: the heat of decomposition of nitroaromatics as a test case
J. Loss Prev. Process Ind.
(2013) Some predictions in the field of the physical thermal stability of nitramines
Thermochim. Acta
(1997)- et al.
Quantitative structure–property relationship for thermal decomposition temperature of ionic liquids
Chem. Eng. Sci.
(2012)
Prediction of the thermal decomposition property of polymers using quantum chemical descriptors
Eur. Polym. J.
On the use of descriptors arising from the conceptual density functional theory for the prediction of chemicals explosibility
Chem. Phys. Lett.
Predictive QSPR models for impact sensitivity of nitroaliphatic compounds
J. Hazard. Mater.
The importance of the Domain of Applicability in QSAR Modeling
J. Mol. Graph. Model.
Initiators (Free Radicals)
Identification and handling of peroxidisable compounds, cahier de notes documentaires
Hygiène et Sécurité du travail
Organic peroxides and peroxy compounds –general description
Oxidation mechanism of diethyl ether: a complex process for a simple molecule
Phys. Chem. Chem. Phys.
Hazard ratings for organic peroxides
Proc. Safety Prog.
Case studies of incidents in runaway reactions and emergency relief
Proc. Safety Prog.
Thermokinetic model simulations for methyl ethyl ketone peroxide contaminated with H2SO4 or NaOH by DSC and VSP2
J. Therm. Anal. Calorim.
Thermal hazard evaluation of lauroyl peroxide mixed with nitric acid
Molecules
Prediction of explosibility of self-reactive materials by calorimetry of a laboratory scale and thermochemical calculations
Sci. Tech. Energ. Mater.
Best practices for QSAR model development, validation, and exploitation
Mol. Inf.
In silico prediction of physicochemical properties: European Commission
Joint Research Centre
Cited by (43)
Determination and discussion of apparent reaction kinetics of liquid diacyl peroxide synthesis in a microreactor
2024, Chemical Engineering ScienceEfficient continuous-flow synthesis of diacyl peroxide in a microreactor and the post-processing of its effluent
2023, Chemical Engineering ScienceDevelopment of small-scale experimental method for vent sizing and observation of runaway reaction
2022, Journal of Loss Prevention in the Process IndustriesCitation Excerpt :In addition, the reactivity and the behavior of the thermal decomposition of chemical substances may differ depending on the thermal history. Thermal behaviors of organic peroxides are continuing research in various aspects such as kinetics, molecular orbital study, CFD, machine learning, etc. (Gong et al., 2022; He et al., 2018; Prana et al., 2014; Lee et al., 2014; Zhang et al., 2022; Wang et al., 2018; Gao et al., 2015). As such, estimating the vent size according to each scenario extracted by risk assessment methods such as HAZOP is important.
First QSPR models to predict the thermal stability of potential self-reactive substances
2022, Process Safety and Environmental Protection