Predicting Thermal Behavior of Chemicals from Molecular Structure

Most chemical processes include potentially hazardous steps mainly due to the nature of the involved chemicals and the operating conditions. In order to properly assess and address the risks, a deep knowledge of the involved chemicals is required. Information can be gathered through the knowledge-to-date and completed with experimental investigations and lab tests.  This information helps setting the appropriate handling and storage conditions for safe operations.

However, accidents still occur nowadays, and according to the EU Major Accidents Hazards Bureau, in most cases, the underlying causes are flaws in the process and safety design. Thus, this crucial preliminary phase should be conducted with care and attention.

If reliable predictive methods were available, time and resources needed for the experimental investigations would be reduced as the tests could be better targeted. This would allow to better focus and analyze more in-depth the results. Moreover, simulations could also offer an opportunity to screen several alternatives in order to identify the less hazardous ones and to conduct experiments only with narrower selections, avoiding thus the exposure and the time consumption from repetitive tests on various chemicals.

The thermal stability of the involved chemicals is one of the key questions to be answered when assessing the process related hazards. It is commonly determined using Differential Scanning Calorimetry (DSC) experiments that allow verifying if the chemicals are stable with respect to heat or if they decompose above certain temperatures. The analysis of DSC thermograms should reveal the onset temperature and the reaction enthalpy of the eventual decomposition the compound could undergo.

We present here a study relying on Quantitative Structure-Property Relationships (QSPR) for predicting the thermal stability of chemicals. The QSPR method is based on empirical correlations between experimental data and numerical molecular descriptors of the examined compounds in order to build predictive models. Thus, in a first step, a large number of chemicals belonging to diverse families were investigated through DSC analysis. Their thermal behavior is defined as a set of properties extracted from DSC thermograms: the curves are assimilated to a succession of Gaussian-like peaks characterized by their position, width, height and asymmetry, the onset temperature, and the released reaction enthalpy.

In a second step, descriptors encoding several aspects of their molecular structure such as constitution, geometry or electronic properties were calculated. The models are then built as multiple linear regressions of the thermal properties as functions of the molecular descriptors. For each property a separate model is produced, offering estimates of that property for a given compound. All these features together give an overview of the considered compound’s thermal behavior without going through the testing phase.

Moreover, by transforming the separate properties back into a curve, an estimated DSC thermogram is reconstructed and predicted. The main advantage of this back-processing step is to recover a graphical representation of the thermal behavior which reveals the decomposition kinetics: generally, a broad peak indicates a slow decomposition following the onset whereas a narrow peak means a rather violent and sudden reactivity.

However, it is important to note that predictive models applied to thermal stability in particular and process safety in general are not intended as a replacement to the experimental investigation, but rather as a helpful complement: they would represent an exploratory tool to estimate these thermal characteristics at an early stage of risk evaluation, a screening tool when several alternatives are in consideration for a similar purpose, and a predictive one when the chemicals of interest are not available for testing.

Resources are key parameters in process safety. All stakeholders involved during the process design, the construction, set-up, commissioning, start-up and exploitation (and later the shut-down) of a process have to consider the inherently safer procedure and resilience engineering. All the developments that will deliver tools allowing decreasing the necessary resources to fulfill these goals are welcome. The prediction of thermal behavior of chemical is one of these tools helping the different stakeholders at the different steps of the process life to get vital information they need regarding thermal stability and where they should concentrate their experimental resources.