Industrial chemical processes can potentially involve thermal risks, as most of the reactions performed are exothermic, the chemicals used are often thermally unstable, and the operating conditions are set to induce high conversion and throughput. Performing an efficient risk assessment and implementing the proper risk mitigation measures are essential to avoid or at least reduce accidents and their potentially disastrous consequences.
An inherently safe process design needs a rigorous thermal risk assessment. Differential Scanning Calorimetry (DSC) is one of the most used analytical methods employed to determine the thermal stability of compounds and mixtures. It allows determining the thermal decompositions characteristics such as its potential energy release, triggering temperature, and kinetic data. DSC experiments can be considered resource efficient as an experiment can be conducted within hours using samples of few milligrams. However it becomes simply impossible to perform a measurement when an intermediate cannot be isolated, is extremely toxic or is physically unavailable. Moreover, when several tests have to be performed, the time and required resource accumulate. Considering how important this information is, it should not be overlooked and this, even in the absence of practically feasible experiments.
Predictions would be the appropriate response to such scenario. This work aims at predicting thermograms of pure compounds, resulting from DSC experiments, using their molecular structures as input. The procedure follows three phases:
Firstly, DSC curves are analyzed and decomposed into five key parameters (onset temperature, peak position, amplitude or maximum heat release rate, width and asymmetry). They are studied separately in order to preserve and later recover the full shape of the DSC curve by minimizing the data loss due to data abstraction. Experimental uncertainty is analyzed and sets the qualitative objectives of the simulations.
Secondly, predictive models are developed for a large number of structurally diverse and thermally reactive chemicals. There are mainly two structure-based approaches of data prediction: Group Contribution (GC) models and Quantitative Structure-Property Relationships (QSPR). Herein, both methods are applied in parallel and resulting models are evaluated relatively to the experimental data and compared to each other. Among the various group contributions frameworks available, the Marrero-Gani system was applied to identify simple and complex groups observed within the study set and determine their contributions. As for the QSPR system, the descriptors were generated to encode as many structural aspects as possible with constitutional, topological, electronic and quantum derived descriptors. For both methods, mathematical models built are linear regressions for simplicity purposes.
In the last phase, two levels of modeling are distinguished: firstly the chemicals are studied without structural classification to develop ‘global models’; then, several subsets are created based on structural similarities in order to develop ‘local models’. These two levels of modeling present in one hand a broader range of application and in the other hand a higher predictive accuracy.
This procedure delivers the entire DSC thermal trail whereas usual abstractions only retain onset temperature and enthalpy. The curve’s shape encloses complementary information regarding the thermal behavior of the tested compound as the decomposition kinetics for instance and therefore should not be discarded. In other respects, it is also an opportunity to compare two structural based approaches of model development (QSPR, GC) as they are usually studied separately on different and not comparable datasets.
Simulating the DSC thermograms from the molecular structure offers several advantageous applications in process design. Besides the previously mentioned possibility to predict DSC curves estimations which cannot be experimentally measured, there is also a time benefit. Prediction can be made at a very early stage of the process design. Even when the compound itself might not be available for preliminary testing its thermal behavior could already be anticipated. Moreover, simulations can allow analyzing several alternatives within limited resources, saving them for considering potential substitution of a hazardous compound by a less hazardous one or modification of the process. The substituted product or the initial product of interest -if no suitable substitution is identified- could be further investigated through a tailored and focused testing phase thanks to the predicted estimations. It is also noteworthy that predictive models help avoiding expendable handling of harmful chemicals.
We propose a method relying on molecular-based approaches to predict the thermal stability and how it could be used to identify thermal threats without necessarily facing them. It is important to stress that predictive models should be handled with precaution when applied to sensitive data such as safety related information. They are also limited to pure compounds, whereas thermal stability is affected by its matrix: other compounds in a mixture or solvents. Thus they are not intended to replace proper experimental investigations, but rather be a helpful tool that allows focusing the experimental work on the most critical compounds. The major benefits of such procedures within process design context are mainly to broaden the number of evaluations within given time and resources, an efficiency gain in testing phase with better resource allocation and valuable timing leading to anticipation.
See more of this Group/Topical: Process Development Division