Data-driven approaches such as deep learning can result in predictive models for material properties with exceptional accuracy and efficiency. However, in many applications, data is sparse, severely limiting their accuracy and applicability. To improve predictions, techniques such as transfer learning and multitask learning have been used. The performance of multitask learning models depends on the strength of the underlying correlations between tasks and the completeness of the data set. Standard multitask models tend to underperform when trained on sparse data sets with weakly correlated properties. To address this gap, we fuse deep-learned embeddings generated by independent pretrained single-task models, resulting in a multitask model that inherits rich, property-specific representations. By reusing (rather than retraining) these embeddings, the resulting fused model outperforms standard multitask models and can be extended with fewer trainable parameters. We demonstrate this technique on a widely used benchmark data set of quantum chemistry data for small molecules as well as a newly compiled sparse data set of experimental data collected from literature and our own quantum chemistry and thermochemical calculations.

Predictive models for the performance of explosives and propellants are important for their design, optimization, and safety. Thermochemical codes can predict some of these properties from fundamental quantities such as density and formation energies that can be obtained from first principles. Models that are simpler to evaluate are desirable for efficient, rapid screening of material screening. In addition, interpretable models can provide insight into the physics and chemistry of these materials that could be useful to direct new synthesis. Current state-of-the-art performance models are based on either the parametrization of physics-based expressions or data-driven approaches with minimal interpretability. We use parsimonious neural networks (PNNs) to discover interpretable models for the specific impulse of propellants and detonation velocity and pressure for explosives using data collected from the open literature. A combination of evolutionary optimization with custom neural networks explores and trains models with objective functions that balance accuracy and complexity. For all three properties of interest, we find interpretable models that are Pareto optimal in the accuracy and simplicity space.

Strain energy is a fundamental measure of the steric and configurational properties of organic molecules. The ability to estimate strain energy through quantum chemical simulations requires at minimum the knowledge of an initial set of nuclear coordinates. In general, such knowledge is not categorically known when screening or generating large numbers of molecule candidates in the context of molecular design. We present a machine learning approach to predict hydrocarbon strain energies using Benson group equivalents. A featurization strategy is crafted by concatenating the molecule group equivalent counts with easily computable molecular fingerprints. The data are obtained from electronic structure calculations we performed on a set of 166 previously synthesized strained hydrocarbons. These data are provided and include gas phase enthalpies of formation and associated optimized atomic coordinates. The strain energy prediction accuracy of several statistical learning methods is evaluated, and their respective merits and limitations are discussed.

Data science and artificial intelligence are playing an increasingly important role in the physical sciences. Unfortunately, in the field of energetic materials data scarcity limits the accuracy and even applicability of ML tools. To address data limitations, we compiled multi-modal data: both experimental and computational results for several properties. We find that multi-task neural networks can learn from multi-modal data and outperform single-task models trained for specific properties. As expected, the improvement is more significant for data-scarce properties. These models are trained using descriptors built from simple molecular information and can be readily applied for large-scale materials screening to explore multiple properties simultaneously. This approach is widely applicable to fields outside energetic materials.

Recent advances have enabled machine learning methodologies developed for large datasets to be applied to the small experimental datasets typically available for chemical systems. Such advances typically involve a data-based approach to transfer learning, where a portion of the experimental data for the property of interest is used to fine-tune a model that is pre-trained on computationally generated data. This transfer learning approach does not work for very small experimental datasets, where there are only enough data for model validation. Here, we develop a physics-informed transfer learning strategy to train a directed-message passing neural network (D-MPNN) model, enabling extrapolation outside of the training domain. We demonstrate this approach by training a D-MPNN model on interpolated vapor pressures and validate the model on an out-of-sample test set of energetic molecule vapor pressures, achieving accuracy comparable to those of experiments.

Data-driven machine learning techniques can be useful for the rapid evaluation of material properties in extreme environments, particularly in cases where direct access to the materials is not possible. Such problems occur in high-throughput material screening and material design approaches where many candidates may not be amenable to direct experimental examination. In this paper, we perform an exhaustive examination of the applicability of machine learning for the estimation of isothermal shock compression properties, specifically the shock Hugoniot, for diverse material systems. A comprehensive analysis is conducted where effects of scarce data, variances in source data, feature choices, and model choices are systematically explored. New modeling strategies are introduced based on feature engineering, including a feature augmentation approach, to mitigate the effects of scarce data. The findings show significant promise of machine learning techniques for design and discovery of materials suited for shock compression applications.

A heterogeneous multiscale method (HMM) to evolve chemistry across disparate temporal scales is presented. The method directly couples a particle-based, microscale coarse-grain chemistry model to a macroscale continuum finite element model of deformation. The macroscale model obtains both the material equation-of-state and instantaneous chemical reaction rates through evaluation of the microscale model subject to distinct boundary conditions. The approach is demonstrated through several simulations of the energetic material 1,3,5-trinitrohexahydro-s-triazine (RDX) subject to slow heating, including a simulation inspired by the scaled thermal explosion (STEX) experiment.

For many experimentally measured chemical properties that cannot be directly computed from first-principles, the existing physics-based models do not extrapolate well to out-of-sample molecules, and experimental datasets themselves are too small for traditional machine learning (ML) approaches. To overcome these limitations, we apply a transfer learning approach, whereby we simultaneously train a multi-target regression model on a small number of molecules with experimentally measured values and a large number of molecules with related computed properties. We demonstrate this methodology on predicting the experimentally measured impact sensitivity of energetic crystals, finding that both characteristics of the computed dataset and model architecture are important to prediction accuracy of the small experimental dataset. Our directed-message passing neural network (D-MPNN) ML model using transfer learning outperforms direct-ML and physics-based models on a diverse test set, and the new methods described here are widely applicable to modeling many other structure-property relationships.

Deep learning has shown great potential for generating molecules with desired properties. But the cost and time required to obtain relevant property data have limited study to only a few classes of materials for which extensive data have already been collected. We develop a deep learning method that combines a generative model with a property prediction model to fuse small data of one class of molecules with larger data in another class. Common low-level physicochemical properties are jointly embedded into a latent space that can be used to design molecules in the smaller class. The chemical space around the molecules in the training set is explored through local gradient ascent optimization. Based on nine molecules from the original training set, nine new molecules are found to have improved properties while remaining structurally similar to the training molecules thereby easing requirements for entirely new synthesis routes. Validation is performed using an equilibrium thermochemistry code to verify the molecules and target properties. A specific example targeting the Chapman-Jouguet velocity and small data for nitrogen-rich molecules is shown. Despite the relative lack of nitrogen-rich molecule data, the results demonstrate that fusing and joint embedding with plentiful low nitrogen molecular data can produce higher generative performance than using the scarce data alone.

We present new capabilities for investigation of microstructure in energetic material response for both explicit large-scale and multiscale simulations. We demonstrate the computational capabilities by studying the effect of porosity on the reactive shock response of a coarse-grain (CG) model of the energetic material cyclotrimethylene trinitramine (RDX), the non-reactive equation of state for a porous representative volume element (RVE) of CG RDX, and utilization of available supercomputing resources for speculative sampling to accelerate hierarchical multiscale simulations. Small amounts of porosity (up to 4 %) are shown to have significant effect on the initiation of reactive CG RDX using large-scale reactive dissipative particle dynamics simulations. Non-reactive RVEs are shown to undergo a porosity-dependent pore collapse at hydrostatic conditions, and an existing automation framework is shown to be easily modified for the incorporation of microstructure while retaining reliable convergence properties. A novel predictive sampling method based on use of kernel density estimators is shown to effectively accelerate time-to-solution in a multiscale simulation, scaling with free CPU cores, while making no assumptions about the underlying physics for the data being analyzed. These multidisciplinary studies of distinct yet connected problems combine to provide methodological insights for high-fidelity modeling of reactive systems with microstructure.