Solvent selection for polymers enabled by generalized chemical fingerprinting and machine learning.

We present machine learning models trained on experimental data to predict room-temperature solubility for any polymer-solvent pair. The new models are a significant advancement over past data-driven work, in terms of protocol, validity, and versatility. A generalizable fingerprinting method is used for the polymers and solvents, making it possible, in principle, to handle any polymer-solvent combination. Our data-driven approach achieves high accuracy when either both the polymer and solvent or just the polymer has been seen during the training phase. Model performance is modest though when a solvent (in a newly queried polymer-solvent pair) is not part of the training set. This is likely because the number of unique solvents in our data set is small (much smaller than the number of polymers). Nevertheless, as the data set increases in size, especially as the solvent set becomes more diverse, the overall predictive performance is expected to improve.

Predicting Crystallization Tendency of Polymers Using Multifidelity Information Fusion and Machine Learning.

The degree of crystallinity of a polymer is a critical parameter that controls a variety of polymer properties. A high degree of crystallinity is associated with excellent mechanical properties crucial for high-performing applications like composites. Low crystallinity promotes ion and gas mobility critical for battery and membrane applications. Experimental determination of the crystallinity for new polymers is time and cost intensive. A data-driven machine learning-based method capable of rapidly predicting the crystallinity could counter these disadvantages and be used to screen polymers for a myriad of applications in a fast, inexpensive fashion. In this work, we developed the first-of-its-kind, data-driven machine learning model to predict the most-likely polymer crystallinity trained on experimental data and theoretical group contribution methods. Since polymer data under consistent processing conditions are unavailable, we tackled process variability by using the "most-likely" polymer values which we refer to as the polymer's tendency to crystallize. Experimental data for polymers' tendency to crystallize is limited by number and diversity, and to tackle this, we augmented experimentation-based data with data using group contribution methods. Therefore, this work utilized two data sets, viz., a high-fidelity, experimental data set for 107 polymers and a more diverse, less accurate low-fidelity data set for 429 polymers which used group contribution methods. We used a multifidelity information fusion strategy to utilize all the information captured in the low-fidelity data set while still predicting at the high-fidelity accuracy. Although this model inherently assumed "typical" processing conditions and estimated the "most-likely" percent crystallinity value, it can help in the estimation of a polymer's tendency to crystallize in a far more cost-effective and efficient manner.

Critical Assessment of the Hildebrand and Hansen Solubility Parameters for Polymers.

Solubility parameter models are widely used to select suitable solvents/nonsolvents for polymers in a variety of processing and engineering applications. In this study, we focus on two well-established models, namely, the Hildebrand and Hansen solubility parameter models. Both models are built on the basis of the notion of "like dissolves like" and identify a liquid as a good solvent for a polymer if the solubility parameters of the liquid and the polymer are close to each other. Here we make a critical and quantitative assessment of the accuracy/utility of these two models by comparing their predictions against actual experimental data. Using a data set of 75 polymers, we find that the Hildebrand model displays a predictive accuracy of 60% for solvents and 76% for nonsolvents. The Hansen model leads to a similar performance; on the basis of a data set of 25 polymers for which Hansen parameters are available, we find that it has an accuracy of 67% for solvents and 76% for nonsolvents. The availability of the Hildebrand parameters for a large polymer data set makes it a widely applicable capability, as the Hildebrand parameter for a new polymer may be determined using this data set and machine learning methods as we have done before; the predicted Hildebrand parameter for a new polymer may then be used to determine suitable solvents and nonsolvents. Such predictions are difficult to make with the Hansen model, as the data set of Hansen parameters for polymers is rather small. Nevertheless, the Hildebrand approach must be used with caution. Our analysis shows that while the Hildebrand model has a predictive accuracy of 70-75% for nonpolar polymers, it performs rather poorly for polar polymers (with an accuracy of 57%). Going forward, determination of solvents and nonsolvents for polymers may benefit by developing classification models built directly on the basis of available experimental data sets rather than utilizing the solubility parameter approach, which is limited in versatility and accuracy.