Machine-Learning-Assisted Understanding of Polymer Nanocomposites Composition–Property Relationship: A Case Study of NanoMine Database

Abstract

The NanoMine database, one of two nodes in the MaterialsMine database, is a new materials data resource that collects annotated data on polymer nanocomposites (PNCs). This work showcases the potential of NanoMine and other materials data resources to assist fundamental materials understanding and therefore rational materials design. This specific case study is built around studying the relationship between the change in the glass transition temperature T g (ΔT g ) and key descriptors of the nanofillers and the polymer matrix in PNCs. We sifted through data from over 2000 experimental samples curated into NanoMine, trained a decision tree classifier to predict the sign of PNC ΔT g , and built a multiple power regression metamodel to predict ΔT g . The successful model used key descriptors including composition, nanoparticle volume fraction, and interfacial surface energy. The results demonstrate the power of using aggregated materials data to gain insight and predictive capability. Further analysis points to the importance of additional analysis of parameters from processing methodologies and continuously adding curated data sets to increase the sample pool size.

Moving beyond silver in point-of-use drinking water pathogen control

“…For example, one can specify a desired nanocomposite glass transition temperature, then use machine learning and artificial intelligence techniques to generate material candidates with varying degrees of synthetic accessibility. 63,64 Applying these same approaches to the disinfection context ( i.e. , enabling computational design for desired disinfection efficacy) would elucidate possibilities beyond the currently available materials, but will rely on mechanistic understanding of current material performance.…”

“…For example, one can specify a desired nanocomposite glass transition temperature, then use machine learning and artificial intelligence techniques to generate material candidates with varying degrees of synthetic accessibility. 63,64 Applying these same approaches to the disinfection context (i.e., enabling computational design for desired disinfection efficacy) would elucidate possibilities beyond the currently available materials, but will rely on mechanistic understanding of current material performance. While the outcome of these approaches can be hypothetical, yet-to-be synthesized materials, there exists a vast array of available synthetic Environmental Science: Water Research & Technology Perspective methods enabling precise atomic control for the assembly of materials.…”

Expanding the Horizons of Machine Learning in Nanomaterials to Chiral Nanostructures

“…[ 82 ] Classification can be also carried out using DT, RF, radial basis function SVM, and linear SVM classifiers, as recently demonstrated by Ma et al to divide polymer nanocomposites based on the glass transition temperature. [ 83 ] The RF and DT classifiers demonstrated classification accuracies of 0.922 and 0.852, respectively, on a dataset of 120 nanocomposite samples. NNs are also commonly used for classification tasks.…”

“…Those models include multipower regression, RF, K-nearest neighbor, support vector (SV), and principal component analysis (PCA) together with NN, etc. [ 83 , 87 – 89 ] As an example, Chen et al employed a RF model to quantitatively analyze the Cu level in carbon black particles. [ 90 ] The authors used laser-induced breakdown spectra which were used in modeling with and without pre-treatment with variable selection methods.…”

Leveraging engineered nanomaterials to support material circularity

“…Machine learning and artificial intelligence methods are already being applied to the design of materials such as metal organic frameworks 65 and polymer nanocomposites. 66 Such approaches highlight the immense value in reducing the vast potential material design space to compositions that meet predefined physicochemical property criteria (e.g., a minimum tensile strength or electrical conductivity). There remains an unrealized and tangible opportunity to establish and incorporate additional criteria that represent environmental benefits and impacts, material hazards, and circular material properties.…”