| POLYMERS Vol.69 No.6 |
| >> Japanese | >> English |
COVER STORY
Big Data and Databases |
| COVER STORY: Highlight Reviews |
| Polymer Informatics for Product Development | Hisaki IKEBATA |
| <Abstract> Recently, some case in material development assisted by Materials informatics (MI) has been reported, however the main target is currently inorganic materials or small organic molecules. There are at least two reason that MI is difficult to be applied for polymer development. One is that data structure of polymer experiment is a little more complicated than that of inorganic materials or small organic molecules. The other reason depends on that complexity of data structure. Polymer researchers should think about many aspects leading to their properties, so it seems difficult to make universal experimental design that has one shared data structure. In this review, the current and future use of polymer informatics are described in the view of machine learning technologies and preparaton of environment for collecting material data. Keywords: Materials Informatics / Polymer Informatics / Machine Learning / Statistical Science |
| Top of the Page▲ |
| Expectations and Challenges of Materials Big Data | Yibin XU |
| <Abstract> Materials informatics (MI) is expected to accelerate the process of materials design and development. More and more research results based on materials data and machine learning have been reported recently. Materials big data is the fundamental of MI. However, to construct materials big data, we face challenges of data capture, data curation, data exchange, etc. Some of these challenges are expected to be solved by the progress of information technology, however, some of them are issues of material science. In this paper, the history of materials data activities is reviewed. Based on our experiences of data collection, database development and materials design by machine learning, conditions of qualified data for MI are concluded. Small data strategies are introduced in order to apply machine learning methods with limited data and to make the best use of existing data. Keywords: Material Data / Database / Big Data / Data Curation / Data Quality / Machine Learning / Small Data |
| Top of the Page▲ |
| Collaborative Application of Simulation and Informatics | Isamu SHIGEMOTO |
| <Abstract> Polymers are difficult to handle in computer simulations because of their slow dynamics of structure formation and strong process dependence of physical properties. As a challenge to such “polymer-likeness”, we will review the theory-based simulation and machine learning-based informatics, as well as the progress and possibility of predicting physical properties through the cooperation of both, with reference to recent papers. Keywords: Molecular Simulations / Coarse-Grained Simulations / Materials Informatics / Machine Learning / Property Prediction |
| Top of the Page▲ |
| COVER STORY: Topics and Products |
| Designing Duarability of Polymeric Materials based on High-Throughput Experiments | Toshiaki TANIIKE |
| <Abstract> The scarcity of learnable datasets is a bottleneck of materials informatics. Here, our recent efforts are outlined on the utilization of high-throughput experiments in studying the durability of polymeric materials. A high-throughput chemiluminescence imaging instrument we newly developed realizes simultaneous determination of the lifetime of 100 samples during thermoxidative degradation. Combining this instrument with a genetic algorithm, additive formulations for polypropylene are explored in a non-empirical fashion. Such a strategy indeed works for discovering novel formulations as well as for studying synergestic effects among additives. Combinations among hindered phenol anti-oxidants bearing different functional groups and molecular weights are suggested to be highly effective for suppressing the thermoxidative degradation of polypropylene at an elevated temperature. Durability data of years are attained in a few months with the aid of high-throughput experimentation. Keywords: High-Throughput Experiment / Polymer Durability / Chemiluminescence Imaging / Additive Formulation |
| Top of the Page▲ |
| Design of Polymer Solar Cells Based on Machine Learning | Akinori SAEKI |
| <Abstract> Materials informatics is a data-driven approach that shows an extremely high throughput efficiency in material exploration of a large molecular space. Thus, a machine learning (ML) technique may be great tool for the rapid screening of next-generation solar cells such as organic photovoltaics (OPV) and perovskite solar cell (PSC), which comprises a complex causality between the fundamental properties (chemical structure, electronic properties, etc) and the device performance (power conversion efficiency, PCE). This article reviews two ML-based research topics: (1) the extraction of an explanatory variable that governs the hole transfer process and PCE of PSC by data science analysis, and (2) the virtual screening of conjugated polymers for OPV by supervised ML. The former exhibits the success in extracting the primary factor, and the latter shows a good agreement between predicted and experimental PCEs. Accordingly, ML supported by manual consideration is expected to be a useful tool for researches in polymer science. Keywords: Materials Informatics / Machine Learning / Organic Photovoltaics / Perovskite Solar Cell / Conjugated Polymer / Random Forest / Artificial Neural Network / Least Absolute Shrinkage and Selection Operator (LASSO) |
| Top of the Page▲ |
| Exploration of Li+/Electron Conducting Polymers by Machine Learning | Kan HATAKEYAMA-SATO |
| <Abstract> Machine learning (ML) is becoming an important approach even for experimental chemists. Here, Li+/electron-conducting polymers were explored by ML techniques. Firstly, an experimental database of Li+-conductors were collected manually (104 cases). Next, all inputted data were converted to numeric arrays by originally written Python scripts, because the ML models can receive only numeric inputs. The relationships between the conductors’ information (e.g., chemical structures) and their ionic conductivity were learned by a probablistic ML model. New conductors were found through the screening from a number of candidate structures. Process informatics were also conducted for PEDOT-PSS, whose conductivity changed dramatically by the different post-treatments. Introduction of graph structured databases and neural network was effective to treat such complicated process information. Keywords: Materials Informatics / Li+-Conducting Polymers / PEDOT-PSS / Deep Learning |
| Top of the Page▲ |
| Experiment-Oriented Materials Informatics for Small Data | Yuya OAKI, Yasuhiko IGARASHI |
| <Abstract> Application of data-scientific methods has attracted much interest in cutting-edge materials science. However, in general, a variety of data-scientific approaches are not easily applied to small data generated by experimental scientists. Our collaborative research has focused on materials informatics used with small-scale experimental data. In the present article, we introduced two examples of experiment-oriented materials informatics for small data. The first example is high-yield syntheses of nanosheets from soft layered organic-inorganic composites through the exfoliation in organic media. The yield prediction model was constructed by sparse modeling, a data-scientific method, combined with chemical perspective on the small-scale experimental data. The high-yield syntheses of a variety of nanosheets were achieved in a minimum number of experiments. The second example is the exploration of a high-performance organic-polymer anode for lithium-ion batteries. The capacity prediction model realized the discovery of a new organic anode with high specific capacity and cyclability. The experiment-oriented materials informatics can be applied to design a variety of advanced functional materials. Keywords: Materials Informatics / Sparse Modeling / Soft Layered Composites / Exfoliation / Nanosheets / Lithium-Ion Battery / Organic Anodes / Conjugated Polymers |
| Top of the Page▲ |
| Development of Automatic Knowledge Extraction System Specilized for MI | Shuichi HIROSE, Hiroki TODA, Yasumitsu ORII |
| <Abstract> Data exploration has occurred in the field of material design, therefore the importance of Materials Informatics (MI) become growing. One of the most pressing challenges in MI is to transform a large volume of documents into knowledge. The process of MI is composed of three steps, (1) conversion of unstructured data such as scientific papers into structured data, (2) natural language processing specialized for materials, and (3) knowledge representation using knowledge graph (KG). This review introduces a concrete method to solve the issue based on polymer data. For example, 400,000 entities were extracted from more than 700 patents, and their relationship is represented in KG. NAGASE & CO., LTD. has been developing MI systems that integrate three components with IBM. It will be provided to customers as SaaS (Software as a Service). The initial version provides an analysis of enzyme and formulation such as formulated epoxies, and the target fields in MI system will be expanded in the future. Keywords: Materials Informatics / Unstructured Data / Structured Data / Natural Language Processing / Knowledge Graph / Polymer / SaaS / Data Exploration |
| Top of the Page▲ |
| Polymer Science and I: A Personal Account |
| Aiming at Original and Straightforward Work | Hiromitsu SOGAWA |
| <Abstract> I want to develop original and straightforward work. As I feel I am just an average talented and decently intelligent person, many years of assiduous effort would be very important. I will keep on trying my best. |
| Top of the Page▲ |
| Front-Line Polymer Science |
| Biomass Conversion Process Innovated by Ionic Liquid | Daisuke HIROSE, Naoki WADA, Kenji TAKAHASHI |
| <Abstract> Environmental problems relating to the carbon dioxide emission caused by incinerating depletable petroleum resources have worsened. Switching from petroleum to renewable resources is required. Lignocellulosic biomass consisting of three major components, such as cellulose, hemicellulose, and lignin, is well known as a renewable resource that can be produced from sunlight, carbon dioxide, and water via photosynthesis. Production of lignocellulose-based materials with low environmental impact requires to change their production processes including biomass conversion as a more efficient one as possible. This mini-review describes recent progress in novel lignocellulose conversion processes by using ionic liquids. Multifunctional roles of ionic liquids such as solvent and catalysts made it possible to continuously simplify processes and developing new modification methods. This unique character of ionic liquids was indispensable for solving the problems of conventional processes. It will strongly contribute to developing new green materials and procedures soon. Keywords: Biomass / Ionic Liquid / Cellulose / Green Chemistry / Biorefinery / Carbohydrate Polymer / Catalyst / Solvent |
| Top of the Page▲ |
| Copyright(C) 2020 The Society of Polymer Science, Japan All Rights Reserved. |