Machine learning application in the life time of materials

                  Xiaojiao Yu 

Abstract:
Materials design and development typically takes several decades from the initial discovery to commercialization with the traditional trial and error development approach. With the accumulation of data from both experimental and computational results, data based machine learning becomes an emerging field in materials discovery, design and property prediction. This manuscript reviews the history of materials science as a disciplinary the most common machine learning method used in materials science, and specifically how they are used in materials discovery, design, synthesis and even failure detection and analysis after materials are deployed in real application. Finally, the limitations of machine learning for application in materials science and challenges in this emerging field is discussed.
Keywords: Machine learning, Materials discovery and design, Materials synthesis, Failure detection

1. Introduction
   Materials science has a long history that can date back to the Bronze age 1. However, only until the 16th century, first book on metallurgy was published, marking the beginning of systematic studies in materials science 2. Researches in materials science were purely empirical until theoretical models were developed. With the advent of computers in the last century, numerical methods to solve theoretical models became available, ranging from DFT (density functional theory) based quantum mechanical modeling of electronic structure for optoelectronic properties calculation, to continuum based finite element modeling for mechanical properties 3-4. Multiscale modeling that bridge various time and spatial scales were also developed in the materials science to better simulate the real complex system 5. Even so, it takes several decades from materials discovery to development and commercialization 6-7. Even though physical modeling can reduce the amount of time by guiding experiment work. The limitation is also obvious. DFT are only used for functional materials optoelectronic property calculation, and that is only limited to materials without defect 8. The assumption itself is far off from reality. New concept such as multiscale modeling is still far away from large scale real industrial application. Traditional ways of materials development are impeding the progress in this field and relevant technological industry.
   With the large amount of complex data generated by experiment, especially simulation results from both published and archived data including materials property value, processing conditions, and microstructural images, analyzing them all becoming increasingly challenging for researchers. Inspired by the human genome initiative, Obama Government launched a Materials Genome Initiative hoping to reduce current materials development time to half 9. With the increase of computing power and the development of machine learning algorithms, materials informatics has increasingly become another paradigm in the field.
   Researchers are already using machine learning method for materials property prediction and discovery. Machine learning forward model are used for materials property prediction after trained on data from experiments and physical simulations. Bhadeshia et al. applied neural network(NN) technique to model creep property and phase structure in steel 10-11. Crystal structure prediction is another area of study for machine learning thanks to the large amount of structural data in crystallographic database. K-nearest- neighbor’s method was used to identify materials’ structure type based on its neighbors’ structure types 12-13. Machine learning is also applied for materials discovery by searching compositional, structural space for desired properties, which is essentially solving a constrained optimization problem. Baerns et al. was able to find an effective multicomponent catalyst for low-temperature oxidation of low- concentration propane with a genetic algorithm and neural network 14. There are a few reviews on machine learning application in materials science already. Dane Morgan and Gerbrand Ceder reviewed the data mining methods in materials development 15. Tim Mueller, Aaron Gilad Kusne, and Rampi Ramprasad also reviewed the progress and application of machine learning in materials science, more specifically in phase diagram, crystal structural and property prediction 16.
   However, their reviews are mostly based on applications in fundamental of materials science. Here, we are taking a more practical approach of reviewing machine learning application in material design, development and stages after deployment. We first discuss data problems specifically in materials science. Then, machine learning concept and most widely used methods are introduced. Up-to-date reviews on machine leaning application in materials discovery, design, de

…(Full text truncated)…