A sparse modeling for small data: Case studies in controlled syntheses of 2D materials

AbstractThis paper proposes an extended version of the interventionist account for causal inference in the practical context of biological mechanism research. This paper studies the details of biological mechanism researchers’ practices of assessing the evidential legitimacy of experimental data, arguing why quantity and variety are two important criteria for this assessment. Because of the nature of biological mechanism research, the epistemic values of these two criteria result from the independence both between the causation of data generation and the causation in question and between different interventions, not techniques. The former independence ensures that the interventions in the causation in question are not affected by the causation that is responsible for data generation. The latter independence ensures the reliability of the final mechanisms not only in the empirical but also the formal aspects. This paper first explores how the researchers use quantity to check the effectiveness of interventions, where they at the same time determine the validity of the difference-making revealed by the results of interventions. Then, this paper draws a distinction between experimental interventions and experimental techniques, so that the reliability of mechanisms, as supported by the variety of evidence, can be safely ensured in the probabilistic sense. The latter process is where the researchers establish evidence of the mechanisms connecting the events of interest. By using case studies, this paper proposes to use ‘intervention’ as the fruitful connecting point of literature between evidence and mechanisms.

Download Full-text

Drained instability of sand in plane strain

Canadian Geotechnical Journal ◽

10.1139/t09-111 ◽

2010 ◽

Vol 47 (4) ◽

pp. 400-412 ◽

Cited By ~ 15

Author(s):

Dariusz Wanatowski ◽

Jian Chu ◽

Wai Lay Loke

Keyword(s):

Experimental Data ◽

Plane Strain ◽

Case Studies ◽

Static Loading ◽

State Parameter ◽

Dam Failure ◽

Triaxial Tests ◽

Granular Soil ◽

Laboratory Studies ◽

Dense Sand

Flowslide or failure of loose granular soil slopes is often explained using liquefaction or instability data obtained from undrained triaxial tests. However, under static loading conditions, the assumption of an undrained condition is not realistic for sand, particularly clean sand. Case studies have indicated that instability of granular soil can occur under essentially drained conditions (e.g., the Wachusett Dam failure in 1907). Laboratory studies on Changi sand by Chu et al. in 2003 have shown that sand can become unstable under completely drained conditions. However, these studies were carried out under axisymmetric conditions and thus, cannot be applied directly to the analysis of slope failures. In this paper, experimental data obtained from plane-strain tests are presented to study the instability behaviour of loose and dense sand under plane-strain conditions. Based on these test data, the conditions for the occurrence of drained instability in plane strain are established. Using the modified state parameter, the conditions for instability under both axisymmetric and plane-strain conditions can be unified. A framework for interpreting the instability conditions of sandy slopes developed under axisymmetric conditions also extends into plane-strain conditions.

Download Full-text

CO2 Reduction to CH4 on Cu-doped Phosphorene: A First-Principles Study

Nanoscale ◽

10.1039/d1nr06066j ◽

2021 ◽

Author(s):

Hongping Zhang ◽

Run Zhang ◽

Chenghua Sun ◽

Yan Jiao ◽

Yaping Zhang

Keyword(s):

Carbon Dioxide ◽

Metal Atom ◽

First Principles ◽

Materials Science ◽

2D Materials ◽

Co2 Reduction ◽

Carbon Dioxide Reduction ◽

First Principles Study

Electrochemical carbon dioxide reduction (CRR) to fuels is one of the significant challenges in materials science and chemistry. Recently, single metal atom catalysts based on 2D materials provide a promising...

Download Full-text

Data-Driven Modeling of Mechanical Properties of Cast Iron Using Fuzzy Logic

Fuzzy Systems and Data Mining VI - Frontiers in Artificial Intelligence and Applications ◽

10.3233/faia200743 ◽

2020 ◽

Author(s):

He Tan ◽

Vladimir Tarasov ◽

Vasileios Fourlakidis ◽

Attila Dioszegi

Keyword(s):

Experimental Data ◽

Fuzzy Logic ◽

Cast Iron ◽

Materials Science ◽

Fatigue Behavior ◽

Relevant Literature ◽

Data Driven ◽

Fuzzy Logic Systems ◽

Data Driven Approach ◽

Data Driven Modeling

For many industries, an understanding of the fatigue behavior of cast iron is important but this topic is still under extensive research in materials science. This paper offers fuzzy logic as a data-driven approach to address the challenge of predicting casting performance. However, data scarcity is an issue when applying a data-driven approach in this field; the presented study tackled this problem. Four fuzzy logic systems were constructed and compared in the study, two based solely upon experimental data and the others combining the same experimental data with data drawn from relevant literature. The study showed that the latter demonstrated a higher accuracy for the prediction of the ultimate tensile strength for cast iron.

Download Full-text

Cross-property deep transfer learning framework for enhanced predictive analytics on small materials data

Nature Communications ◽

10.1038/s41467-021-26921-5 ◽

2021 ◽

Vol 12 (1) ◽

Author(s):

Vishu Gupta ◽

Kamal Choudhary ◽

Francesca Tavazza ◽

Carelyn Campbell ◽

Wei-keng Liao ◽

...

Keyword(s):

Transfer Learning ◽

Predictive Models ◽

Materials Science ◽

Predictive Analytics ◽

Large Datasets ◽

Small Data ◽

Widespread Application ◽

Learning Framework ◽

Large Databases ◽

Physical Attributes

AbstractArtificial intelligence (AI) and machine learning (ML) have been increasingly used in materials science to build predictive models and accelerate discovery. For selected properties, availability of large databases has also facilitated application of deep learning (DL) and transfer learning (TL). However, unavailability of large datasets for a majority of properties prohibits widespread application of DL/TL. We present a cross-property deep-transfer-learning framework that leverages models trained on large datasets to build models on small datasets of different properties. We test the proposed framework on 39 computational and two experimental datasets and find that the TL models with only elemental fractions as input outperform ML/DL models trained from scratch even when they are allowed to use physical attributes as input, for 27/39 (≈ 69%) computational and both the experimental datasets. We believe that the proposed framework can be widely useful to tackle the small data challenge in applying AI/ML in materials science.

Download Full-text

JAVA™-Based Design Calculator for Integral Snap-Fits

Volume 2: 19th Computers and Information in Engineering Conference ◽

10.1115/detc99/cie-9075 ◽

1999 ◽

Author(s):

Jung S. Oh ◽

Dean Q. Lewis ◽

Daeyong Lee ◽

Gary A. Gabriele

Keyword(s):

Experimental Data ◽

Operating System ◽

Case Studies ◽

Design Process ◽

Design Tool ◽

The Internet ◽

Specific Loading ◽

Different Types ◽

Integral Attachment ◽

A Company

Abstract Many different types of snap-fits have been developed to replace conventional fasteners, and research efforts have been made to characterize their performance. It is often tedious to look for design equations for unique types of snap-fits to calculate the insertion and retention forces. If found, these equations tend to be long, complex, and difficult to use. For this reason, a snap-fit calculator has been created to help in designing integral attachment features. Studies of seven most commonly used snap-fits (annular snap, bayonet-and-finger, cantilever hook, cantilever-hole, compressive hook, L-shaped hook, and U-shaped, hook) were used to provide the equations implemented in this snap-fit calculator, more fasteners than any other snap-fit calculator available. This tool aids in designing snap-fits to meet specific loading requirements by allowing the designer to size the feature to obtain desired estimates for maximum insertion and retention forces. The software for this design tool was written in JAVA™ language that is independent of operating system platforms and can be distributed at a company site-wide over an intranet or worldwide over the Internet. This makes it easily accessible to a user, and universal upgrades can be achieved by simply updating the software at the server location. Designers will find this tool to be useful in the design process and the most convenient way to estimate the performance of snap-fits. This paper describes the development and operation of the IFP snap-fit calculator including several case studies comparing the calculated results to experimental data.

Download Full-text