A5101506103- Thermochemical Biomass Conversion Processes
- Granular flow and fluidized beds
- Heat transfer and supercritical fluids
- Advanced Text Analysis Techniques
- Advanced Combustion Engine Technologies
- Advanced Malware Detection Techniques
- Topic Modeling
- Natural Language Processing Techniques
- Lattice Boltzmann Simulation Studies
- Energetic Materials and Combustion
- Cyclone Separators and Fluid Dynamics
- Network Security and Intrusion Detection
- Chemical Thermodynamics and Molecular Structure
- Mineral Processing and Grinding
- Machine Learning in Materials Science
- Web Data Mining and Analysis
- Thermal and Kinetic Analysis
- Computer Graphics and Visualization Techniques
- Catalysis and Oxidation Reactions
- Recommender Systems and Techniques
- Embedded Systems and FPGA Design
- Fluid Dynamics and Heat Transfer
- Lignin and Wood Chemistry
- Advanced Graph Neural Networks
- Petroleum Processing and Analysis
University of Chinese Academy of Sciences
2010-2024
Institute of Process Engineering
2012-2024
Chinese Academy of Sciences
2005-2024
State Key Laboratory of Chemical Engineering
2023
State Key Laboratory of Multiphase Complex Systems
2023
Harbin University of Science and Technology
2023
China Centre for Resources Satellite Data and Application
2021
Ministry of Natural Resources
2021
Institute of Information Engineering
2015-2020
National Security Technologies (United States)
2015
Joint extraction of entities and relations is an important task in natural language processing (NLP), which aims to capture all relational triplets from plain texts. This a big challenge due some the extracted one sentence may have overlapping entities. Most existing methods perform entity recognition followed by relation detection between every possible pairs, usually suffers numerous redundant operations. In this paper, we propose relation-specific attention network (RSAN) handle issue....
The methodology development and applications of ReaxFF molecular dynamics (ReaxFF MD) in unraveling the complex reactions kinetics for pyrolysis oxidation organic systems are reviewed. Particular attention is given to large-scale MD simulation method ∼10,000 atoms practical strategies overcome temporal–spatial limits as much possible. High-performance computing codes running on CPU cluster/supercomputers GPU were overviewed. GPU-enabled code like GMD-Reax revolutionizing simulations run...
Social recommendation, which aims to exploit social information improve the quality of a recommender system, has attracted an increasing amount attention in recent years. A large portion existing recommendation models are based on tractable assumption that users consider same factors make decisions both systems and networks. However, this is not concert with real-world situations, since usually show different preferences scenarios. In paper, we investigate how differences between user...
Deep understanding of the detailed coal pyrolysis process is very important for clean utilization. The overall stages in were investigated by ReaxFF MD simulations large-scale models combined with reaction analysis a cheminformatics approach. Analysis slow heat-up molecular dynamics (MD) shows that Liulin can be divided into four based on thermal cleavage bridge bonds: activation stage structure (Stage-I), primary (Stage-IIA), secondary (Stage-IIB), and recombination dominated (Stage-III)....
This work presents long time ReaxFF MD simulations of fuel-rich combustion for up to 10 ns explore the initial mechanism soot nanoparticle formation. A 24-component rocket propellant 1 (RP-1) model based on major components RP-1 fuel was employed. Simulations were performed by GPU-accelerated code GMD-Reax, and reactions therein revealed with aid VARxMD. Simulated evolution physical chemical properties largest molecule exhibits overall structural transitions three stages incipient ring...
Proper description of chemical structure diversity is necessary for a coal model in exploring the pyrolysis mechanism by reactive molecular dynamics (ReaxFF-MD) simulation. This paper presents strategy constructing large and reasonable models manually with varied structures. A multicomponent containing 23,898 atoms was constructed Fugu subbituminous following proposed on basis characterization data obtained from proximate ultimate analysis, 13C NMR, solvent extraction experiments. The...
The initial conversion pathways of linkages and their linked monomer units in lignin pyrolysis were investigated comprehensively by ReaxFF MD simulations facilitated the unique VARxMD for reaction analysis. simulated molecular model contains 15 920 atoms was constructed on basis Adler's softwood model. uncover ratio various aryl monomers. For rings α-O-4, β-O-4 α-O-4 & β-5, Cα/Cβ ether bond cracking dominates pathway accounting at least up to 80% consumption. linkage β–β γ-O-α, both Cα–O its...
Empirical correlations for bubble diameter and velocity are incapable of predicting the local behaviors fairly because impact hydrodynamics on bubbles in fluidized beds. Based image processing, a novel identification method with an adaptive threshold was proposed to distinguish characterize The information regarding properties can thus be extracted using big data from highly resolved simulations. Accordingly, deep neural network trained accurately predict properties, where inputs were...
Reactions of linkages and monomer rings in hardwood, softwood, kraft lignin pyrolysis were investigated using reactive force field (ReaxFF) molecular dynamics (MD) simulations. Four large models with diverse oxygen-containing groups constructed on the basis two-dimensional proposed by Adler Freudenberg for softwood lignin, Nimz hardwood Marton lignin. Large-scale ReaxFF MD simulations performed GMD-Reax code analyzed VARxMD code. The breaking Cα/Cβ–O bonds α/β–O–4 is dominant pathway to...
Document-level relation extraction (RE) poses new challenges over its sentence-level counterpart since it requires an adequate comprehension of the whole document and multi-hop reasoning ability across multiple sentences to reach final result. In this paper, we propose a novel graph-based model with Dual-tier Heterogeneous Graph (DHG) for document-level RE. particular, DHG is composed structure modeling layer followed by layer. The major advantage that capable not only capturing both...
Overall scenarios first revealed clearly by ReaxFF MD for self-heating, similar product distributions and altered kinetics in the three-stage decomposition of CL-20 its cocrystals under adiabatic conditions.
Understanding the underlying mechanisms on sensitivity-decrease of CL-20/TNT cocrystal is essential for wide applications promising high-energetic CL-20. This work presents chemical scenario thermolysis obtained from ReaxFF molecular dynamics simulations. Facilitated by unique VARxMD reaction analysis, interplay reactions between CL-20 and TNT responsible was first revealed. The early response to thermal stimulus dominated N-NO2 bond cleavage NO2 formation C-N scission leading ring-opening...
Extracting relations from plain text is an important task with wide application. Most existing methods formulate it as a supervised problem and utilize one-hot hard labels the sole target in training, neglecting rich semantic information among relations. In this paper, we aim to explore supervision soft relation extraction, which makes possible integrate prior knowledge. Specifically, bipartite graph first devised discover type constraints between entities based on entire corpus. Then,...
This work compares pyrolysis reactions of 3- and 4-component surrogate models RP-3 aviation fuel by a ReaxFF molecular dynamics (MD) simulation method. To evaluate the reactivity two models, multi-component baseline model that consists 45 components was constructed as representative real fuel. Reactive MD simulations were performed for simple using GPU code GMD-Reax. Reaction pathways analyzed with aid unique software VARxMD. The main product yield initial reaction in heat-up are found...
Exploring the physical mechanisms of complex systems and making effective use them are keys to dealing with complexity world. The emergence big data enhancement computing power, in conjunction improvement optimization algorithms, leading development artificial intelligence (AI) driven by deep learning. However, learning fails reveal underlying logic connotations problems being solved. Mesoscience provides a concept understand mechanism spatiotemporal multiscale structure systems, its...
An efficient and highly scalable bond-order potential code has been developed for the molecular dynamics simulation of bulk silicon, reaching 1.87 Pflops (floating point operations per second) in single precision on 7168 graphic processing units (GPUs) Tianhe-1A system. Furthermore, by coupling GPUs central units, we also simulated surface reconstruction crystalline silicon at sub-millimeter scale with more than 110 billion atoms, 1.17 plus 92.1 Tflops double entire Such simulations can...
The reactive molecular dynamics using ReaxFF provides an effective means to generate global reactions for pyrolysis of realistic fuel mixtures. from large-scale simulations a mixture may be characterized by multiple reaction sites, explosion intermediate species structures, and scattered contribution diversified pathways product species. This work proposes approach SRG-Reax aiming at generating skeleton networks based on patterns or classes centers huge obtained MD pyrolysis. (Skeleton...