arXiv Science⌕ Search

arXiv subjects

Shuvodeep De

Publications and source records attributed to Shuvodeep De.

3 recordsLinked to original sources

SiC-BN composites with extreme thermal and environmental stability

Ceramic materials that maintain structural integrity and resist thermal-shock under extreme conditions are critical for many applications. Here we show that spark plasma sintered composites containing equal weight fractions of hexagonal silicon carbide (6H-SiC) and boron nitride (h-BN) exhibit exceptional stability under extreme high-temperature oxidizing conditions and unusually stable temperature-independent thermal conductivity across a broad temperature range. Under high-velocity oxygen-fuel torch exposure in air, it withstands prolonged heating and thermal-cycling at surface temperatures reaching >1500 °C, whereas pure 6H-SiC and h-BN rapidly fractures or erodes. Microstructural evaluation shows the formation of a multi-phase surface silica layer that further protects the material from oxidation. Finite-element analysis heat-flux map shows that intrinsic porosity mitigates thermal stress, while ReaxFF molecular dynamics simulations reveal thermally driven interfacial reconstruction and the formation of a Si-N/C-B-rich bonding network, enhancing the interfacial cohesion at elevated temperatures. This lightweight, easily machinable ceramic composite will be a promising structural material in extreme environment applications.

cond-mat.mtrl-sci↗

Sentiment Analysis on Movie Reviews: A Deep Dive into Modern Techniques and Open Challenges

This paper presents a comprehensive survey of sentiment analysis methods for movie reviews, a benchmark task that has played a central role in advancing natural language processing. We review the evolution of techniques from early lexicon-based and classical machine learning approaches to modern deep learning architectures and large language models, covering widely used datasets such as IMDb, Rotten Tomatoes, and SST-2, and models ranging from Naive Bayes and support vector machines to LSTM networks, BERT, and attention-based transformers. Beyond summarizing prior work, this survey differentiates itself by offering a comparative, challenge-driven analysis of how these modeling paradigms address domain-specific issues such as sarcasm, negation, contextual ambiguity, and domain shift, which remain open problems in existing literature. Unlike earlier reviews that focus primarily on text-only pipelines, we also synthesize recent advances in multimodal sentiment analysis that integrate textual, audio, and visual cues from movie trailers and clips. In addition, we examine emerging concerns related to interpretability, fairness, and robustness that are often underexplored in prior surveys, and we outline future research directions including zero-shot and few-shot learning, hybrid symbolic--neural models, and real-time deployment considerations. Overall, this abstract provides a domain-focused roadmap that highlights both established solutions and unresolved challenges toward building more accurate, generalizable, and explainable sentiment analysis systems for movie review data.

cs.IT↗

Algorithms for 2D Mesh Decomposition in Distributed Design Optimization

Optimization of thin-walled structures like an aircraft wing, aircraft fuselage or submarine hull often involves dividing the shell surface into numerous localized panels, each characterized by its own set of design variables. The process of extracting information about a localized panel (nodal coordinates, mesh connectivity) from a finite element model, input file is usually a problem-specific task. In this work, a generalized process to extract localized panels from the two-dimensional (2D) mesh is discussed. The process employs set operations on elemental connectivity information and is independent of nodal coordinates. Thus, it is capable of extracting panel of any shape given the boundary and thus can be used during optimization of a wide range of structures. A method to create stiffeners on the resulting local panels is also presented, and the effect of stiffener element size on buckling is studied. The local panel extraction process is demonstrated by integrating it into a distributed MDO framework for optimization of an aircraft wing having curvilinear spars and ribs (SpaRibs). A range of examples is included wherein the process is used to create panels on the wing-skin, bounded by adjacent SpaRibs.

math.NA↗