Accelerating Computational Fluid Dynamics with Unsupervised Learning-Based Mesh Adaptation

우리나라 식생활에 밀접한 관련을 가지고 있는 채소인 양파의 수급불균형 해결을 위한 생산량 예측 모형 개발의 노력이 많은 연구를 통해 이뤄지고 있다. 하지만 양파의 수확기와 저장 가능성을 고려해 봤을 때 생산량 예측만으로는 수급불균형 해결이 어렵다. 따라서 본 논문에서는 양파의 생산량 정보와 가격의 다양한 요인이 포함되어 있으며 일상에서 쉽게 접할 수 있는 인터넷 기사를 이용하여 가격 예측을 위한 감성사전을 구축하고자 한다. 양파 기사는 2012년부터 2016년까지의 데이터를 사용하였고 도매시장 가격을 통한 문서구분을 통해 4가지 TF-IDF를 비교하여 적합한 TF-IDF를 사용하였다. 분석을 위하여 분할적 군집분석 중 k-means 군집, 밀도기반군집(DBSCAN; density based spatial cluster applications with noise), 가우시안혼합분포군집(GMM; Gaussian mixture model) 군집을 통하여 가격에 대한 긍정/부정 단어를 구분한 결과 GMM 군집이 의미 있는 긍정, 부정, 무정의 3개의 사전으로 구성되었다. 구축된 사전의 합리성을 비교하기 위하여 가격 상승 기사와 가격 하락 기사의 분류에 로지스틱 회귀분석을 적용한 결과 85.7%의 정확도로 구축된 사전의 합리성을 확인할 수 있었다.Many researches are accomplished as a result of the efforts of developing the production predicting model to solve the supply imbalance of onions which are vegetables very closely related to Korean food. But considering the possibility of storing onions, it is very difficult to solve the supply imbalance of onions only with predicting the production. So, this paper s purpose is trying to build a sentiment dictionary to predict the price of onions by using the internet articles which include the informations about the production of onions and various factors of the price, and these articles are very easy to access on our daily lives. Articles about onions are from 2012 to 2016, using TF-IDF for comparing with four kinds of TF-IDFs through the documents classification of wholesale prices of onions. As a result of classifying the positive/negative words for price by k-means clustering, DBSCAN (density based spatial cluster application with noise) clustering, GMM (Gaussian mixture model) clustering which are partitional clustering, GMM clustering is composed with three meaningful dictionaries. To compare the reasonability of these built dictionary, applying classified articles about the rise and drop of the price on logistic regression, and it shows 85.7% accuracy.

In conventional full-duplex communications, dedicated symbols are transmitted to estimate both the self-interference channel and the desired signal channel in order to perform self-interference cancellation (SIC) and to detect the desired signal. However, inaccurate channel estimation will produce residual self-interference and degrade the detection performance. In this paper, we exploit Gaussian mixture model (GMM) clustering to design a full-duplex transceiver (FDT), which is able to detect the desired signal without requiring digital domain channel estimations and SIC. The frame structure of the designed FDT contains two successive phases: labelling phase and data transmission phase. In particular, the designed FDT performs cluster labelling in the labelling phase and performs GMM clustering based on expectation-maximization (EM) algorithm in the data transmission phase. Furthermore, the labelling symbol design is optimized for the overhead reduction and clustering accuracy improving. Finally, simulation results show that, the bit error rate (BER) of the designed FDT is closed to the performance of the FDT with a maximum likelihood (ML) detector and perfect channel knowledge, meanwhile is superior to the BER performance of the FDT with a ML detector and a least square (LS) channel estimator.

(English) This thesis deals with finite element methods to solve compressible flow problems, an important branch of Computational Fluid Dynamics (CFD) whose applications are widespread in many areas of engineering and science. In spite of the increasing amount of computational resources made available for the scientific and engineering research communities, the numerical simulation of complex compressible phenomena in many practical applications is still a challenge. These type of flow problems are extremely demanding in what concerns numerical computations and memory requirements. In particular, in this thesis we investigate the possibility to solve the underlying algebraic systems in a decoupled manner, a technique usually called fractional step or segregation method. Although segregation techniques have been broadly studied and analyzed for the incompressible Navier-Stokes equations, allowing for the separate resolution of velocity and pressure unkonwns, much less has been explored for compressible problems. The interest on this type of technique not only comes from the fact that it permits a segregated calculation of the problem unkonwns (usually leading to better conditioned systems) but from the associated reduction of the computational cost. We study three different problems inside the compressible CFD research branch in separated chapters: the isentropic Navier-Stokes equations, the Navier-Stokes problem written in primitive variables (velocity, pressure and temperature), and the navier-Stokes problem written in the classical formulation with conservative variables (momentum, density, total energy). For each of these problems, first we propose a finite element stabilized formulation framed within the Variational MultiScale concept, which allows to use equal interpolation spaces for all the variables in play. Second, and once space and time discretizations are selected, we derive fractional step methods up to second order in time. Finally, all the schemes are implemented in a parallel multiphysics code and representative simulations are carried out in order to analyze the performance of the proposed techniques. (Español) Esta tesis se centra en la aplicación del método de elementos finitos a problemas de flujo compresible, unas de las ramas más importantes dentro de la mecánica de fluidos computacional (CFD, por sus siglas en inglés), y cuya aplicación es amplia en distintas áreas de ciencia e ingeniería. A pesar del aumento de los recursos computacionales que se ponen a disposición de la comunidad científica, la solución numérica de problemas complejos de flujo compresible para distintas aplicaciones prácticas es aún un reto. Este tipo de problemas de flujo son extremadamente exigentes en lo que se refiere al cálculo numérico y los requisitos de memoria computacionales. En particular, en esta tesis se estudia la posibilidad de resolver los sistemas algebraicos que subyacen tras la aplicación del método de elementos finitos de una forma desacoplada, una técnica que usualmente se denomina de paso fraccionado o método de segregación. Aunque las técnicas de segregación han sido ampliamente estudiadas y analizadas para el problema incompresible de Navier-Stokes, lo que permite un cálculo separado de las variables velocidad y presión, poco se ha explorado su aplicación a problemas de flujo compresible. El interés en este tipo de técnicas radica no solo en el hecho de que permite un cálculo segregado de las incógnitas del problema (generalmente dando lugar además a sistemas algebraicos mejor condicionados), sino también en la reducción de coste computacional que ofrecen. Se estudian tres problemas distintos dentro de la rama de investigación en CFD de flujos compresibles: las ecuaciones isentrópicas de Navier-Stokes, el problema compresible de Navier-Stokes escrito en variables primitivas (velocidad, presión y temperatura) y el problema compresible de Navier-Stokes planteado de manera clásica en variables conservativas (momento, densidad y energía total). Para cada uno de estos casos, primero se propone una formulación de elementos finitos, estabilizada bajo el marco de trabajo de las multiescalas variacionales (VMS, por sus siglas en inglés), lo que en la práctica permite utilizar la misma interpolación para las variables del problema. En segundo lugar, y una vez las discretizaciones en tiempo y espacio han sido seleccionadas, se proponen métodos de paso fraccionado para cada problema de hasta segundo orden en tiempo. Finalmente, todos los algoritmos se implementan en una plataforma de código paralelo para problemas multifísicos y se llevan a cabo simulaciones representativas con el fin de analizar el rendimiento de las técnicas propuestas.

We introduce a method based on Gaussian mixture model (GMM) clustering and level-set to automatically detect intraretina fluid on diabetic retinopathy (DR) from spectral domain optical coherence tomography (SD-OCT) images in this paper. First, each B-scan is segmented using GMM clustering. The original clustering results are refined using location and thickness information. Then, the spatial information among every consecutive five B-scans is used to search potential fluid. Finally, the improved level-set method is used to obtain the accurate boundaries. The high sensitivity and accuracy demonstrated here show its potential for detection of fluid.

Research on FT-transformer-based intelligent discrimination model for roof stability classification of retreating coal roadways

In this work we propose a strategy to jointly estimate activity and borders of oncological lesions in PET/CT. The starting step is constituted by a lesion contouring on PET image volume and by a Gaussian Mixture Model (GMM) clustering which individuates a set of regions in the lesion area (lesion, uncertainty, lesion spillout, organ). A maximum likelihood (A WOSEM) reconstruction step refines regions' borders and estimates a mean convergence activity for the lesion region. It applies a model of the scanner Point Spread Function (PSF) to recover blurring and it contemporaneously works on regional basis functions and single voxels. The area outside the four regions is frozen (i.e. not updated). The algorithm was validated on an anthropomorphic phantom in which lesions have been simulated with zeolites (clinoptilolite samples, volume 0.6 - 5.2 ml) loaded with 18F-FDG. Zeolite borders for ground truth definition were derived segmenting zeolites on coregistered CT images. For each zeolite, three different initial contouring were considered, corresponding to volumes about 100%, 60% and 140% of the true volume: the GMM clustering was able to robustly delineate regions independently from the initial contouring (variations 0.75) and in estimating zeolite activity (activity error 1ml). Suboptimal results were found for zeolites at the border of the axial FOV, since the PSF model, supposed invariant to axial shift, was inadequate at the axial borders. The proposed strategy appears promising and can be proposed as a general approach for a semi-automatic quantification and segmentation of lesions previously detected on standard clinical images. It will be further validated on data sets provided with a ground truth.

Deep clustering was applied to unlabeled, automatically detected signals in a coral reef soundscape to distinguish fish pulse calls from segments of whale song. Deep embedded clustering (DEC) learned latent features of the signals and formed clusters using fixed-length power spectrograms of the signals. Handpicked spectral and temporal features were also extracted and clustered with Gaussian mixture models (GMM) and conventional clustering. DEC, GMM, and conventional clustering were tested on simulated datasets of fish pulse calls and whale song units with randomized bandwidth, duration, and SNR. Both GMM and DEC achieved high accuracy and identified clusters with fish calls, whale song units, and simultaneous fish calls and whale song units. Conventional clustering methods K-means and hierarchical agglomerative clustering had low accuracy in scenarios with unequally sized clusters or multiple signals. Experimentally detected fish pulse calls and whale song units recorded near Hawaii in February–March 2020 were clustered with DEC, GMM, and conventional clustering. DEC demonstrated the highest accuracy on a small, manually labeled dataset and successfully separated clusters of fish calls and whale song units. GMM achieved high recall for detecting whale song units but accuracy. Both GMM and DEC overpredicted the number of whale song unit signals.

In rarefied gas flows, discontinuity phenomena such as velocity slip and temperature jump commonly appear in the gas layer adjacent to a solid boundary. Due to the physical complexity of the interactions at the gas-solid interface, particularly in the case of systems with local nonequilibrium state, boundary models with limited number of parameters cannot completely describe the reflection of gas molecules at the boundary. In this work, the Gaussian mixture (GM) model, which is an unsupervised machine learning technique, is employed to construct a statistical gas-solid surface scattering model based on the collisional data obtained from molecular dynamics (MD) simulations. The GM model is applied to study Couette flow for different inert gases (Ar and He) confined between two parallel infinite gold walls at different temperatures. A direct comparison between the results obtained from the GM model and the Cercignani-Lampis-Lord (CLL) scattering kernel against the MD collisional data in terms of the distribution of the predicted postcollisional velocities, and accommodation coefficients has shown that the results from the GM model are an excellent match with the MD results outperforming the CLL scattering kernel. As an example, for He gas, while the predicted energy accommodation coefficient by the CLL model is more than two times higher than the MD predictions, the value computed by the GM model is in excellent agreement with the MD results. This superior performance of the GM model confirms its high potential to derive a generalized boundary condition in systems encountered with highly nonequilibrium and complex gas flow conditions.

This study aimed to assess the impact and vulnerability of climate change by classifying 26 clusters of meteorologically homogeneous regions. To determine the optimal clustering method, both K-means and Gaussian Mixture Model (GMM) clustering were analyzed using the effective storage capacity to watershed area ratio. The optimal number of clusters was derived based on several evaluation metrics, including the Silhouette Score, Calinski-Harabasz Index, Akaike Information Criterion (AIC), and Bayesian Information Criterion (BIC). Ultimately, GMM clustering was identified as the optimal method, with the best clustering results obtained at k = 4 for an effective storage capacity of 100,000 to 400,000 tons and k = 5 for an effective storage capacity of 400,000 to 10,000,000 tons. Additionally, standard reservoirs applicable to agricultural production infrastructure design standards were identified based on homogeneous weather region clusters, the optimal clustering method, and centroid results. The findings of this study can serve as fundamental data for the development and revision of design standards, contributing to more climate-resilient agricultural infrastructure.

Urban hydrological systems are increasingly vulnerable to rainfall variability and extreme precipitation events, necessitating robust analytical frameworks for spatial classification and predictive diagnostics. This study addresses a key gap in urban climate analytics by applying unsupervised machine learning models to classify rainfall patterns across six major Indian cities, Bangalore, Chennai, Delhi, Hyderabad, Kolkata, and Mumbai, each representing distinct climatic zones and monsoon regimes. Using gridded daily rainfall data from the India Meteorological Department (IMD), spanning 1981–2023 at a spatial resolution of 0.25° × 0.25°, the analysis reveals intensifying rainfall trends in Mumbai and Kolkata, consistent with urbanisation‐induced climatic perturbations. Notable extreme events in Chennai (2005, 2015, 2021) and Bangalore (2005, 2017, 2022) are linked to cyclonic activity and elevated atmospheric moisture fluxes. Six clustering models, DBSCAN, Spectral Clustering, K‐Means, OPTICS, Gaussian Mixture Model (GMM), and Agglomerative Clustering, were applied to evaluate spatial coherence and anomaly detection. DBSCAN achieved the highest silhouette score (0.82), followed by Spectral Clustering (0.80), K‐Means (0.70), and OPTICS (0.44; Accuracy: 0.06). GMM and Agglomerative Clustering provided comparative insights into probabilistic and hierarchical rainfall structures. Despite their utility in spatial classification, all models produced negative R 2 values in time‐series forecasting, reaffirming the limitations of clustering algorithms for temporal prediction tasks. Future work should explore hybrid deep learning approaches such as LSTM or ARIMA to better capture sequential dependencies and enhance operational forecasting. This study presents a scalable framework integrating climate trend diagnostics with AI‐based classification, contributing to improved urban rainfall pattern detection, flood risk assessment, and climate‐resilient planning.

In this paper, we propose a novel method which utilizes the Gaussian Mixture Model (GMM) to estimate the leakage distribution of a circuit. Our proposed method assumes that the leakage distribution can be represented using the GMM which can cover any continuous function. After the GMM clustering using the leakage simulation data, the leakage distribution of the input circuit can be obtained. The experimental results with the K-S test showed that the proposed method exhibited 1.82e+05 times larger p-value and 7.74e-01 times smaller K-S statistics compared to the state-of-the-art benchmark method on average.

Machine learning (ML) is revolutionizing industries by providing innovative solutions that drive sustainable economic growth and market efficiency. The integration of ML techniques in various sectors, including finance, healthcare, manufacturing, and energy, offers significant potential for enhancing productivity, optimizing resource allocation, and fostering innovation. This review explores the transformative impact of ML on economic growth and market efficiency, emphasizing its role in predictive analytics, decision-making, and automation. In finance, ML algorithms are employed to predict market trends, manage risks, and detect fraudulent activities. By analyzing vast amounts of data in real-time, ML models enable financial institutions to make informed decisions, reduce operational costs, and improve customer experiences. In healthcare, ML applications range from diagnosing diseases and predicting patient outcomes to optimizing treatment plans and managing healthcare resources, thereby improving patient care and reducing costs. The manufacturing sector benefits from ML through predictive maintenance, quality control, and supply chain optimization. ML models analyze data from sensors and production lines to predict equipment failures, ensuring timely maintenance and minimizing downtime. This leads to increased productivity and reduced waste, contributing to economic sustainability. Energy management is another critical area where ML drives efficiency. ML algorithms optimize energy consumption, forecast renewable energy production, and enhance grid management. By integrating ML into energy systems, organizations can reduce their carbon footprint and operational costs while promoting sustainable energy practices. Moreover, ML enhances market efficiency by enabling businesses to understand consumer behavior and preferences. Retailers use ML to personalize marketing strategies, optimize pricing, and manage inventory effectively. This not only boosts sales but also improves customer satisfaction and loyalty. Despite the significant benefits, the adoption of ML presents challenges such as data privacy concerns, algorithmic bias, and the need for skilled professionals. Addressing these challenges requires robust regulatory frameworks, ethical guidelines, and investment in education and training. In conclusion, harnessing ML techniques is pivotal for driving sustainable economic growth and market efficiency. By leveraging ML's capabilities in predictive analytics, decision-making, and automation, industries can achieve higher productivity, optimized resource utilization, and innovative solutions that contribute to long-term economic sustainability and competitiveness. As ML continues to evolve, its potential to transform various sectors and promote sustainable development remains immense.

Indoor environmental quality is crucial for human health and comfort, necessitating precise and efficient computational methods to optimise indoor climate parameters. Recent advancements in machine learning (ML) and computational fluid dynamics (CFD) are promising. However, applying ML to complex building airflow presents challenges. This research aims to investigate the integration of ML with CFD in the context of built environment applications using a systematic review approach. It highlights a critical knowledge gap: the need to synthesise innovative approaches that address the limitations of indoor modelling using data-driven ML methods. The review examines contemporary literature, identifying current developments and suggesting potential future directions. It delves into the innovations in combining ML with CFD to predict thermal comfort and indoor air quality, uncovering key limitations such as the lack of high-quality experimental data for training and validation, the computational complexity of detailed CFD simulations, and the interpretability issues of ‘black-box’ ML models. The emergence of data-driven techniques in fluid mechanics offers promising prospects for modelling in the built environment. Future research should focus on incorporating physics-based rules in ML models, adapting turbulence closure models for indoor flows, and enhancing model validation using real-world datasets. The research emphasises the synergistic relationship between ML and CFD; it proposes pathways to overcome current limitations, aiming to enhance the precision and efficiency of indoor environment modelling through their integration.

Computational Fluid Dynamics (CFD) is regarded as an important tool for analyzing the flow field, thermal environment, and air quality around the built environment. However, for built environment applications, the high computational cost of CFD hinders large-scale scenario simulation and efficient design optimization. In the field of built environment research, surrogate modeling has become a key technology to connect the needs of high-fidelity CFD simulation and rapid prediction, whereas the low-dimensional nature of traditional surrogate models is unable to match the physical complexity and prediction needs of built flow fields. Therefore, combining machine learning (ML) with CFD to predict flow fields in built environments offers a promising way to increase simulation speed while maintaining reasonable accuracy. This review briefly reviews traditional surrogate models and focuses on ML-based surrogate models, especially the specific application of neural network architectures in rapidly predicting flow fields in the built environment. The review indicates that ML accelerates the three core aspects of CFD, namely mesh preprocessing, numerical solving, and post-processing visualization, in order to achieve efficient coupled CFD simulation. Although ML surrogate models still face challenges such as data availability, multi-physics field coupling, and generalization capability, the emergence of physical information-driven data enhancement techniques effectively alleviates the above problems. Meanwhile, the integration of traditional methods with ML can further enhance the comprehensive performance of surrogate models. Notably, the online ministry of trained ML models using transfer learning strategies deserves further research. These advances will provide an important basis for advancing efficient and accurate operational solutions in sustainable building design and operation.

This study aims to improve early bowel cancer detection by introducing a non-visual approach to lesion characterisation, with a focus on precision and reliability. The study makes use of the sensitive dynamics of a robotic capsule self-propelling in the bowel to biomechanically characterise encountered lesions. The method incorporates a two-phase machine learning involving supervised regression as first phase and unsupervised clustering as final phase. The initial phase employs multi-layer perceptron (MLP), support vector regression, and Gaussian process regression to predict the stiffness of lesions, expressed as Young’s modulus (E) from capsule displacement signals. The final phase involving Gaussian mixture model (GMM) clustering achieved over 89 % accuracy for both simulation and experiment. The GMM demonstrated improved precision and reliability when incorporating predictions from more accurate regression models, such as MLP, though it struggled with high and closely related E-values.