1. Ph.D Theses

Permanent URI for this collectionhttps://idr.nitk.ac.in/handle/1/11

Browse

Now showing 1 - 20 of 1091

3d Convolutional Neural Network Architectures for Volumetric Medical Image Segmentation
(National Institute of Technology Karnataka, Surathkal., 2024) S. Niyas
Computer-aided medical image analysis plays a critical role in supporting medical practitioners with expert clinical diagnoses and determining optimal treatment plans. Currently, convolu tional neural networks (CNNs) are widely regarded as the preferred method for automated medical image analysis due to their ability to autonomously learn relevant features from train ing data. However, most cutting-edge semantic image segmentation techniques rely on two dimensional (2D) CNN models, which do not fully exploit the inter-slice information available in cross-sectional imaging modalities, such as MRI volumes. This limitation underscores the need for more advanced approaches to better utilize the three-dimensional (3D) data inherent in these imaging techniques. In this thesis, we present a comprehensive evaluation of various techniques employed in 3D deep learning for medical image segmentation. With the rapid advancements in 3D imaging systems and excellent hardware and software support to process large volumes of data, 3D deep learning methods are gaining popularity in medical image segmentation. However, traditional 3D CNN-based segmentation models require substantial computational resources, extensive memory, and typically larger datasets than 2D CNN approaches. To address these challenges, we propose a 3D CNN segmentation model that e!ciently extracts information across slices and mitigates several limitations associated with traditional 3D CNN techniques. The method aims to retain the advantages of both 2D CNN and 3D CNN methods by e”ectively designing input data slices and the CNN architecture. In this study, we proposed a shallow sliced stacking approach to reduce the depth of input 3D data to maintain a good segmentation accuracy with minimum computation overhead and model complexity. Incorporating residual connections in the encoder path also facilitates the extraction of multi-scale features without significantly increasing the model complexity. Accurate diagnosis of various medical conditions often requires the simultaneous analysis of multiple image characteristics. For instance, Focal Cortical Dysplasia (FCD) lesion detec viii tion can be significantly enhanced by incorporating data on cortical thickness maps along with f luid-Attenuated Inversion Recovery (FLAIR) Magnetic Resonance Imaging (MRI) scans. Ad ditionally, employing multi-axis analysis of 3D cross-sectional imaging can substantially improve diagnostic performance. Inspired by these concepts, we propose a 3D deep learning model em ploying a multi-view, dual encoder-decoder architecture. The model also incorporates various architecture-wise enhancements, including an end-to-end cascaded approach for transitioning from coarse to fine segmentation, 3D Attention modules for maintaining consistency between encoder and decoder pairs, and dual-task learning. In our study, we apply this model to pro cess FLAIR MRI volumes alongside corresponding cortical thickness maps, aiming to e”ectively detect FCD lesions. Generative Adversarial Networks (GANs) have significantly impacted the field of image anal ysis, and they have been successfully employed for tasks such as image segmentation. Hence, this study also proposes a 3D attention-driven Vox2Vox CNN network that leverages the power of a 3D GAN to accurately segment acute stroke lesion cores in Computed Tomography Per fusion (CTP) scans. This methodology also incorporates valuable insights derived from our prior models relevant to this research. The segmentation framework incorporates two super vised GAN components: a generator and a discriminator. The generator module is designed to process 3D slices from CTP maps and learn to generate 3D binary prediction masks that closely match the ground truth for stroke lesions. Concurrently, the discriminator module is trained to distinguish between the outputs generated by the generator and the actual ground truth. Overall, this thesis demonstrates the e!cacy of 3D deep learning in identifying malig nancies from cross-sectional imaging modalities, including CT and MRI, thereby enhancing the capabilities of automated Computer-Aided Detection (CAD) systems.
A Comprehensive Investigation of Thermal and Flow-Resistance Behaviour of Metal Based Porous Media
(National Institute of Technology Karnataka, Surathkal., 2024) G., Trilok; N., Gnanasekaran
This thesis work presents numerical investigation of metal based porous media such as metal foams and stacked wire mesh porous structures, focusing on addressing the issue of incurred flow resistance that is always accompanied with the enhanced heat transfer associated with such media. Key features of porous medium such as their twin structural properties (porosity and pore density), thickness and method of formation (stacking types in terms wire mesh porous structures) are identified as potential influencing parameters that play a key role in the thermo-hydraulic phenomenon. In the first part, influence of porosity and pore density of porous media is demonstrated for their combined effect on flow resistance and heat transfer enhancement behavior. Significance of considering both of these twin structural properties in analyzing the characteristics of porous medium particularly in forced convection regime is further emphasized through Nusselt number correlations. In the second part, thickness of porous medium is considered as another parameter along with the structural properties, and various trade-off scenarios between enhanced heat transfer and incurred flow resistance is comprehensively analyzed. TOPSIS A multi-objective, multi-attribute decision making technique is utilized in this regard, and unique potentials of a porous medium corresponding to its various combination of structural and thickness conditions are evaluated in terms of their ability to minimize flow resistance and maximize heat transfer. In the last part, potentials of stacked wire mesh porous structures are investigated for their various trade-off scenarios between enhanced heat transfer and incurred flow resistance. Expressions pertaining to key morphological features such as porosity, pore density and specific surface area of wire mesh porous structures of various stacking types are derived and used in the porous media modeling to comprehensively analyze the phenomenon of increased pressure drop with increase in heat transfer corresponding to variations in structural properties (porosity and pore density), stacking types and thickness scenarios.
A Framework for Enhancing Sustainable Competency of Small and Medium Contractors in the Ethiopian Construction Industry
(National Institute of Technology Karnataka, Surathkal, 2024) Bekele, Abraham Aboneh; Mahesh, Gangadhar
Small and medium contractors (SMCs) are vital in promoting socioeconomic development, particularly in developing economies, as they constitute a significant portion of the construction industry (CI). Their significance lies in their ability to create employment opportunities, generate revenue, develop infrastructure, and have strong links with other sectors of the economy, which have multiple effects on the country's growth. While acknowledging their significant importance and contribution, it is evident that there is a need to enhance and maintain their competency in light of various challenges affecting their growth. This research aims to devise a sustainable competency development framework for enhancing the competitiveness of SMCs in the Ethiopian CI and establish management mechanisms to facilitate their business sustenance. The specific objectives are to: identify factors affecting sustainable competency of SMCs; assess the effectiveness of the development programs in enhancing progress in the CI; evaluate the prevailing opportunities to create sustainable SMCs and develop appropriate improvement mechanisms to exploit these opportunities; and develop sustainable competency development framework for SMCs. The study employed qualitative and quantitative research methods. This approach allowed for gathering input from industry stakeholders, which was then used to develop the framework. The findings of this study provide a comprehensive understanding of the factors impacting the sustainable competency of SMCs in Ethiopia. The study identified the major underlying factors or challenges, such as the lack of project management skills; low-profit margin due to high competition; inability to access plants and equipment; and the inability to access financial resources emanating from endogenic core sources. Additionally, the study also identified factors or challenges stemming from exogenic core sources including unfavourable financial policy, lack of trust between parties in the industry, and uncertainty in supplies of materials and prices. Furthermore, the study's findings offer valuable insights into potential improvements that could enhance the prospects of sustainable SMCs development in Ethiopia. These improvements encompass encouraging local construction material producers and enhancing their capacity, advocating for an industry-based education system, introducing sector-specific financing programs, and implementing project planning, scheduling, and performance tracking practices. The study's findings highlight priority areas for enhancing competitiveness, providing valuable guidance for policymakers, regulators, entrepreneurs, and other stakeholders in making informed decisions.
A Framework for Human Activity and Behavioural Pattern Recognition in Multimodal Sensor Smart Home Environment
(National Institute of Technology Karnataka, Surathkal, 2024) Kolkar, Ranjit; Geetha V.
Human Activity Recognition (HAR) has become a subject of significant interest due to its potential applications in various fields, including healthcare, sports, and user profiling. There are four main types of sensor-based HAR: wearable, ambient, camera, and hybrid sensor-based recognition. Smartphones, with their built-in sensors, have emerged as valuable tools for HAR, and other sensors like Passive Infrared (PIR), load sensors, smart switches, and smartwatches are extensively used in HAR systems along with vision-based sensors. Despite advancements, accurately recognizing human activities remains challenging due to the complexity and diversity of sensors used and the intricate nature of human activities. Each sensor type has advantages and limitations, making selecting appropriate sensors a challenging task requiring a comprehensive understanding of their characteristics. While there are existing applications of HAR, there are still significant opportunities to address various challenges. This work addresses several challenges in improving recognition efficiency, integrating multimodal sensors, achieving synchronization between heterogeneous sensors, collecting long-hour data using these sensors, and developing a cost-effective framework for human activity recognition and behavioral patterns in the daily life of an elderly person. The thesis work addresses the challenges and develops a framework for HAR and behavioural pattern recognition using multimodal sensors in a smart home environment. First, we design and develop a deep learning-based solution to recognize the activities based on sensors present in the smartphone. Later, we create and curate a dataset for long-hour human activities in a multimodal sensor-equipped smart home environment and follow to design and develop a human behavioural pattern recognition system in a smart home environment. The first work focuses on comparing the performance of various deep learning models Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU) for HAR using smartphone-based sensors. The study explored various datasets and recognition models, providing valuable insights into the overall HAR architecture. The primary objective of this research is to accurately recognize basic human activities such as walking, sitting, standing, going upstairs, going downstairs, and lying down. The models were trained and evaluated on well-known datasets like Wireless Sensor Data Mining (WISDM) and University of California, Irvine, Human Activity Recognition (UCI-HAR). Through rigorous experimentation, the performance of the models on these datasets was significantly improved using the GRU model, laying the foundation for the subsequent research objectives. Additionally, the thesis proposed a novel approach called Spider Monkey Optimization (SMO)-based deep neural network to enhance HAR’s accuracy and precision further. The proposed system was evaluated on various datasets involving similar activities, including UCI-HAR, WISDM, Royal Institute of Technology (KTH) action, and Physical Activity Monitoring using Accelerometers, Gyroscopes, and Magnetometers (PAMAP2). The optimization improved the performance and the reduced training time, making it practical for real-world applications. The second work in the thesis involves the collection of long-hour datasets using a multimodal approach. It has been observed from the literature and our previous work that understanding human behaviour patterns solely based on basic activities and smartphone sensors is challenging. Therefore, in this work, we combined smartphone sensor data with ambient sensors to better understand the user’s context. The context includes room occupancy detection using PIR sensors, water bottle level indication using load sensors, and monitoring the status of the TV, bathroom lights, and mirror bulb lights using smart switches. We derived a broader range of activities beyond the basic ones by combining and proposing a hybrid sensor-based data collection approach for two individuals over an extended period. The third work in the thesis also proposes a novel priority-based labelling technique for data segmentation to retain user context while labelling. This enhanced dataset enables us to gain valuable insights into human behaviour patterns in dayto- day life. Additionally, through a comprehensive analysis of user data, we can derive the user’s personality and provide feedback on their behaviour patterns to improve or analyze activities performed over time. The research identifies various applications, such as elderly monitoring systems, personality identification, and behaviour analysis, all aimed at improving health and well-being. KEYWORDS: HAR, SMO, Wearable sensors Smartphone sensors, Deep learning, Ambient sensors, Internet of Things (IoT), PIR, Elderly monitoring, User profiling, Behaviour patterns.
A Less Invasive and Computationally Efficient Silent Speech Interface Using Facial Electromyography
(National Institute of Technology Karnataka, Surathkal, 2023) Abdullah, Asif; CMC, Krishnan
Silent Speech Interface (SSI) is one of the promising areas of Human Computer Interaction (HCI) research. The Surface Electromyography (SEMG) based SSI is a technique where the electric activity of facial muscles are used to detect speech. The existing SSI techniques use computationally expensive methods and complex machine learning algorithms for the identification of silently uttered speech. The increased computational expense prevents real time implementation of SSI models especially in cost efficient applications such as communicative assistance for laryngectomy patients. Thus the objective of this research work is to develop a less complex and computationally less expensive SEMG based SSI model with superior accuracy. To achieve this goal, investigations are done on many feature extraction methods to check if they are suitable for SEMG based SSI. Detrended Fluctuation Analysis (DFA) is found to be promising for the recognition of silent speech using SEMG. The use of computationally less expensive classification algorithms was envisioned in this research work to develop a simpler and faster SSI model. The research identified K Nearest Neighbors and Decision Trees as suitable pattern recognition algorithms for this work. The number of channels associated with SEMG based SSI is also a matter of important concern. A state-of-the-art model uses seven channels of SEMG data for the recognition of silent speech. Considering the use of some unipolar electrodes along with the bipolar ones, the number of electrodes to be accommodated on the face usually ranges from eight to twelve. For practical applications this is a high number especially in the case of medical conditions faced by laryngectomy patients. Too many number of electrodes on the subject’s face creates inconvenience to the user who have undergone laryngectomy. It can hinder facial movement and can also contribute to the occurrence of cross talk between different facial muscles. Thus the reduction of number of channels is necessary and hence it is included as an important objective of this research work. The effectiveness of using DFA for successful channel reduction is investigated thoroughly. The analysis using DFA is also compared with channel reduction performed on models that employ existing state-of-the-art methods. The availability of reliable data is vital for every researcher to carry out fruitful research. But as far as SEMG based SSI is considered, data availability is a major concern. There are very few reliable data sets (with sufficient vocabulary) available for SEMG based research. This is primarily due to the popular research orientation towards acoustic speech recognition. Thus the creation of an extensive database is a promising aspect to consider and the initial steps to that cause is also considered as an important goal of this research work. Hardware purchase and assembly, drafting of a detailed data acquisition methodology, and a sample data collection is done as part of the work.
A Study on Acyclic Edge Coloring and Domination Number of a Graph
(National Institute of Technology Karnataka, Surathkal., 2024) Kulamarva, Shashanka; Hegde, Suresh M.; Basavaraju, Manu
Let G =(V,E) be a graph with n vertices and let C be a given set of colors. A proper edge coloring of the graph G with the colors from the setC is a function f : E →C such that f(e1) ̸ = f(e2) for any adjacent edges e1 and e2. An acyclic edge coloring of a graph is a proper edge coloring without any bichro matic cycles. The acyclic chromatic index of a graph G denoted by a′(G), is the min imum positive integer k such that G has an acyclic edge coloring with k colors. It was conjectured by Fiamˇc´ ık (1978) (and independently by Alon, Sudakov and Zaks (2001)) that a′(G) ≤ ∆+2 for any graph G with maximum degree ∆. Linear arboricity of a graph G, denoted by la(G), is the minimum number of linear forests into which the edges of G can be partitioned. It was conjectured by Akiyama, Exoo and Harary (1980) that for any graph G, la(G) ≤ ⌈∆+1 2 ⌉. Agraph is said to be chordless if no cycle in the graph contains a chord. By a result of Basavaraju and Chandran (2010), if G is chordless, then a′(G) ≤ ∆+1. Machado, de Figueiredo and Trotignon (2013) proved that the chromatic index of a chordless graph is ∆ when ∆ ≥ 3. In this thesis, it is proved that for any chordless graph G, a′(G) = ∆, when ∆ ≥ 3. One can see that this is an improvement over the result of Machado et al. (2013), since any acyclic edge coloring is also a proper edge coloring and we are using the same number of colors. The thesis also provides the sketch of a polynomial-time algorithm that takes a chordless graph G as an input and returns an optimal acyclic edge coloring of G as the output. As a byproduct, one can prove that la(G) = ⌈∆ 2⌉, when ∆ ≥3. To obtain the result on the acyclic chromatic index of a chordless graph, an extremal structure in chordless graphs has been proved which is a refinement of the structure given by Machado et al. (2013) in the case of chromatic index. This extremal structure might be of independent interest. Agraph G is said to be k-degenerate if every subgraph of G has a vertex of degree at most k. Basavaraju and Chandran (2010) proved that the acyclic edge coloring con jecture is true for 2-degenerate graphs. In the thesis, it is proved that for a 3-degenerate graph G, a′(G) ≤ ∆+5, thereby bringing the upper bound closer to the conjectured bound. The thesis also considers the class of k-degenerate graphs with k ≥ 4 and improves iii the existing upper bound given by Fiedorowicz (2011) for the acyclic chromatic index of a k-degenerate graph. Aset of vertices D in graph G is said to be a dominating set if every vertex which is not in D is adjacent to a vertex in D. The size of a minimum dominating set of G is said to be the domination number of G and is denoted by γ(G). A classical upper bound for the domination number of a graph G having no iso lated vertices is ⌊n 2 ⌋. However, for several families of graphs, we have γ(G) ≤ ⌊√ n⌋ which gives a substantially improved upper bound. By the multiplicative version of the Nordhaus-Gaddum type result for the domination number of a graph, for any graph G with n vertices and Gbeing the complement of G, we have γ(G)γ( will imply that for every graph G, either γ(G) ≤ ⌊√ n⌋ or γ( G) ≤n. This result G) ≤⌊√ n⌋ or both. The thesis presents some conditions necessary for a graph G to have γ(G) ≤ ⌊√ and some conditions sufficient for a graph G to have γ(G) ≤ ⌊√ of all connected graphs G with γ(G) = ⌊√ n⌋, n⌋. A characterization n⌋ is also provided. Further, the thesis also provides proof for the statement that if the condition: rad(G) =diam(G) =rad( G)=diam( G)=2 is not satisfied for a graph G, then the task of deciding whether it is γ(G) ≤ ⌊√ G) ≤⌊√ γ( n⌋ or n⌋ can be done in polynomial time. The thesis concludes with a conjec ture that a slightly weaker decision problem can be solved in polynomial time for any arbitrary graph. Along with the above mentioned results, the thesis also introduces a new termi nology of t-complementary self-centered graphs and a new concept of freeable colors, thereby contributing to the literature of acyclic edge coloring and the domination in graphs. This concept of freeable colors turns out to be a very useful tool for proving the upper bounds on the acyclic chromatic index.
A Study on Certain Positivity Classes of Operators in Hilbert Spaces
(National Institute of Technology Karnataka, Surathkal., 2024) A., Rashid; P., Sam Jhonson
In mathematical optimization theory, the linear complementarity problem, which is stated as, given a vector q in a finite-dimensional real vector space and an n×n real matrix A, then, finding a vector x that satisfies the system of inequalities x ≥ 0, q+Ax ≥ 0, xT(q+Ax) = 0, plays a vital role in many areas such as bimatrix game theory, mar ket equilibrium, computational complexity, and many more. The nature of the solution of the linear complementarity problem can be discussed with the help of the matrix A involved in the problem. The sign-reversing is a property of matrices along with a given vector, which is stated as an n×n matrix A reverses the sign of a vector x in an n-dimensional real vector space if it satisfies xi(Ax)i ≤ 0 for all index i. The con cept of sign-reversing is a useful tool to identify and characterize certain matrix classes involved in linear complementarity problems. The sign-reversing set of a matrix A is defined as {x : xi(Ax)i ≤ 0, ∀i}. In this thesis, we characterize the sign-reversing set of an arbitrary square matrix A in terms of the null spaces of the matrices DA−A−D, where D is a diagonal matrix such that 0 ≤ D≤I. The matrices which have convex sign-reversing sets include large classes of matrices; we discuss a subclass of matrices in which the convexity of the sign-reversing set is characterized. Areal square matrix A is called a P-matrix if all its principal minors are positive. In 2016, Kannan and Sivakumar extended the notion of P-matrix to infinite-dimensional Banach spaces relative to a given Schauder basis by using the sign non-reversal prop erty of matrices. Motivated by their work, we discuss P-operators on separable real Hilbert spaces with the help of the inner product structure of the Hilbert spaces. We also investigate P-operators relative to various orthonormal bases. In this thesis, we define the concept of a sign-reversing set for operators on separa ble Hilbert spaces with the help of the inner product structure of the Hilbert spaces rel ative to a given orthonormal basis. We also generalize some special classes of matrices to operators in infinite-dimensional Hilbert spaces with the help of the sign-reversing property of operators. The thesis also studies the spectral theory of certain positive operators under consideration.
A Study on the Effect of Oil Price on Stock Returns of the Renewable Energy Firms in India
(National Institute of Technology Karnataka, Surathkal., 2023) Mishra, Lalatendu; H., Acharya Rajesh
The renewable energy share is increasing the total energy consumption. A substantial growth in the renewable energy generation is seen in India which is going to be one of the most renewable energy producer in the world. Like other sector, renewable energy sector’s stock return reaction to crude oil price is studied which is growing due to the increasing importance of renewable energy sector. This is an incremental study at firm level and in the context of India. The study period was from January 2009 to March 2022. All renewable energy firm listed in the National stock Exchange of India are considered for the study. The renewable energy sector or industry can be divided into sub-sectors such as solar, wind, etc. But, most Indian renewable energy firms deal in different business activities. Thus, all renewable energy firms are categorized into four sub-group, such as firms dealing in renewable energy products and services, standalone firms dealing in renewable energy production, firms dealing in both renewable and non-renewable energy production, firms dealing in renewable energy and other businesses. For crude oil price, the Indian basket oil price is considered. In first objective, the effect of oil price change has a direct relationship with stock return in all market conditions in the whole sample of Indian renewable energy firms. The direct effect of oil price change on stock return sees a declining trend in the case of firms dealing in renewable energy and other business activities. In remaining sub groups of renewable energy firms, the direct effect of oil price is seen without any particular trend. And this effect is found in all market conditions. In the second objective, oil price uncertainty has an indirect effect in bearish market conditions and a direct effect on stock returns of Indian renewable energy firm in the extreme bullish market conditions. In the sub groups, oil price uncertainty has an indirect effect in bearish market conditions and a direct effect during normal and bullish market conditions on stock returns. No asymmetric effect of oil price uncertainty on stock return is observed. In third objective, the oil price asymmetric effect only on stock returns of the standalone renewable energy production firms is seen in the short run. In the long run, the asymmetric effect is found in the whole sample and other sub-groups of renewable energy firms, except firms dealing in production of renewable and non-renewable energy firms. In fourth objective, oil supply shock has a direct relationship with stock return only in bullish market condition in the whole sample of renewable energy firms. This pattern is seen in the sub-group of firms. In normal market condition, demand specific to oil price shock has a direct effect on stock return. This interaction is seen in bearish market conditions also. In the case of standalone renewable energy firms, this relationship has a strong presence. In the remaining market conditions, the structural oil shocks have an inverse relationship with stock return. The result of the study will also be helpful to the investors and portfolio managers to maximize the investment return by observing the possible movement in stock prices of the renewable energy sector firms. The government and the regulatory authorities can understand the behavior of the renewable energy sector stocks in response to oil price. They could frame appropriate policies to ensure investment growth in this sector.
A Study on the Role of Hyperparameter Optimization (HPO) iN Image Classification Problems Using Automl Models
(National Institute of Technology Karnataka, Surathkal, 2024) Vincent, Amala Mary.; P. Jidesh
Hyperparameter optimization (HPO) has a profound impact on the performance of machine learning models. This work investigates various HPO techniques and offers valuable insights into their effectiveness, particularly in the context of image classification with the aid of AutoML models. The study places a special focus on Bayesian optimization and introduces the innovative application of genetic algorithms, differential evolution, and covariance matrix adaptation - evolutionary strategy for acquisition function optimization. Comparative analysis reveals that these evolutionary variants significantly enhance the performance of standard Bayesian optimization, with genetic algorithms emerging as less effective for acquisition function optimization. In real-time scenarios, where deep learning models often involve a multitude of hyperparameters, finding the optimal configuration becomes even more challenging. To illustrate the practical implications of HPO, the thesis presents two real-time case studies. The first case study centers on the analysis of land use and land cover changes in the Ernakulam district of Kerala using sentinel-2 images from 2019 and 2023. Employing machine learning models aided by evolutionary-based hyperparameter optimization, the study successfully tracks spatial and temporal changes in land use patterns. Furthermore, a post-classification change detection was carried out to elucidate the scale and pace of urban growth following the COVID-19 pandemic. The second study addresses flood vulnerability prediction in Kerala, India, leveraging machine learning models and AutoML systems. A three-dimensional convolutional neural network (CNN), coupled with Bayesian optimization and evolutionary algorithms like differential evolution and covariance matrix adaptation, leads to enhanced accuracy, precision, recall, AUC, and kappa scores compared to AutoML models. This thesis also ventures into the realm of meta-learning, a potential approach for tackling the data scarcity inherent in real-time image classification applications. Focusing on metric-based meta-learning models, it explores the integration of hyperparameter optimization techniques to enhance the performance of these models. The innovative bilevel optimization framework, combining meta-learning and hyperparameter optimization, is introduced, with Bayesian optimization and its recent variants employed i to fine-tune hyperparameters. Computational experiments conducted on Omniglot and ImageNet datasets provide insights into the impact of different hyperparameter optimization algorithms, with meta-testing accuracy serving as the basis for conclusions. In summary, the thesis focuses on the critical role of hyperparameter optimization in improving the machine learning model performance, emphasizing its practical applications through real-time case studies. Additionally, it explores the synergy of metalearning and hyperparameter optimization, offering a comprehensive view of cuttingedge techniques in the field.
AB Initio Studies of the Ground State Structure and Properties of Boron Carbides and Ruthenium Carbides
(National Institute of Technology Karnataka, Surathkal, 2016) G, Harikrishnan; K. M, Ajith
This work investigates the ground state structure and properties of Boron Carbides (B12C3 and B13C2 stoichiometries) and Ruthenium Carbides (RuC, Ru2C and Ru3C stoichiometries), each belonging to a class of hard materials. Exhaustive crystal structure search using evolutionary algorithm and density functional theory is performed in each of these stoichiometries. The lowest energy structures emerging from the structure search are further relaxed and their ground properties are computed using DFT. The work in B12C3 stoichiometry provides the first independent confirmation using structure search that B11Cp(CBC) is the ground state structure of this stoichiometry. It is established that mechanically and dynamically stable structures with base-centered monoclinic symmetry can be at thermodynamical equilibrium at temperatures up to 660 K in B12C3, raising the possibility of identifying the monoclinic symmetry in experimental measurements. A demonstration of experimentally identifiable signatures of monoclinic symmetry is provided through the computed cumulative infrared spectrum of some of the systems. The work in B13C2 stoichiometry has conclusively solved the long standing problem of the discrepancy between the DFT calculations and the experimental observations over the semiconducting nature of B13C2. The remarkable success of a newly identified 30-atomcell structure in explaining many of the experimental data on B12C3 and B13C2 provides the first definitive evidence that structures with larger unit cells, are associated with crystals of these stoichiometries even at the ground state. The work in Ruthenium Carbide stoichiometries has gathered into a coherent perspective the widely varying structures proposed from experimental reports of synthesis, computational modeling and crystal structure search and provided conclusive structural candidates to be pursued in experiments. The study of the pressure-induced variation of their stability and properties has set indicators and benchmarks for future experimental investigations. The estimation of hardness of all the systems has underlined their importance in many applications, with nearly superhard values for some of them.
Acoustic Emission Signal Based Investigations Involving Laboratory and Field Studies Related To Partial Discharges & Hot-Spots in Power Transformers
(National Institute of Technology Karnataka, Surathkal, 2017) Shanker, Tangella Bhavani; Punekar, G. S.; Nagamani, H. N.
Power transformers are important and vital components of ac power systems. It is essential to monitor the condition of these transformers periodically in order to ascertain the performance for continuous operation for its expected average life of 25-30 years. The defects in power transformers lead to the deterioration of insulation and eventual premature failure. The deterioration of insulation of power transformers can be assessed by carrying out the condition monitoring tests periodically. The condition monitoring test techniques can be off-line or on-line. The off-line test techniques are being followed as given in IEEE Std. 62(1995). These tests require outage of the transformer, thereby causing interruption of power supply. Whereas, on-line test techniques do not require any outage. Hence, on-line diagnostic techniques have gained importance. Literature review shows application of Acoustic Emission (AE) detection technique as a promising on-line tool for condition monitoring/diagnosis of the power transformers. The general guidelines for the application of AE technique for this purpose are outlined in IEEE Std. C57.127 (2007). Few typical case studies of AE signal measurements are discussed involving (i) two identical transformers, (ii) same transformer on different occasions (years) in power stations in India are reported. Some case studies with AE signals, involving On-Load Tap Changer (OLTC) and cooling system pump are also reported. These case studies also help in comprehending the efficacy of integrating the Dissolved Gas Analysis (DGA) data with the AE test results. Laboratory experimental work is carried out by simulating the most probable defects like Partial Discharge (PD) and hot-spots (leading to heat-waves) in order to capture AE signals in the range of 0-500 kHz. The classification and characterization of the defects based on the energy distribution of AE signals over the different frequency ranges is carried out using Discrete Wavelet Transform (DWT) utilizing the MATLAB toolbox. The eight-level decomposition revealed that the dominant frequency ranges for the energy distribution of the AE signals due to PD and heat-wave are 125 kHz-250 kHz and 62.5 kHz–125 kHz, respectively. The AE signal data from the transformers (field test) involving PD and hotspots are also analyzed using DWT. The laboratory based characterization of PD and heatwave got validated through the analysis of field data. The proposed method of identifying defects by AE signal analysis using DWT would complement the DGA of the transformeroil. Thus this would be a better substitute for DGA based analysis as AE based technique can be adopted in real time. The Acoustic Emission Partial Discharge (AEPD) signal parameters such as discharge magnitude and peak frequencies are studied using Fast Fourier Transform (FFT) to understand the behavior of AE signals at temperatures ranging from 30°C to 75°C. The results reported are intended to give an understanding of behavior of AEPD signals over the entire working temperature range of a transformer. At temperatures above 65°C a reduction in AEPD magnitude and peak frequencies are observed. Such behavior is noticed and probably being reported for the first time. An attempt is also made to explain the same.
Acoustic Scene Classification Using Speech Features
(National Institute of Technology Karnataka, Surathkal, 2020) Mulimani, Manjunath.; Koolagudi, Shashidhar G.
Currently, smart devices like smartphones, laptops, tablets, etc., need human intervention in the effective delivery of the services. They are capable of recognizing stuff like speech, music, images, characters and so on. To make smart systems behave as intelligent ones, we need to build a capacity in them, to understand and respond to the surrounding situation accordingly, without human intervention. Enabling the devices to sense the environment in which they are present through analysis of sound is the main objective of the Acoustic Scene Classification. The initial step in analyzing the surroundings is recognition of acoustic events present in day-to-day environment. Such acoustic events are broadly categorized into two types: monophonic and polyphonic. Monophonic acoustic events correspond to the non-overlapped events; in other words, at most one acoustic event is active in a given time. Polyphonic acoustic events correspond to the overlapped events; in other words, multiple acoustic events occur at the same time instance. In this work, we aim to develop the systems for automatic recognition of monophonic and polyphonic acoustic events along with corresponding acoustic scene. Applications of this research work include context-aware mobile devices, robots, intelligent monitoring systems, assistive technologies for hearing-aids and so on. Some of the important issues in this research area are, identifying acoustic event specific features for acoustic event characterization and recognition, optimization of the existing algorithms, developing robust mechanisms for acoustic event recognition in noisy environments, making the-state-of-the-art methods working on big data, developing a joint model that recognizes both acoustic events followed by corresponding scenes etc. Some of the existing approaches towards solutions have major limitations of using known traditional speech features, that are sensitive to noise, use of features from two-dimensional Time-Frequency Representations (TFRs) for recognizing the acoustic events, that demand high computational time;use of deep learning models, that require substantially huge amount of training data. Many novel approaches have been presented in this thesis for recognition of monophonic acoustic events, polyphonic acoustic events and scenes. Two main challenges associated with the real-time Acoustic Event Classification (AEC) are addressed in this thesis. The first one is the effective recognition of acoustic events in noisy environments, and the second one is the use of MapReduce programming model on Hadoop distributed environment to reduce computational complexity. In this thesis, the features are extracted from the spectrograms, which are robust compared to the traditional speech features. Further, an improved Convolutional Recurrent Neural Network (CRNN) and a Deep Neural Network-Driven feature learning models are proposed for Polyphonic Acoustic Event Detection (AED) in real-life recordings. Finally, binaural features are explored to train Kervolutional Recurrent Neural Network (KRNN), which recognizes both acoustic events and a respective scene of an audio signal. Detailed experimental evaluation is carried out to compare the performance of each of the proposed approaches against baseline and state-of-the-art systems.
Adaptive Distance Relay for Statcom Connected Transmission Lines - Development of Dsp Based Relay Hardware, Relaying Schemes and Hil Testing Procedures
(National Institute of Technology Karnataka, Surathkal, 2013) M.V., Sham; Panduranga Vittal, K.
Flexible AC Transmission System (FACTS) devices are used to enhance the transient stability limit and power transfer capacity of the existing transmission lines. Static Synchronous Compensator (STATCOM) a shunt type FACTS device is used to maintain the voltage at the point of common coupling on the transmission lines. A STATCOM has fast response of about 1-2 fundamental cycles, which matches with the typical response time of the protection subsystem. Hence, its functional characteristics and associated control system introduce dynamic changes during fault conditions in a transmission line. It is important that distance relays perform correctly irrespective of such dynamic changes introduced during faults, as it defeats the purpose of STATCOM installation. The work presented in this thesis is aimed at detailed study on the influence of STATCOM on the performance of distance relay under normal and abnormal operating conditions of the power systems. The work also put forth adaptive distance relaying schemes to mitigate the adverse of impact of STATCOM on distance relay. Its performance is compared with the conventional standalone mho type distance relay, through simulations on a realistic study power system using EMTDC/PSCAD package. A relay hardware to implement, the adaptive relaying scheme has been developed using TMS320F28335 digital signal processor and a simultaneous sampling ADS8556 analog to digital converter. The real time hardware in the loop test bench has been developed, using Doble F6150 power system simulator, to test the performance of the newly developed relaying schemes and relay hardware. The simulation results obtained from EMTDC/PSCAD are used as test signals for this purpose. The evaluation results have clearly demonstrated, the efficacy of the adaptive relaying schemes in mitigating the adverse impact of STATCOM on the distance relay performance.
Adaptive Resource Management in SLA Aware Elastic Clouds
(National Institute of Technology Karnataka, Surathkal, 2019) S, Anithakumari; Chandrasekaran, K.
In recent years, there has been an increasing interest in solving the over-provisioning and under-provisioning of elastic cloud resources because of the Service Level Agreement (SLA) violation problem. The recent studies have reported that federated cloud services may serve as a better elastic cloud model over a single provider model. A major problem with the federated cloud is the interoperability between multiple cloud service providers. Therefore in this thesis, a proactive SLA aware adaptive resource management approach is proposed for elastic cloud services. Aim of this thesis is to develop a suitable SLA monitoring framework to predict the SLA violations and adaptively allocate the cloud resources to improve the elasticity. It achieves the mutual benefits for cloud consumers and service providers by means of calculating and reducing penalty cost. Our framework has been implemented and validated on a private cloud using OpenNebula 4.0. The results have shown that the proposed proactive approach has significantly reduced the SLA violations compared to a reactive approach. As an additional contribution, the presented work solves the interoperability issues of the federated cloud using an innovative SLA matching algorithm. The simulation results of this work show that the said approach performs better than its counterparts.
Advanced Slope Monitoring System Todevelop Trigger Action Response Plan (Tarp) in Opencast Coal Mines Using Internet of Things (Iot)
(National Institute of Technology Karnataka, Surathkal., 2024) M., Sathish Kumar; K., Ram Chandar
In India, coal is the main energy source used to generate electricity and for other industrial purposes. Since coal-based thermal power plants account for a sizable share of India's electricity generation, the demand for coal rises drastically. India currently imports coal from other nations to meet its domestic needs. In India more than 96% coal is produced from opencast mines. Opencast (OC) mines are progressively becoming deeper to meet the increasing demand of coal. Lager and deeper opencast mines result in unstable slopes, leading to slope failures, which pose a major challenge. Slope stability and its monitoring is a serious issue in OC mines. In current scenario, conventional methods are being used for slope monitoring in opencast mine. Such monitoring typically requires a person to be physically present at the site and can only be carried out during the day. On the other hand, Slope Stability Radar (SSR) and Light Detection and Ranging (LiDAR) can monitor slope movements effectively but these are expensive, works on day time only and physical presence of persons required. In order to address this ambiguity, an advanced slope monitoring system is essential. This system should utilize low-cost sensors to monitor parameters affecting slope stability and provide early alerts by analyzing sensor data in real-time. Based on a comprehensive literature review identified, moisture content, vibrations, and displacement are the key factors contributing to slope instability. So, in this study, a slope monitoring system was designed using soil moisture, vibration and displacement sensors to detect and monitor these parameters. This system incorporates Wireless Sensor Networks (WSN) and Cloud Computing Technologies (CCT), enabling early warning alerts via email and SMS when pre-set threshold values are exceeded. The system was developed and implemented in three case study opencast mines. Additionally, a Trigger Action Response Plane (TARP) was formulated based on the rate of displacement and total displacement by the means of Wireless Sensor Networks, existing total station monitoring system, numerical modelling parametric study, that provides guidelines for actions to be taken at various levels of slope stability based on alerts from the monitoring system aided to develop a i user-friendly Advanced Slope Monitoring System (ASMS) software. This system was evaluated for its functions and performance through a laboratory experimentation on a physical slope model after the sensors were calibrated using reliable instruments. Based on the laboratory experiments, soil moisture sensors recorded a maximum of 82% and minimum of 25%. For vibration sensor, a maximum of 80 Hz and minimum of 0 Hz was detected, and 0.25 mm and 5.3 mm displacement is recorded without load and with load condition respectively. While evaluating the effect of soil moisture and vibration on slope displacement, it is identified that the moisture content in the slope has more impact on slope displacement than vibration. Laboratory investigations gave encouraging results on reliability and effectiveness of the developed system to perform field investigations in three different mines. From the field investigations, Kakatiya Khani Opencast-2 Project case study recorded the highest average rate of displacement of 2.12 mm/day, 75% moisture content and 36 Hz vibration. Khairagura Opencast Project recorded 3.27 mm/day rate of displacement, soil moisture content 78% and 28 Hz vibration, at Srirampur Opencast 2 Project rate of displacement is 3.57 mm/day with moisture 82% and 30 Hz vibration. Data collected from the mines of existing total station monitoring system and previous slope failure cases revealed the following observations, upto 50 mm displacement slopes are stable, between 50 mm to 100 mm cracks are generated and from 100 mm to 150 mm indicates potential failures are observed and above 150 mm failures observed. Later, Slope displacement obtained from Wireless Sensor Networks system of case study mines were compared with the displacement readings from the total station and numerical model of the slope that was being monitored. Results obtained at case study-1 mine for displacement through Wireless Sensor Networks system is 25.50 mm (minimum) and 46.80 mm (maximum), through total station monitoring system is 27 mm (minimum) and 49.30 mm (maximum). Similarly, the minimum and maximum displacement through numerical modeling are 29.77 mm and 46.26 mm ii respectively. The percentage of error while comparing with Wireless Sensor Networks and Total Station is below 11.47%, and WSN and NMM methods is not more than 16.75%. Hence, Wireless Sensor Networks based slope monitoring system data is very reliable. Parametric study conducted using numerical modelling studies with varying rock properties and slope geometry. A regression equation is developed for displacement and Factor of Safety (FoS). Advanced Slope Monitoring System (ASMS) software is developed based on the derived equations to track the behavior of slopes in opencast mines. Trigger Action Response Plan (TARP) has been developed based on the field investigations of case study mines. Level 1 indicates, for displacement rates below 0.3 mm/day and total slope movement under 10 mm, recommends a weekly monitoring. Level 2 indicates, rate of displacement between 0.3 mm/day to 10 mm/day and total displacement between 10 mm to 50 mm suggests weekly monitoring and slope indicates no cracks, Level 3 indicates, rates between 10 mm/day to 50 mm/day, and 50 mm to 100 mm total displacement, suggests monitoring every two days and slope indicates with crack. Level 4, rate of displacement between 50 mm/day to 100 mm/day, and total displacement between 100 mm to 150 mm recommends daily monitoring indicating potential failure. Level 5 for rate of displacement exceeding 100 mm/day and total displacement exceeding 150 mm failure takes place and suggesting clearing the area.
Advanced Spectral Spatial Approaches for Dimensionality Reduction of Hyperspectral Data
(National Institute of Technology Karnataka, Surathkal, 2024) C, DEEPA; SHETTY, AMBA; NARASIMHADHAN, A.V.
Recent advances in sensor technology have enabled the collection of large data in hyperspectral remote sensing. Although rich spectral information is captured in hundreds of narrow contiguous bands, the hyperspectral data possess several limitations such as mixed pixels, high intraclass variability, interclass similarity, and the curse of dimensionality which restricts the potential of conventional machine learning classifiers. Dimensionality reduction (DR) and incorporation of spatial information can be taken into account to increase the interpretability of hyperspectral data. The thesis mainly focuses on the implementation of different approaches for DR of hyperspectral data to address the curse of dimensionality, limited samples and labelled data issues inherent in hyperspectral data. First, a quality measure based on the co-ranking matrix has been proposed for the performance evaluation of 15 DR techniques for mineral exploration. The selection of appropriate techniques for a particular task is challenging due to the diversity and ever-increasing number of DR techniques. A few important aspects in this regard have been explored in detail. Clustering is performed using the K-means algorithm and the relationship between the quality index and clustering accuracy has been examined concurrently for the first time in hyperspectral remote sensing. Furthermore, the loss of quality in the process of DR has also been analyzed which provides sufficient input for the end-user to select an appropriate DR technique. Second, the ability of the Convolutional Neural Network (CNN) for supervised learning of hyperspectral data is explored. A fast and compact hybrid CNN which combines the strengths of 3D and 2D convolutions to extract joint spectral-spatial information has been proposed to analyze the impact of different feature extraction techniques on classification performance. The effect of input patch size on final results has been well demonstrated. A detailed investigation of classification accuracy, execution time, and comparison with nine state-of-the-art approaches has been demonstrated. ii Next, a novel deep feature selection strategy using autoencoders inspired by knowledge distillation has been implemented for the model compression and selection of informative bands. The potential of convolutional autoencoders has been well explored in selecting discriminative bands. Sensitivity analysis tests and different applications have been considered to verify the generalization capability of the proposed model. The potential of unsupervised learning schemes has been discussed in detail. Finally, a generator model based on Generative Adversarial Networks (GAN) has been proposed for virtual sample generation and compact representation of hyperspectral data. The training instability issue in Vanilla GAN has been addressed by the effective implementation of deep convolutional GANs. By comparing the spectra of the generated hyperspectral images to the corresponding real ones, the quality of the images is assessed. The potential of augmented data for improvement in classification accuracy has also been investigated.
Aerodynamic Performance of Leading Edge Protuberances at Low Reynolds Number
(National Institute Of Technology Karnataka, Surathkal, 2024) Chittepu, Jayapal Reddy; A, Sathyabhama
Wind energy plays a pivotal role in the renewable energy sector, offering sustainable and clean electricity, thus mitigating climate change by reducing greenhouse gas emissions and diversifying the energy mix away from fossil fuels. With increasing electricity demand, environmental concerns, and technological advancements, wind energy has gained prominence. The ambition for green energy heavily relies on tapping into the largely untapped potential of Small Horizontal Axia Wind Turbines (SHAWT). These turbines often face challenges such as low wind speed, small rotor size, and laminar separation, resulting in poor performance. Enhancing the aerodynamic efficiency of small wind turbine blades is crucial for increasing power and overall effectiveness. Various active and passive techniques are available to improve turbine blade aerodynamics. One passive method involves adopting leading-edge modifications inspired by Humpback Whale tubercles. These modifications, particularly effective in low Reynolds number flows, enhance aerodynamic performance by elevating the maximum lift coefficient and delaying stall. In the first section of thesis, a detailed investigation on the aerodynamic performance of two distinct airfoils, E216 and SG6043, operating at 100K Reynolds numbers is presented. The emphasis is primarily on the effect of leading-edge protuberances on the aerodynamic properties of these airfoils. Numerical simulations in ANSYS FLUENT using the SST k - 𝜔 turbulent flow model and experimental analyses in a subsonic wind tunnel using a sensitive three-component force balance were carried out. Three protuberance shapes were investigated: sinusoidal, slot, and triangular, with amplitudes (A) of 0.03c, 0.06c, and 0.11c and wavelengths (W) of 0.11c, 0.21c, and 0.43c relative to the chord length (c) of the airfoil. The variation in amplitude and wavelength combinations resulted in nine distinct models. The numerical study examined fifty six models, including baseline and protuberance models of E216 and SG6043 airfoils. Out of these, wind tunnel experiments were conducted on the baseline model as well as one model each of the three protuberance shapes to validate the numerical findings, totalling eight models for validation. The studies covered an angle of attack range of 0° to 20°. Numerical results showed that the sinusoidal protuberances caused a 2° stall delay with lower amplitudes, improving E216 A4.5W64.5 Clmax by 2.88% and (L/D)max by 6.22%. However, for SG6043, Clmax decreased by 10.21%, and (L/D)max dropped by 4.38%. Triangular protuberances also delayed stall by 2° to 4° for E216 A4.5 and A9, enhancing Clmax and (L/D)max. The E216 A4.5W64.5 model exhibited an 11.2% Clmax increase and a 14.43% rise in (L/D)max at stall angle, while SG6043 A4.5W64.5 showed a 4.37% Clmax decrease and a 1.92% (L/D)max drop at stall angle. Slot protuberances also delayed stall by 2° to 4°. The E216 airfoil demonstrated improved Clmax and stall delay, while SG6043 enhanced Clmax but reduced (L/D)max. Slot and triangular A4.5 models excelled in stall delay and post-stall performance, favoring low amplitude and high wavelength configurations. Further, the study was extended to investigate the effects of Reynolds numbers on E216 airfoil experimentally using strain gauges and data acquisition arrangement in subsonic wind tunnel facility. The Reynolds numbers considered for experimental investigation are 50K, 100K, and 150K. The study explored Reynolds number effects on protuberances, revealing minimal impact at 50K Reynolds numbers. However, at 100K and 150K Reynolds numbers, improvements were evident: enhanced post-stall lift coefficients and stall delay compared to the baseline. At 150K Reynolds number, slot, triangular, and sinusoidal protuberances showed notable increases in maximum lift coefficients by 29%, 23%, and 13%, respectively, compared to the baseline. Furthermore, this study was extended to experimentally investigate the aerodynamic behaviour of the E216 airfoil with protuberances under dynamic conditions. This investigation made use of a stepper motor, strain gauges, and a data acquisition system in a subsonic wind tunnel facility. The reduced frequencies (k) considered in this study are 0.025, 0.05, and 0.065. In essence, introducing protuberances to the airfoil not only influences lift coefficient patterns but also affects the hysteresis loop size during dynamic stall. Moreover, protuberance models exhibited smoother post-stall behavior compared to the baseline. Overall, slot and triangular protuberances notably enhanced Clmax compared to both the baseline and sinusoidal protuberance models. The smoke and tuft flow visualization techniques revealed important insights into flow patterns, assisting in understanding the flow physics. The smoke flow study observed increased post-stall lift with increasing angles of attack due to the merging of secondary flow with primary flow. While sinusoidal and triangular profiles exhibited similar behavior, triangular leading edges effectively guided flow to trough regions, resulting in a larger secondary flow volume. Triangular protuberances proved more effective for drag reduction on the SG6043 airfoil, particularly at a Reynolds number of 100K, and contributed to post-stall lift improvement. According to the tuft flow visualization, separation of flow was observed at different distances from the leading edge in protuberance models, around 22% for the sinusoidal model and 24% for the triangular and slot configurations. Significantly, both the slot and triangular protuberance models showcased improved attached flow within the troughs, leading to elevated lift coefficients and stall delay. The investigation was extended further to examine the aerodynamic performance of Small Horizontal Axis Wind Turbines (SHAWT) with protuberance blades, comparing them with the baseline blades. The aerodynamic performance of wind turbines featuring various protuberance blade models – sinusoidal, triangular, and slot designs, were tested and contrasted with the baseline. Based on the results, the enhancement in power coefficient (CP) for protuberance configurations is as follows: 2.7 % for sinusoidal, 5.2 % for triangular, and 7.6 % for slot blades compared to baseline blades. The slot protuberances blade model displayed the highest CP value among the three shapes. This suggests that wind turbines incorporating slot, triangular, and sinusoidal protuberances achieved improved aerodynamic efficiency compared to baseline blades.
Aerodynamic Performance of Low Aspect Ratio Turbine Blade in the Presence of Purge Flow
(National Institute of Technology Karnataka, Surathkal, 2021) Babu, Sushanlal.; S, Anish.
In aero engines, purge flow is directly fed from the compressor which bypasses the combustion chamber and introduced into the disk space between blade rows to prevent the hot ingress. Higher quantity of purge gas fed through the disk space can provide additional thermal protection to passage endwall and blade surfaces. Moreover interaction of the purge air with the mainstream flow can alter the flow characteristics of turbine blade passage. The objective of the present investigation is to understand the secondary vortices and its aerodynamic behavior within a low aspect ratio turbine blade passage in the presence of purge flow. An attempt is made to understand the influence of velocity ratios and purge ejection angles on these secondary vortices. The objective is broadened by investigating the unsteadiness generated by upstream wakes over the secondary vortex formations in th presence of purge flow. Further the thesis aims to judge the feasibility of implementing endwall contouring to curb the additional losses generated by the purge flow. To accomplish these objectives, a combination of experimental measurements and computational simulations are executed on a common blade geometry. The most reliable commercial software ANSYS CFX which solves three dimensional Reynolds Averaged Navier Stokes Equations together with Shear Stress Transport (SST) turbulence model has been used to carry out computational simulations. Along with steady state analysis, in order to reveal the time dependent nature of the flow variables, transient analysis has been conducted for certain selected computational domains. The numerical results are validated with experimental measurements obtained at the blade exit region using five hole probe and Scanivalve. The experimental analysis is conducted for the base case without purge (BC) and base case with purge (BCp) configurations having flat endwalls. vi In the present analysis, it is observed that with an increase in the velocity ratio, the mass averaged total pressure losses also increases. In an effort to reduce the losses, purge ejection angle is reduced to 350 from 900 with a step size of 150. Significant loss reduction and improved endwall protection are observed at lower ejection angles. Numerical investigation of upstream disturbances/wakes explore the interaction effects of two additional vortices, viz. the cylinder vortex (Vc) and the purge vortex (Vp). Steady state analysis shows an increase in the underturning at blade exit due to the squeezing of the pressure side leg of horseshoe vortex (PSL) towards the pressure surface by the cylinder vortices (Vp). The unsteady analysis reveals the formation of filament shaped wake structures which breaks into smaller vortical structures at the blade leading edge for stagnation wake configuration (STW). On the contrary, in midpassage wake configuration (MW), the obstruction created by the purge flow causes the upper portion of cylinder vortices bend forward, creating a shearing action along the spanwise direction. Investigation of contoured endwall geometries shows that, endwall curvature either accelerate or decelerate the flow thereby a control over the endwall static pressure can be obtained. Out of three contoured endwalls investigated, the stagnation zones generated at the contour valleys has resulted in the additional loss generation for the first two profiles. Reduced valley depth and optimum hump height of the third configuration has effectively redistributed the endwall static pressure. Moreover an increase in the static pressure distribution at the endwall near to pressure surface has eliminated the pressure side bubble formation. Computational results of URANS (Unsteady Reynolds Averaged Navier Stokes) simulations are obtained for analyzing transient behaviour of pressure side bubble, with more emphasis on its migration on pressure surface and across the blade passage.
Ai-Based Clinical Decision Support Systems Using Multimodal Healthcare Data
(National Institute of Technology Karnataka, Surathkal, 2022) Mayya, Veena; S, Sowmya Kamath
Healthcare analytics is a branch of data science that examines underlying patterns in healthcare data in order to identify ways in which clinical care can be improved – in terms of patient care, cost optimization, and hospital management. Towards this end, Clinical Decision Support Systems (CDSS) have received extensive re- search attention over the years. CDSS are intended to influence clinical decision making during patient care. CDSS can be defined as “a link between health obser- vations and health-related knowledge that influences treatment choices by clinicians for improved healthcare delivery”.A CDSS is intended to aid physicians and other health care professionals with clinical decision-making tasks based on automated analysis of patient data and other sources of information. CDSS is an evolving system with the potential for wide applicability to improve patient outcomes and healthcare resource utilization. Recent breakthroughs in healthcare analytics have seen an emerging trend in the application of artificial intelligence approaches to assist essential applications such as disease prediction, disease code assignment, disease phenotyping, and disease-related lesion segmentation. Despite the signifi- cant benefits offered by CDSSs, there are several issues that need to be overcome to achieve their full potential. There is substantial scope for improvement in terms of patient data modelling methodologies and prediction models, particularly for unstructured clinical data. This thesis discusses several approaches for developing decision support sys- tems towards patient-centric predictive analytics on large multimodal healthcare data. Clinical data in the form of unstructured text, which is rich in patient- specific information sources, has largely remained unexplored and could be poten- tially used to facilitate effective CDSS development. Effective code assignment for patient clinical records in a hospital plays a significant role in the process of stan- dardizing medical records, mainly for streamlining clinical care delivery, billing, and managing insurance claims. The current practice employed is manual cod- ing, usually carried out by trained medical coders, making the process subjective, error-prone, inexact, and time-consuming. To alleviate this cost-intensive pro- iii cess, intelligent coding systems built on patients’ unstructured electronic medical records (EMR) are critical. Towards this, various deep learning models have been proposed for improving the diagnostic coding system performance that makes use of patient clinical reports and discharge summaries. The approach involved multi channel convolution networks and label attention transformer architectures for au- tomatic assignment of diagnostic codes. The label attention mechanism enabled the direct extraction of textual evidence in medical documents that mapped to the diagnostic codes. Medical imaging data like ultrasound, magnetic resonance imaging, computed tomography, positron emission tomography, X-ray, retinal photography, slit lamp microscopy, etc., play an important role in the early detection, diagnosis, and treatment of diseases. Presently, most imaging modalities are manually inter- preted by expert clinicians for disease diagnosis. With the exponential increase in the volume of chronic patients, this process of manual inspection and interpre- tation increases the cognitive and diagnostic burden on healthcare professionals. Recently, machine learning and deep learning techniques have been utilized for designing computer based analysis systems for medical images. Ophthalmology, pathology, radiology, and oncology are a few fields where deep learning techniques have been successfully leveraged for interpreting imaging data. Ophthalmology was the first field to be revolutionized and most explored in health care. To- wards this, various deep learning models have been proposed for improving the performance of ocular disease detection systems that make use of fundoscopy and slit-lamp microscopy imaging data. Patient data is recorded in multiple formats, including unstructured clinical notes, structured EHRs, and diagnostic images, resulting in multimodal data that together accounts for patients’ demographic information, past histories of illness and medical procedures performed, diseases diagnosed, etc. Most existing works limit their models to a single modality of data, like structured textual, unstruc- tured textual, or imaging medical data, and very few works have utilized multi- modal medical data. To address this, various deep learning models were designed that can learn disease representations from multimodal patient data for early dis- ease prediction. Scalability is ensured by incorporating content based learning models for automatically generating diagnosis reports of identified lung diseases, reducing radiologists’ cognitive burden.
Algorithms for Color Normalization and Segmentation of Liver Cancer Histopathology Images
(National Institute of Technology Karnataka, Surathkal, 2021) Roy, Santanu.; Lal, Shyam.
With the advent of Computer Assisted Diagnosis (CAD), accuracy of cancer detection from histopathology images is significantly increased. However, color variation in CAD system is inevitable due to variability of stain concentration and manual tissue sectioning. Small variation in color may lead to misclassification of cancer cells. Therefore, color normalization is the first step of Computer Assisted Diagnosis (CAD), in order to reduce the inter-variability of background color among a set of source images. In this thesis, first a novel color normalization method is proposed for Hematoxylin and Eosin (H and E) stained histopathology images. Conventional Reinhard algorithm is modified in our proposed method by incorporating fuzzy logic. Moreover, mathematically it is proved that our proposed method satisfies all three hypotheses of color normalization. Furthermore, several quality metrics are estimated locally for evaluating the performance of various color normalization methods. Experimental result reveals that our proposed method has outperformed all other benchmark methods. The second step of CAD is nuclei segmentation which is the most significant step since it enables the classification task computationally efficient and simple. However, automatic nuclei detection is fraught with problems due to highly textured nuclei boundary and various size and shapes of nuclei present in histopathology images. In this thesis, a novel edge detection technique is proposed for segmenting the nuclei regions in liver cancer Hematoxylin and Eosin (H and E) stained histopathology images, based on the notion of computing local standard deviation value. Moreover, the edge-detected image is converted into a binary image by using local Otsu thresholding and thereafter, it is refined by an adaptive morphological filter. The experimental result indicates that proposed segmentation method overcomes the limitations of existing unsupervised methods and subsequently its performance is also comparable with deep neural models. To the best of our knowledge, our proposed method is the only unsupervised method iii which achieves nuclei detection accuracy closest to 1 (0.9516). Furthermore, two more quality metrics are computed in order to measure the performance of nuclei segmentation methods quantitatively. The mean value of quality metrics reveals that our proposed segmentation method outperforms other existing methods both qualitatively and quantitatively.