Search Results (1 - 25 of 80 Results)

Sort By  
Sort Dir
 
Results per page  

Akkala, ArjunDevelopment of Artificial Neural Networks Based Interpolation Techniques for the Modeling and Estimation of Radon Concentrations in Ohio
Master of Science, University of Toledo, 2010, Engineering (Computer Science)

Radon is a chemically inert, naturally occurring radioactive gas. It is one of the main causes of lung cancer second to smoking, and accounts for about 25,000 deaths every year in the US alone according to the National Cancer Institute. In order to initiate preventative measures to reduce the deaths caused by radon inhalation, it is helpful to have radon concentration data for each locality, e.g. zip code. However, such data are not available for every zip code in Ohio, owing to several reasons including inapproachability. In places where data is unavailable, radon concentrations must be estimated using interpolation techniques to take appropriate preventive measures against cancer.

This thesis proposes new interpolation techniques based on Artificial Neural Networks utilizing the available knowledge in terms of Radon concentration data and Uranium concentration data for modeling and predicting Radon concentrations in Ohio, US. Several models were first trained and then validated using available data to identify the best model for each technique. Model accuracies using the proposed approaches were proven to be significantly better in comparison to conventional interpolation techniques such as Kriging and Radial Basis Functions.

Committee:

Vijay Devabhaktuni, PhD (Advisor); Ashok Kumar, PhD (Advisor); Mohammed Niamat, PhD (Committee Member)

Subjects:

Environmental Engineering

Keywords:

Artificial neural networks; Interpolation; Modeling; Ohio; Zip code; Radon; Uranium; Knowledge Based Neural Network; Source Difference Method; Prior Knowledge Input, Space Mapped Neural Network

Paheding, SidikeProgressively Expanded Neural Network for Automatic Material Identification in Hyperspectral Imagery
Doctor of Philosophy (Ph.D.), University of Dayton, 2016, Electrical and Computer Engineering
The science of hyperspectral remote sensing focuses on the exploitation of the spectral signatures of various materials to enhance capabilities including object detection, recognition, and material characterization. Hyperspectral imagery (HSI) has been extensively used for object detection and identification applications since it provides plenty of spectral information to uniquely identify materials by their reflectance spectra. HSI-based object detection algorithms can be generally classified into stochastic and deterministic approaches. Deterministic approaches are comparatively simple to apply since it is usually based on direct spectral similarity such as spectral angles or spectral correlation. In contrast, stochastic algorithms require statistical modeling and estimation for target class and non-target class. Over the decades, many single class object detection methods have been proposed in the literature, however, deterministic multiclass object detection in HSI has not been explored. In this work, we propose a deterministic multiclass object detection scheme, named class-associative spectral fringe-adjusted joint transform correlation. Human brain is capable of simultaneously processing high volumes of multi-modal data received every second of the day. In contrast, a machine sees input data simply as random binary numbers. Although machines are computationally efficient, they are inferior when comes to data abstraction and interpretation. Thus, mimicking the learning strength of human brain has been current trend in artificial intelligence. In this work, we present a biological inspired neural network, named progressively expanded neural network (PEN Net), based on nonlinear transformation of input neurons to a feature space for better pattern differentiation. In PEN Net, discrete fixed excitations are disassembled and scattered in the feature space as a nonlinear line. Each disassembled element on the line corresponds to a pattern with similar features. Unlike the conventional neural network where hidden neurons need to be iteratively adjusted to achieve better accuracy, our proposed PEN Net does not require hidden neurons tuning which achieves better computational efficiency, and it has also shown superior performance in HSI classification tasks compared to the state-of-the-arts. Spectral-spatial features based HSI classification framework has shown stronger strength compared to spectral-only based methods. In our lastly proposed technique, PEN Net is incorporated with multiscale spatial features (i.e., multiscale complete local binary pattern) to perform a spectral-spatial classification of HSI. Several experiments demonstrate excellent performance of our proposed technique compared to the more recent developed approaches.

Committee:

Vijayan Asari (Advisor); Raul Ordonez (Committee Member); Eric Balster (Committee Member); Muhammad Islam (Committee Member)

Subjects:

Computer Engineering; Electrical Engineering; Remote Sensing

Keywords:

Hyperspectral imagery; neural network; object detection; classification; joint transform correlation; progressively expanded neural network; spectral-spatial features;

Jagirdar, SureshInvestigation into Regression Analysis of Multivariate Additional Value and Missing Value Data Models Using Artificial Neural Networks and Imputation Techniques
Master of Science (MS), Ohio University, 2008, Industrial and Systems Engineering (Engineering and Technology)

Missing data or insufficient data is a major concern in statistical analysis, especially when the problem to be addressed is related to prediction. The present study is a quantitative analysis of data based on a 'regression technique' for the modeling and prediction of three different numerical data sets, named: Boston Housing, Saginaw Bay,and Reportable Outages. The goal is to enhance the regression results by using an artificial neural network to either impute missing data or supplement existing data using non-linear model. The present prediction problem focuses on developing a hybrid model to create additional data using a combination of model selection techniques such as cross validation estimate and regularization theory in order to reduce the effect of over fitting. The student model is constructed using the knowledge extracted from the teacher model. The knowledge is in the form of learning obtained through an Artificial Neural Network training of the data set/s. The learning derived from the training is also used to build a reverse engineered model to address the missing value data problem.

In the case of Reportable Outages, the student model achieved is the closest nonlinear model to the teacher model and succeeded in enhancing the regression analysis for the prediction problem. The Boston Housing student model contains a significant amount of correlation among the variables, which need additional data relational techniques to address the correlation. The enhancement of linear regression for the case of prediction for Saginaw Bay data is limited to nature and its complex processes.

The overall results achieved are encouraging and show promise for developing a model to create a needed data when data is highly correlated. More data cases need to be investigated using the reverse engineering technique of the Artificial Neural Network for predicting missing value data.

Committee:

Gary R. Weckman (Committee Chair); David Koonce (Committee Member); Scherwa Diana (Committee Member); Andrew Snow (Committee Member)

Subjects:

Industrial Engineering

Keywords:

Artificial Neural Network; Data Addition and Knowledge Extraction; Reverse Engineered Artificial Neural Network; Correlation Breakdown

Kamalasadan, SukumarA New Generation of Adaptive Control: An Intelligent Supervisory Loop Approach
Doctor of Philosophy in Engineering, University of Toledo, 2004, Electrical Engineering
A new class of intelligent adaptive control for systems with complex and multimodal dynamics including scheduled and unscheduled ‘Jumps’, is developed. Those systems are often under the challenge of unforeseen changes due to wide range of operations and/or external influences. The underlying structural feature is an introduction of an Intelligent Supervisory Loop (ISL) to augment the Model Reference Adaptive Control (MRAC) framework. Four novel design formulations are developed which evolve from different methods of conceiving ISL, structured into intelligent control algorithms, and then investigated with comprehensive simulation models of a single link flexible robotic manipulator as well as a six degree of freedom F16 fighter aircraft. The first scheme is a Fuzzy Multiple Reference Model Adaptive Controller (FMRMAC). It consists of a fuzzy logic switching strategy introduced to the MRAC framework. The second is a novel Neural Network Parallel Adaptive Controller (NNPAC) for systems with unmodeled dynamics and mode swings. It consists of an online growing dynamic radial basis neural network, which controls the plant in parallel with a direct MRAC. The third scheme is a novel Neural Network Parallel Fuzzy Adaptive Controller (NNPFAC) for dynamic ‘Jump’ systems showing scheduled mode switching and unmodeled dynamics. The scheme consists of a growing online dynamic Neural Network (NN) controller in parallel with a direct MRAC, and a fuzzy multiple reference model generator. The fourth scheme is a Composite Parallel Multiple Reference Model Adaptive Controller (CPMRMAC) for systems showing unscheduled mode switching and unmodeled dynamics. The scheme consists of an online growing dynamic NN controller in parallel with a direct MRAC, and an NN multiple reference model generator. Extensive feasibility simulation studies and investigations have been conducted on the four proposed schemes, and with results consistently showing that the four design formulations developed in this research, for implementing intelligent supervisory loops into the MRAC framework, are feasible, effective and have immense potential for complex systems control. Even though those two systems are specific in nature, they are true representatives of an important and challenging class of dynamic systems that require the new generation of adaptive controllers developed in this project work.

Committee:

Adel Ghandakly (Advisor)

Keywords:

Intelligent Adaptive Control; Intelligent Supervisory Loop Approach; Fuzzy Multiple Reference Model Adaptive Control; Neural Network Parallel Adaptive Control; Neural Network Parallel Fuzzy Adaptive Control

Kramer, Gregory RobertAn analysis of neutral drift's effect on the evolution of a CTRNN locomotion controller with noisy fitness evaluation
Doctor of Philosophy (PhD), Wright State University, 2007, Computer Science and Engineering PhD
This dissertation focuses on the evolution of Continuous Time Recurrent Neural Networks (CTRNNs) as controllers for control systems. Existing research suggests that the process of neutral drift can greatly benefit evolution for problems whose fitness landscapes contain large-scale neutral networks. CTRNNs are known to be highly degenerate, providing a possible source of large-scale landscape neutrality, and existing research suggests that neutral drift benefits the evolution of simple CTRNNs. However, there has been no in-depth examination of the effects of neutral drift on complex CTRNN controllers, especially in the presence of noisy fitness evaluation. To address this problem, this dissertation presents an analysis of the effect of neutral drift on the evolution of a complex CTRNN locomotion controller for a simulated hexapod robot in the presence of noisy fitness evaluations. In particular, two stochastic hill-climber-based EAs are examined and compared, one that does not engage in neutral drift, and one that does. The experimental results show that while neutral drift provides a significant advantage early in the evolutionary process, the later effects of noisy fitness evaluations seriously degrades the utility of neutral drift, and overall, there is no significant difference between the non-drifting and drifting EAs. These results provide evidence that large-scale neutral networks do exist in complex CTRNN fitness landscapes and highlight the important role that noisy fitness evaluations play in influencing evolutionary performance.

Committee:

John Gallagher (Advisor)

Subjects:

Computer Science

Keywords:

Continuous Time Recurrent Neural Network; Neutrality; Neutral Drift; Artificial Evolution; Evolutionary Computation; Neural Network; Controller; Evolutionary Algorithm; Noisy Fitness Evaluation

Borundiya, Amit ParasmalImplementation of Hopfield Neural Network Using Double Gate MOSFET
Master of Science (MS), Ohio University, 2008, Electrical Engineering & Computer Science (Engineering and Technology)
Hopfield Neural Network has been used to solve the constraints satisfaction problems. To make these networks solve problem in real time, independent of the size, would require building a massively parallel structure. A CMOS circuit can be used to construct such network to find the solution. Current CMOS technology is reaching its physical limitation in deep submicron regime and new devices are explored which can provide scalability in accordance to Moore's; Law. To further increase the network capacity double gate transistors can be used. Double gate MOSFET model of the hysteresis neuron proposed in this thesis utilizes 8 transistors as compared to 60 transistors needed with an operational amplifier's; model. This structure not only reduces the count of transistors by 86% but also demonstrates that larger circuits of double gate MOSFETs can be built, bolstering the faith in double gate MOSFET devices as a possible substitute of CMOS devices in a near future.

Committee:

Janusz A. Starzyk, PhD (Committee Chair); Savas Kaya, PhD (Committee Member); Jeffery Dill, PhD (Committee Member); Xiaoping Shen, PhD (Committee Member)

Subjects:

Electrical Engineering

Keywords:

Hysteresis Hopfield Neural Network using double gate MOSFET; double gate mosfet neural network; N-Queen with double gate MOSFET

Chen, SongqingDevelopment of a neural network based software package for the automatic recognition of license plate characters
Master of Science (MS), Ohio University, 1992, Industrial and Manufacturing Systems Engineering (Engineering)

This research studies the techniques being used in character recognition. Software has been developed to automatically recognize license plate characters. The program is written in Borland C++ 2.0, and runs on Microsoft Windows 3.0. The system consists of two main parts: the preprocessor and the neural network. The preprocessor separates all characters from a picture. The neural network has a three-layer architecture using a back-propagation training method. Instead of using the derivative of the sigmoid transfer function, a differential step-size function is applied to the output neurons to solve local minima problems in training. The neural network recognizes the characters from the preprocessor.

To run the system, an IBM 286 PC or compatible with at least 1M memory is required. An image digitizer system CapCalc was used to generate an image file in 512×480 format with 0-255 gray levels. It recognizes both white-on-black and black-on-white pictures. The pictures of the license plates should be fairly clear and complete.

The neural network was trained with 145 characters from 24 license plate pictures of different characters and fonts. The pictures included 15 black-on- white and 9 white-on-black. The system was tested with additional 35 license plate pictures. It was able to recognize about half of the characters on the 5 pictures which met the requirements. On a 386 PS/2 model 70, it took about 16 seconds to recognize one picture. The system demonstrated the feasibility of automatically recognizing license plate characters using image-processing techniques and a neural network. This was an exploratory lab study, and was not intended to be a realistic model of a biological vision system.

Committee:

Helmut Zwahlen (Advisor)

Subjects:

Engineering, Industrial

Keywords:

Neural network based software; Automatic recognition; License plate characters

Pappada, Scott MichaelPrediction of Glucose for Enhancement of Treatment and Outcome: A Neural Network Model Approach
Doctor of Philosophy in Engineering, University of Toledo, 2010, Bioengineering

Critical care (e.g. trauma and cardiothoracic surgical) and diabetic patients are prone to variability in glucose concentration on a daily basis. Hypoglycemic and hyperglycemic glucose values in these patient populations have been associated with decreased patient outcomes. In diabetic patients, persistently elevated glucose values are associated with development of long term complications such as, but not limited to retinopathy, neuropathy, and nephropathy. In the critical care patient population, elevated glucose has been correlated to increases in mortality, length of stay in the intensive care unit (ICU), and morbidities. The maintenance of tight glycemic control in these patients without severe hypoglycemia or glycemic variability appears to improve outcomes in these patients.

Various factors are associated with future glycemic excursions such as, but not limited to: lifestyle/activities (e.g. sleep-wake cycles), emotional factors (e.g. stress), nutritional intake, medication dosages, and ICU medical records (in critical care patients). In the field of diabetes research, models for prediction of glucose and/or models used to maintain tight glycemic control have been the focus of research. In the critical care patient population, very little research into development of such models has been completed to date.

Multiple factors affect or are indicators of future glucose concentration. A suitable modeling technique needs to incorporate the effect of such factors for accurate prediction of glucose. A modeling technique well suited for this task is a neural network model.A neural network is an adapative modeling technique, which learns and updates model parameters based on determining patterns/trends existent in input data.This adapative capability, makes neural network modeling well suited for prediction of glucose where multiple factors impact future glycemic excursions.

This dissertation summarizes the development and optimization of various neural network model architectures for the real-time prediction of glucose in diabetic and critical care patients. Neural network models were configured to predict glucose using prediction horizons >60 minutes, which have not been attained in many predictive models to date. The performance of the neural network model is assessed via determination of overall model error, percentage of glycemic extremes predicted, and clinical acceptability of model predictions as determined via Clarke Error Grid Analysis.

Committee:

Brent Cameron, PhD (Advisor); Ronald Fournier, PhD (Committee Member); Thomas Papadimos, MD (Committee Member); Marilyn Borst, MD (Committee Member); William Olorunto, MD (Committee Member)

Subjects:

Bioinformatics; Computer Science; Engineering

Keywords:

prediction of glucose; diabetes; critical care; neural network; closed loop insulin delivery; glucose; real-time prediction;

Li, ZheA Neural Network Based Distributed Intrusion Detection System on Cloud Platform
Master of Science in Engineering, University of Toledo, 2013, Engineering (Computer Science)
Intrusion detection system (IDS) is an important component to maintain network security. Also, as the cloud platform is quickly evolving and becoming more popular in our everyday life, it is useful and necessary to build an effective IDS for the cloud. However, existing intrusion detection techniques will be likely to face challenges when deployed on the cloud platform. The pre-determined IDS architecture may lead to overloading of a part of the cloud due to the extra detection overhead. This thesis proposes a neural network based IDS, which is a distributed system with an adaptive architecture, so as to make full use of the available resources without overloading any single machine in the cloud. Moreover, with the machine learning ability from the neural network, the proposed IDS can detect new types of attacks with fairly accurate results. Evaluation of the proposed IDS with the KDD dataset on a physical cloud testbed shows that it is a promising approach to detecting attacks in the cloud infrastructure.

Committee:

Lingfeng Wang (Committee Chair); Weiqing Sun (Committee Co-Chair); Richard Molyet (Committee Member)

Subjects:

Computer Engineering; Computer Science

Keywords:

Distributed IDS; Neural network; Cloud security; Anomaly detection

Horton, Jennifer LeighPush Recovery: A Machine Learning Approach to Reactive Stepping
Master of Science, The Ohio State University, 2013, Electrical and Computer Engineering
When robots are integrated into the real world, chances are they will not be able to completely avoid situations in which they are bumped or pushed unexpectedly. In these situations, the robot could potentially damage itself, damage its surroundings, or fail to perform its tasking unless it is able to take active countermeasures to prevent or recover from falling. One such countermeasure, referred to as reactive stepping, involves a robot taking a series of steps in order to regain balance and recover from a push. Research into reactive stepping typically focuses on choosing which step to take. This thesis proposes a machine learning approach to reactive stepping. This approach leverages neural networks to calculate a series of steps that return the robot to a stable position. It was theorized that the robot would become stable if it always chose the step resulting in the highest reduction of energy. Theories were tested using a compass model that incorporated parameters and constraints realistic of an actual humanoid robot. The machine learning approach using neural networks performed favorably in both computation time and push recovery effectiveness when compared with the linear least squares, nearest interpolation, and linear interpolation methods. Results showed that when using neural networks to calculate the best step for an arbitrary push within the defined range, the compass model was able to successfully recover from 97% of the pushes applied. The procedure was kept very general and could be used to implement reactive stepping on physical robots, or other robot models.

Committee:

Yuan Zheng (Advisor); David Orin (Committee Member)

Subjects:

Computer Science; Engineering; Robotics

Keywords:

Push Recovery; Reactive Stepping; Machine Learning; Neural Network; Bipedal Robot; Compass Model; Robot

Parks, Brandon ScottSearch for the Higgs Boson in the ZHvvbb̄ Channel at CDF Run II
Doctor of Philosophy, The Ohio State University, 2008, Physics
This analysis focuses on a low mass Higgs boson search with 1.7 fb-1 of data. The focus is on Higgs events in which it is produced in association with a W or Z boson. Such events are expected to leave a distinct signature of large missing transverse energy for either a Z → vv decay or a leptonic W decay in which the lepton goes undetected, as well as jets with taggable secondary vertices from the H → bb̄ decay. Utilizing a new track based technique for removing QCD multi-jet processes as well as a neural network discriminant, an expected limit of 8.3 times the Standard Model prediction at the 95% CL for a Higgs boson mass of 115 GeV/c2 is calculated, with an observed limit of 8.0*SM.

Committee:

Brian Winer, PhD (Advisor); Richard Hughes, PhD (Committee Member); Junko Shigemitsu, PhD (Committee Member); Linn Van Woerkom, PhD (Committee Member); Andrew Gould, PhD (Committee Member)

Subjects:

Physics

Keywords:

Higgs; Tevatron; CDF; Fermilab; Associated Production; Neural Network

Park, Gwang HoonHandwritten digit and script recognition using density based random vector functional link network
Doctor of Philosophy, Case Western Reserve University, 1995, Electrical Engineering
A new formation of a neural network called a Density Based Random Vector Functional Link Network (DBRVFLN) is introduced to solve high dimensional real-world problems. It is a hybrid technique which uses the combination of a priori knowledge of the problem and randomness to prepare unknown factors. Simple but powerful feature extraction methods for handwritten digit recognition and script recognition are introduced. Handwritten digit recognition systems using neural networks are introduced. For the script recognition task, a global approach which uses whole sets of features of the image and an analytical approach from nonsegmented sequence of features via letters to a word using neural networks are designed and explored. The recognition systems are based on conventional preprocessing methodologies, novel feature extraction and reduction algorithms and quadratic approaches of neural networks such as Radial Basis Function Neural Network(RBNN) and DBRVFLN. The recognition systems are tested using unconstrained real-world databases. In the handwritten digit recognition task, the performance of the recognition system using DBRVFLN is better than that of RBNN if there is enough priori knowledge. To attain 1% substitution error rates, the current recognition system needs to tolerate rejection rat es of about 11%∼12%. In the global approach of script recognition task, the ability to construct a filter for one word is tested. While keeping a very low substitution error rate of 0.74%, the recognition system which uses DBRVFLN rejects 11.48% and has better performance with 2.78% in the rejection ratio than the system using RBNN even if they have the same number of enhancement nodes. The experimental results show that the random vector enhancements for the unknown factor in DBRVFLN act very nicely as decision enhancers and that they do improve the classification performances in comparison with RBNN, in the same recognition system

Committee:

Yoh-Han Pao (Advisor)

Keywords:

Density based random vector functional link network (DBRVFLN); Neural network

Neu, Christopher C.A Search for the Higgs Boson in proton - antiproton collisions at center-of-mass energy of 1.8 TeV
Doctor of Philosophy, The Ohio State University, 2003, Physics

Although the Standard Model of fundamental particles and their interactions has enjoyed much success over the past quarter-century, a portion of the theory has eluded experimental verification. Mass is a manifest quality of the constituents of our universe; it remains to be understood however in the context of the Standard Model why the fundamental particles have the masses they possess. The imposition of mass is a consequence of the breaking of the symmetry that unifies the weak and electromagnetic forces; electroweak symmetry breaking is accomplished in the Standard Model via the Higgs mechanism. This elegant portion of the theory not only provides the dynamics for the symmetry breaking, but also predicts a physically observable scalar particle, the Higgs boson, which is yet to be discovered.

A new search for the Higgs boson has been performed in the proton – antiproton collisions at center-of-mass energy of 1.8 TeV provided by the Tevatron accelerator at Fermi National Accelerator Laboratory. In this analysis, a neural network was utilized to aid in the rejection of collision events that share the equivalent signature as Higgs events but are produced via other, less interesting production mechanisms. The neural network was implemented as part of an advanced event selection that in simulation studies was shown to provide a 34% increase in signal sensitivity over conventional methods. When the technique is applied to the data collected by the CDF collaboration during the Tevatron’s Run 1 (1992—1995), an excess of events is identified above the background expectation. The limit on the Higgs production cross section is calculated for six Higgs mass hypotheses in the range 100 GeV/c2 < MH < 150 GeV/c2. The WH production cross section upper limit was determined to be 18—22 pb in the range MH < 130 GeV/c2 at 95% confidence; the limit in the range MH > 130 GeV/c2 is considerably larger. The measured limit is a factor of two larger than the limit from a priori studies. The Standard Model theory prediction is approximately two orders of magnitude lower than this upper limit in the MH range.

Committee:

Brian Winer (Advisor); Richard Hughes (Other); Eric Braaten (Other); Linn Van Woerkem (Other)

Keywords:

Higgs; Higgs boson; Fermilab; Tevatron; Electroweak symmetry breaking; Origin of mass; Neural Network

Foltz, John WendellThe Relationships Between Microstructure, Tensile Properties and Fatigue Life in Ti-5Al-5V-5Mo-3Cr-0.4Fe (Ti-5553)
Doctor of Philosophy, The Ohio State University, 2010, Materials Science and Engineering

β-titanium alloys are being increasingly used in airframes as a way to decrease the weight of the aircraft. As a result of this movement, Ti-5Al-5V-5Mo-3Cr-0.4Fe (Timetal 555), a high-strength β titanium alloy, is being used on the current generation of landing gear. This alloy features good combinations of strength, ductility, toughness and fatigue life in α+β processed conditions, but little is known about β-processed conditions. Recent work by the Center for the Accelerated Maturation of Materials (CAMM) research group at The Ohio State University has improved the tensile property knowledge base for β-processed conditions in this alloy, and this thesis augments the aforementioned development with description of how microstructure affects fatigue life.

In this work, β-processed microstructures have been produced in a GleebleTM thermomechanical simulator and subsequently characterized with a combination of electron and optical microscopy techniques. Four-point bending fatigue tests have been carried out on the material to characterize fatigue life. All the microstructural conditions have been fatigue tested with the maximum test stress equal to 90% of the measured yield strength. The subsequent results from tensile tests, fatigue tests, and microstructural quantification have been analyzed using Bayesian neural networks in an attempt to predict fatigue life using microstructural and tensile inputs. Good correlation has been developed between lifetime predictions and experimental results using microstructure and tensile inputs. Trained Bayesian neural networks have also been used in a predictive fashion to explore functional dependencies between these inputs and fatigue life.

In this work, one section discusses the thermal treatments that led to the observed microstructures, and the possible sequence of precipitation that led to these microstructures. The thesis then describes the implications of microstructure on fatigue life and implications of tensile properties on fatigue life. Several additional experiments are then described that highlight possible causes for the observed dependence of microstructure on fatigue life, including fractographic evidence to provide support of microstructural dependencies.

Committee:

James Williams (Advisor); Hamish Fraser (Committee Member); Katharine Flores (Committee Member)

Subjects:

Materials Science

Keywords:

titanium; fatigue; beta alloy; microstructure; neural network; fracture; tensile properties; bayesian;

Siegel, DavidPrognostics and Health Assessment of a Multi-Regime System using a Residual Clustering Health Monitoring Approach
PhD, University of Cincinnati, 2013, Engineering and Applied Science: Mechanical Engineering
Monitoring the health condition of machinery has been an area of research for quite some time. Despites several advancements, the application of conventional signal analysis and pattern recognition methods face several challenges when the operating variables such as load, speed, and temperature vary considerably for the monitored asset. The residual clustering approach addresses the multi-regime monitoring challenge by first modeling the baseline non-linear correlation relationship in the measured signal features and by providing predicted signal features. Calculating the residual signal features allows one to normalize the effect of the operating variables, since one is considering how the response of the system compares with the predicted response based on the baseline behavior. In many instances the degradation signature of a component or system is more pronounced under certain operating conditions. The clustering portion of the residual clustering method specifically addresses the regime dependent signature aspect and bases the health value on the monitoring regime in which the degradation signature is more prevalent. This dissertation work highlights the mathematical framework and provides guidance on the appropriate processing methods for each portion of the approach. From simulation studies and wind speed data, the results highlight that the auto-associative neural network method provides the lowest prediction error when compared with regression, neural network, and principal component analysis methods. The results from this dissertation work also imply that the selection of the clustering algorithm does not significantly affect the calculated health value, and in general, most clustering algorithms appear suitable for detecting the problem using the residual clustering approach. The feasibility of the residual clustering approach is demonstrated in three case studies. For the wind speed sensor health monitoring case study, the residual clustering method provides the most accurate health assessment of the wind speed sensors when compared with the other methods used by the 24 participants in the Prognostics and Health Management 2011 Data Challenge. The residual clustering approach also outperformed other multi-regime health monitoring methods such as a mixture distribution overlap method for the gearbox case study. The residual clustering method was also able to provide an early detection of a problem on the wind turbine rotor shaft with 26 days of advanced warning. The rotor shaft health value using the residual clustering approach had the most monotonic health trend when compared with three other multi-regime health monitoring methods for the wind turbine drivetrain case study. The dissertation work shows that the residual clustering approach is fundamentally sound and should be considered along with the existing methods for multi-regime condition monitoring applications. The method appears to outperform many of the existing methods, and would be an appropriate monitoring algorithm if there is a nominal amount of correlation in the measured signals. Additional refinement of the approach can look into more sophisticated methods for threshold setting along with integrating a feature selection method into the residual clustering framework. In addition, algorithms for diagnosis and remaining useful life estimation for multi-regime condition monitoring applications would also require additional research and development work.

Committee:

Jay Lee, Ph.D. (Committee Chair); Canh Ly, Ph.D. (Committee Member); Teik Lim, Ph.D. (Committee Member); David Thompson, Ph.D. (Committee Member)

Subjects:

Mechanics

Keywords:

Residual Clustering;Health Monitoring;Auto-Associative Neural Network;Multi-Regime;Sensor Health;Gearbox Condition Monitoring

Gnanasekar, NithyakumaranTemperature and Hourly Precipitation Prediction System for Road Bridge using Artificial Neural Networks
MS, University of Cincinnati, 2015, Engineering and Applied Science: Computer Science
UCII has designed and built the Weather monitoring system for Bridges, at multiple locations in North America. The Weather Monitoring System collects data from the local weather stations, airports around the bridge and the sensors installed on the bridge. This data is analyzed for inclement weather conditions on and around the bridge and whenever an abnormal behavior is detected, “Alarms” are sent out via email. Using the large amount of data collected from various stations and sensors correlation between every variable is studied. This detailed study is later used to build Eight Hour Ahead Temperature Prediction system and a Four Hours Ahead Hourly Precipitation Prediction System. The Prediction system uses machine learning techniques to predict new data based on prior data. The Atmospheric variables collected such as temperature, pressure, humidity, hourly precipitation, wind speed, wind direction, solar radiation is analyzed. The correlation between the afore-mentioned variables on time scale is also analyzed and is discussed in detail which aided in building the Prediction models. The prediction system uses an Artificial Neural network to train and provide predictions. The prediction system was designed and built to be used alongside Weather Monitoring system built by UCII and to further its intelligence coefficient in predicting inclement weather condition.

Committee:

Arthur Helmicki, Ph.D. (Committee Chair); Victor Hunt, Ph.D. (Committee Member); Paul Talaga, Ph.D. (Committee Member)

Subjects:

Engineering

Keywords:

Artificial Neural Network;Road way bridge;local weather prediction;weather monitoring system

Pusuluri, Sai TejaExploring Neural Network Models with Hierarchical Memories and Their Use in Modeling Biological Systems
Doctor of Philosophy (PhD), Ohio University, 2017, Physics and Astronomy (Arts and Sciences)
Energy landscapes are often used as metaphors for phenomena in biology, social sciences and finance. Different methods have been implemented in the past for the construction of energy landscapes. Neural network models based on spin glass physics provide an excellent mathematical framework for the construction of energy landscapes. This framework uses a minimal number of parameters and constructs the landscape using data from the actual phenomena. In the past neural network models were used to mimic the storage and retrieval process of memories (patterns) in the brain. With advances in the field now, these models are being used in machine learning, deep learning and modeling of complex phenomena. Most of the past literature focuses on increasing the storage capacity and stability of stored patterns in the network but does not study these models from a modeling perspective or an energy landscape perspective. This dissertation focuses on neural network models both from a modeling perspective and from an energy landscape perspective. I firstly show how the cellular interconversion phenomenon can be modeled as a transition between attractor states on an epigenetic landscape constructed using neural network models. The model allows the identification of a reaction coordinate of cellular interconversion by analyzing experimental and simulation time course data. Monte Carlo simulations of the model show that the initial phase of cellular interconversion is a Poisson process and the later phase of cellular interconversion is a deterministic process. Secondly, I explore the static features of landscapes generated using neural network models, such as sizes of basins of attraction and densities of metastable states. The simulation results show that the static landscape features are strongly dependent on the correlation strength and correlation structure between patterns. Using different hierarchical structures of the correlation between patterns affects the landscape features. These results show how the static landscape features can be controlled by adjusting the correlations between patterns. Finally, I explore the dynamical features of landscapes generated using neural network models such as the stability of minima and the transition rates between minima. The results from this project show that the stability depends on the correlations between patterns. It is also found that the transition rates between minima strongly depend on the type of bias applied and the correlation between patterns. The results from this part of the dissertation can be useful in engineering an energy landscape without even having the complete information about the associated minima of the landscape.

Committee:

Horacio Castillo (Advisor)

Subjects:

Bioinformatics; Condensed Matter Physics; Genetics; Physics

Keywords:

Neural Network Models; Spin Glass Physics; Gene Regulatory Networks; Epigenetics

Sullivan, John B.The application of an artificial neural network to a turning movement detector system
Master of Science (MS), Ohio University, 1991, Industrial and Manufacturing Systems Engineering (Engineering)
The application of an artificial neural network to a turning movement detector system

Committee:

Donald Scheck (Advisor)

Subjects:

Engineering, Industrial

Keywords:

Artificial Neural Network; Turning Movement Detector System

Bishop, Russell C.A Method for Generating Robot Control Systems
Master of Science in Engineering, Youngstown State University, 2008, Department of Electrical and Computer Engineering
This thesis presents a method of generating neural-network based control systems for walking robots. A genetic learning rule is combined with a physics simulation and scoring system in order to find appropriate weights for these networks. This approach produces highly robust neural-network control mechanisms that are capable of handling a wide variety of conditions, such as rough terrain and randomly varying robot proportions. In each of two test runs, the system was able to make the robot walk approximately 1.75 meters (5.8 body lengths) in the physics simulation, over very rough terrain, in 14 seconds of simulation-world time.

Committee:

Jalal Jalali, PhD (Advisor); Frank Li, PhD (Committee Member); Philip Munro, PhD (Committee Member)

Subjects:

Robots

Keywords:

genetic algorithm; neural network; walking robot

Nelapati, PraneethAn Intelligent SOP Navigation System with Two Mobile Receivers
Master of Science in Electrical Engineering, University of Toledo, 2011, College of Engineering

In this thesis an Intelligent Signals of opportunity (SOP) system is proposed. The intelligence integrated in the system is based on Self Cloned Evolutionary Neural Networks. Signals of Opportunity (SOP) are highly efficient and effective alternative for the Global Positioning System (GPS). The conventional SOP system has one fixed base receiver and one mobile receiver. SOP helps navigation by precisely measuring the time difference between the signals arriving at two points, namely base station and mobile unit. The principal difficulty involved in establishing the SOP is the time and cost required for setting up the base stations. A novel approach proposed in this thesis eliminates the need for fixed base receiver and replaces it with the mobile receiver.

The proposed system was simulated for both the Amplitude Modulated and Frequency Modulated Systems. It has been demonstrated that improved SOP system is 94.9 % more efficient than the conventional method for the Amplitude Modulated (AM) signals. It is 91.43% more efficient over conventional system for Frequency Modulated (FM) signals. The intelligence built into the system identifies the most appropriate signal available in a given locale. In an urban area, where the available signal resources are abundant, the SOP system signal selection process is time consuming and thus blocks the navigational service occasionally. To overcome the problem, intelligence is built into the system using Evolutionary Neural Network which trains based on the past analysis of the system. To make the learning in the neural network faster than a conventional evolutionary algorithm, an algorithms based on selective cloning is used.

Committee:

Kaur Devinder, PhD (Committee Chair); Salari Ezzatollah, PhD (Committee Member); Henry Ledgard, PhD (Committee Member)

Subjects:

Computer Science; Electrical Engineering

Keywords:

Signals of Opportunity; Navigation Systems; GPS; Neural Network; Genetic Algorithm ; Selective Cloning

Pappala, SwethaDevice Specific Key Generation Technique for Anti-Counterfeiting Methods Using FPGA Based Physically Unclonable Functions and Artificial Intelligence
Master of Science in Electrical Engineering, University of Toledo, 2012, Electrical Engineering

Anti-counterfeiting techniques have entered a new era with the implementation of critical designs and confidential information transfer protocols. The complexity in developing security mechanisms and routing protocols for embedded systems continues to increase; on the other hand, cost and size constraints have been lowered. Trustworthy authentication of a device is of extreme importance for secure protocols. Methodologies for preventing IC piracy have been developed that require a unique signature key for every fabricated chip. Physically Unclonable Functions (PUFs) can be used for such signature generation.

This research implements a key generation process using a novel Ring Oscillator PUF (ROPUFs) design followed by an error correcting code, and a hashing algorithm. The key generation process has been implemented in three phases: ROPUF, Error Correction Process, and a Hashing Algorithm. The ROPUF design takes advantage of the unique characteristic properties of FPGAs. In this work, the ROPUFs are implemented using LUTs, multiplexers and flip flops that are the basic components of the FPGA architecture. The PUF design is followed by an error correction process to rectify any noisy bits in the response due to drastic environmental changes like temperature and voltage fluctuations. Artificial Neural Networks are used for the error correction process. The latter part of the research deals with a hashing function that has been implemented to enhance the security of the key generation process. The hashing function redresses the response bits of the PUF unit to mask the challenge-response pairs.

The proposed PUF circuit is implemented on 5 Xilinx Spartan 2 XC2S100 FPGAs, and an Agilent 16801A Logic Analyzer is used to obtain the PUF responses. The intra-chip and inter-chip responses are analyzed and plotted using Hamming distances. The overall uniqueness of the responses is found to be 49.0625% which is higher when compared to the previous implementations of the conventional ROPUF circuit (43.40%), and the earlier chain-implementation (48.51%). The inter-chip and intra-chip uniqueness factor for the proposed design are 47.929% and 41.91% respectively.

Artificial Neural Networks are tested using the PUF responses of various lengths. The failure rates of the proposed method are below 1 ppm which is lower than the failure rate of BCH codes which is typically 4.8 ppm. The SHA-256 algorithm is optimized using parallel processing techniques to give better throughput results. The delay is reduced to 45 clock cycles.

Committee:

Mohammed Niamat, PhD (Committee Chair); Weiqing Sun, PhD (Committee Co-Chair); Mansoor Alam, PhD (Committee Member)

Subjects:

Computer Science; Electrical Engineering

Keywords:

FPGA; PUF; Cryptography; Security; Error Correcting Code; Neural Network.

WOLFE, GLENN APERFORMANCE MACRO-MODELING TECHNIQUES FOR FAST ANALOG CIRCUIT SYNTHESIS
PhD, University of Cincinnati, 2004, Engineering : Computer Engineering
This work focuses on the development of accurate and efficient performance parameter macro-models for use in the synthesis of analog circuits. Once constructed the mathematical models may be used as substitutes for full SPICE simulation, providing efficient computation of performance parameter estimates. In this thesis, we explore various modeling architectures, develop and apply two unique sampling methodologies for adaptively improving model quality, and attempt to apply the sizing rules methodology in order to perform dimensional reduction and ensure proper operation of analog circuits. In order to properly create an analog performance model, a training data set is needed to create the model, and an independent validation data set is needed to verify the accuracy of the model. The training and validation data sets are comprised of discretely sampled points in the design space. Various methods exist for generating these sample points. A static sampler does not take into account the shape of the function under scrutiny, whereas an adaptive sampler strives to reduce modeling error through strategic placement of costly sample points. Two unique adaptive sampling methodologies are developed and are applied to various analog circuit performance metrics. It is shown experimentally that both adaptive samplers are capable of improving maximum modeling errors for various performance metrics and analog topologies. Strategic placement of costly sample points improves model quality while reducing the time needed to create the performance models. Adaptive sampling also alleviates human intervention during model construction, realizing an automatic framework for sampling and modeling performance parameters. The sizing rules method and feasibility region modeling are analyzed and applied to analog performance macro-modeling in an attempt to automatically reduce the dimensionality of the design space, simplify performance parameter behavior, and ensure proper DC biasing. A feasibility region is a portion of the design space satisfying design space and electrical space inequality constraints generated by the sizing rules method. Experimental evidence indicates that the sizing rules method alone does not sufficiently constrain a circuit to facilitate the creation of accurate analog performance macro-models. Additional, manually derived design constraints are required to enable the development of accurate performance parameter models.

Committee:

Dr. Ranga Vemuri (Advisor)

Keywords:

analog synthesis; analog sizing; analog performance macro-modeling; neural network; pseudo-cubic spline; feasibility regions; feasible region modeling; adaptive sampling

Wu, XiaomingApproximation using linear fitting neural network: Polynomial approach and gaussian approach
Master of Science (MS), Ohio University, 1991, Electrical Engineering & Computer Science (Engineering and Technology)
Approximation using linear fitting neural network: Polynomial approach and gaussian approach

Committee:

Henryk Lozykowski (Advisor)

Keywords:

Approximation; Linear Fitting Neural Network; Polynomial Approach; Gaussian Approach

Collins, Peter ChancellorA combinatorial approach to the development of composition-microstructure-property relationships in titanium alloys using directed laser deposition
Doctor of Philosophy, The Ohio State University, 2004, Materials Science and Engineering
The Laser Engineered Net Shaping (LENS™) system, a type of directed laser manufacturing, has been used to create compositionally graded materials. Using elemental blends, it is possible to quickly vary composition, thus allowing fundamental aspects of phase transformations and microstructural development for particular alloy systems to be explored. In this work, it is shown that the use of elemental blends has been refined, such that bulk homogeneous specimens can be produced. When tested, the mechanical properties are equivalent to conventionally prepared specimens. Additionally, when elemental blends are used in LENS™ process, it is possible to deposit compositionally graded materials. In addition to the increase in design flexibility that such compositionally graded, net shape, unitized structures offer, they also afford the capability to rapidly explore composition-microstructure-property relationships in a variety of alloy systems. This research effort focuses on the titanium alloy system. Several composition gradients based on different classes of alloys (designated a, a+b, and b alloys) have been produced with the LENS™. Once deposited, such composition gradients have been exploited in two ways. Firstly, binary gradients (based on the Ti-xV and Ti-xMo systems) have been heat treated, allowing the relationships between thermal histories and microstructural features (i.e. phase composition and volume fraction) to be explored. Neural networks have been used to aid in the interpretation of strengthening mechanisms in these binary titanium alloy systems. Secondly, digitized steps in composition have been achieved in the Ti-xAl-yV system. Thus, alloy compositions in the neighborhood of Ti-6Al-4V, the most widely used titanium alloy, have been explored. The results of this have allowed for the investigation of composition-microstructure-property relationships in Ti-6-4 based systems.

Committee:

Hamish Fraser (Advisor)

Subjects:

Engineering, Materials Science

Keywords:

combinatorial method; combinatorial approach; laser deposition; directed laser deposition; LENS; titanium; molybdenum; Ti-6-4; Ti-6Al-4V; Timetal 21S; composition; microstructure; property; relationships; neural network; fuzzy logic

Huang, I-Wen EvanUniform Corrosion and General Dissolution of Aluminum Alloys 2024-T3, 6061-T6, and 7075-T6
Doctor of Philosophy, The Ohio State University, 2016, Materials Science and Engineering
Uniform corrosion and general dissolution of aluminum alloys was not as well-studied in the past, although it was known for causing significant amount of weight loss. This work comprises four chapters to understand uniform corrosion of aluminum alloys 2024-T3, 6061-T6, and 7075-T6. A preliminary weight loss experiment was performed for distinguishing corrosion induced weight loss attributed to uniform corrosion and pitting corrosion. The result suggested that uniform corrosion generated a greater mass loss than pitting corrosion. First, to understand uniform corrosion mechanism and kinetics in different environments, a series of static immersion tests in NaCl solutions were performed to provide quantitative measurement of uniform corrosion. Thereafter, uniform corrosion development as a function of temperature, pH, Cl-, and time was investigated to understand the influence of environmental factors. Faster uniform corrosion rate has been found at lower temperature (20 and 40°C) than at higher temperature (60 and 80°C) due to accelerated corrosion product formation at high temperatures inhibiting corrosion reactions. Electrochemical tests including along with scanning electron microscopy (SEM) were utilized to study the temperature effect. Second, in order to further understand the uniform corrosion influence on pit growth kinetics, a long term exposures for 180 days in both immersion and ASTM-B117 test were performed. Uniform corrosion induced surface recession was found to have limited impact on pit geometry regardless of exposure methods. It was also found that the competition for limited cathodic current from uniform corrosion the primary rate limiting factor for pit growth. Very large pits were found after uniform corrosion growth reached a plateau due to corrosion product coverage. Also, optical microscopy and focused ion beam (FIB) imaging has provided more insights of distinctive pitting geometry and subsurface damages found from immersion samples and B117 samples. Although uniform corrosion was studied in various electrolytes, the pH impact was still difficult to discern due to ongoing cathodic reactions that changed electrolyte pH with time. Therefore, buffered pH electrolytes with pH values of 3, 5, 8, and 10 were prepared static immersion tests. Electrochemical experiments were performed in each buffered pH conditions for understanding corrosion mechanisms. Uniform corrosion was found exhibiting higher corrosion rate in buffered acidic and alkaline electrolytes due to pH- and temperature-dependent corrosion product precipitation. Observations were supported by electrochemical, SEM, and EDS observations. Due to the complexity of corrosion data, a reliable corrosion prediction based on empirical observations could be challenging. Artificial neural network (ANN) modeling was used for corrosion data pattern recognition by mimicking human neural network systems. Predictive models were developed based on corrosion data acquired in this study. The model was adaptable through iteratively update its prediction by error minimization during the training phase. Trained ANN model can predict uniform corrosion successfully. In addition to ANN, fuzzy curve analysis was utilized to rank the influence of each input (temperature, pH, Cl-, and time). For example, temperature and pH were found to be the most influential parameters to uniform corrosion. This information can provide feedback for ANN improvement, also known as “data pruning”.

Committee:

Rudolph Buchheit (Advisor); Gerald Frankel (Committee Member); Jenifer Locke (Committee Member); Christopher Taylor (Committee Member)

Subjects:

Engineering; Materials Science; Metallurgy

Keywords:

uniform corrosion, pitting, aluminum alloys, artificial neural network

Next Page