Master of Science / Department of Electrical and Computer Engineering / Balasubramaniam Natarajan / Decentralized sensor networks are collections of individual local sensors that observe a common phenomenon, quantize their observations, and send this quantized information to a central processor (fusion center) which then makes a global decision about the phenomenon. Most of the existing literature in this field consider only the data fusion aspect of this problem, i.e., the statistical hypothesis testing and optimal combining of the information obtained by the local sensors. In this thesis, we look at both the data detection and the data fusion aspects of the decentralized sensor networks. By data detection, we refer to the communication problem of transmitting quantized information from the local sensors to the fusion center through a multiple access channel. This work first analyzes the data fusion problem in decentralized sensor network when the sensor observations are corrupted by additive white gaussian noise. We optimize both local decision rules and fusion rule for this case. After that, we consider same problem when the observations are corrupted by correlated gaussian noise. We propose a novel parallel genetic algorithm which simultaneously optimizes both the local decision and fusion rules and show that our algorithm matches the results from prior work with considerably less computational cost. We also demonstrate that, irrespective of the fusion rule, the system can provide equivalent performance with an appropriate choice of local decision rules. The second part of this work analyzes the data detection problem in distributed sensor networks. We characterize this problem as a multiple input multiple output (MIMO) system problem, where the local sensors represent the multiple input nodes and the fusion center(s) represent the output nodes. This set up, where the number of input nodes (sensors) is greater than the number of output nodes (fusion center(s)), is known as an overloaded array in MIMO terminology. We use a genetic algorithm to solve this overloaded array problem.

Doctor of Philosophy / Department of Electrical and Computer Engineering / James E. DeVault / Prakash Krishnaswami / Most state-of-the-art systems are multidisciplinary in nature and encompass a wide range of components from domains such as electronics, mechanics, hydraulics, etc. Design considerations and design parameters of the system can come from any or a combination of these domains. The traditional optimization approach for multidisciplinary systems utilizes sequential optimization, wherein each subsystem is optimized in isolation in a predetermined order, assuming that the designs of the other subsystems remain fixed. This often leads to system designs that are suboptimal. In recent years emphasis has been placed on development of an integrated scheme for analysis and design of multidisciplinary systems. An important aspect is the software architecture required to support such a scheme. This dissertation presents MIXEDMODELS (Multidisciplinary Integrated eXtensible Engine for Driving Metamodeling, Optimization and DEsign of Large-scale Systems) - a unified analysis and design tool for multidisciplinary systems that is based on a procedural, symbolic-numeric architecture. This architecture offers great modeling flexibility at the component level, allowing any engineer to add components in his/her domain of expertise to the platform in a modular fashion. The symbolic engine in the MIXEDMODELS platform synthesizes the system governing equations as a unified set of nonlinear differential-algebraic equations (DAEs). These equations are differentiated with respect to design variables to obtain an additional set of DAEs that describe the sensitivity coefficients of the system state variables. This combined set of DAEs is solved numerically to obtain the solution for the state variables and the state sensitivity coefficients of the system. Finally, knowing the system performance functions, their design sensitivity coefficients can be calculated by using the values of the state variables and state sensitivity coefficients obtained from the DAEs. For ease in error control and software implementation, sensitivity analysis formulation described in this work uses direct differentiation approach as opposed to the adjoint variable approach. The MIXEDMODELS capabilities are demonstrated through several numerical examples and the results indicate that the MIXEDMODELS formulation and architecture is effective in terms of accuracy, modeling convenience, computational efficiency, and the ability to simulate the behavior of a general class of multidisciplinary systems.

Master of Science / Department of Mechanical and Nuclear Engineering / Warren N. White / This thesis describes the process of testing and implementing ultrasonic transducers for ball position feedback on a ball-and-beam apparatus. Also included are specifications for equipment to allow feedback and command signals to be wireless, not hardwired to the control computer. The author presents various ball-and-beam configurations as well as details about the specific configuration used for this work. These details include choices in sensors, materials, hardware, construction, and controller. After the apparatus has been described, the author provides information to support claims about system performance. The conclusions presented specify the necessary hardware to make the system wireless and indicate that acoustic sensors can complete a successful ball-on-beam balancing system.

Master of Science / Department of Architectural Engineering and Construction Science / Fred L. Hasler / This paper discusses photopic and scotopic vision of the human eye and the implications that could result in the design process of the lighting industry. The incorporation of scotopic vision in lighting effects the perceived illumination in all settings; but these affects and benefits are seen more prevalently at night, as this is when scotopic vision is utilized by the eye the most. The paper will begin with an overview of the eye including discussions of exactly what photopic and scotopic vision are, as well as how the eye works. This will lay a foundation for the paper to help the reader better comprehend and understand the remainder of the content. After the groundwork has been laid, the factors that affect how the eye perceives light will be discussed. These factors include pupil size and color of the light. A discussion of the basis for current lighting industry design and how light levels are measured will follow. Once these topics have been fully explored, there will be a discussion of the changes that could occur in the lighting industry if scotopic vision is taken into account. Increased energy efficiency would result if the scotopic vision is incorporated, resulting from the decrease in needed total lumen output. There have been a few applications that have utilized the effects of the scotopic vision in their design, these cases will be presented. Following the case study discussions, will be a discussion of a survey conducted by myself on the change-out of high pressure sodium (HPS) fixtures to LED fixtures in the downtown Poyntz Avenue area of Manhattan, KS. After all studies have been reviewed, conclusions and correlations among them will be explored. Following this analysis, suggestions will be given to improve the way lighting is designed in the industry.

Master of Science / Department of Electrical and Computer Engineering / Dwight D. Day / Micro-hydro turbines generate power for small villages and industries in Afghanistan. They usually produce less than 100 kW of power. Currently the flow into the turbine is controlled manually and the voltage is controlled automatically with an electronic load controller. Excess power not used by the village is dumped into a community water heater. For larger sites that have a reservoir and/or large variable load throughout the day and night, the turbine needs to be fitted with an automatic flow control system to conserve water in the reservoir or deal with the variable loads. Large turbines usually use hydraulic governors that automatically adjust the flow of water into the turbine. For micro-hydro sized plants this method would be too expensive and be difficult to build and maintain locally. For this reason, a 3 phase AC induction motor will be used to move the internal flow control valve of the turbine. Because a sudden change in load is possible (30 – 40%) for micro-hydro plants, the electronic load controller will also be needed to respond to quick changes in load so that the village voltage does not exceed 220V. This report documents the process of building a test system comprising of a dynamic resistive load, microcontroller controlled resistive load, a three phase AC generator and a DC Motor. Where the dynamic resistive load represents the load of the village, the computer controlled resistive load would represent the community water heater, the three phase AC generator represents the Generator on site and the DC Motor together with its DC input voltage would emulate the turbine and its water flow respectively. The DC input voltage would be also controlled with a PWM signal through a delay loop to represent the water gate delay effects on the turbine as close as possible. With this, it would be possible to completely build and test a control system that emulates the dynamics of a water turbine generator.

Doctor of Philosophy / Department of Biological & Agricultural Engineering / Naiqian Zhang / Suspended sediment concentration (SSC) in water is one of the most important parameters to evaluate water quality. Monitoring SSC provides important information on determining sediment transport for soil erosion research and soil/water conservation practices. Sediment mass transported at a given time can be assessed by simultaneous SSC and water flow velocity measurements. Fouling, including bio-fouling, has damaging impact on optical SSC measurements over the long term. In this study, an inexpensive, real-time, self-cleaning, optical sediment and flow velocity sensor was developed. Laboratory experiments were conducted on a previously designed SSC sensor. A light modulation algorithm was designed to reduce the influence of ambient light, especially sunlight, on measurement accuracy. Statistical models to predict SSC based on measured light intensities were established and compared with neural network models. The statistical analysis showed that soil texture played an important role in SSC measurement accuracy while the designed sensor was capable of reducing the effect of water color on sensor performance. Neural-network models can further remove the influence of soil texture type on SSC measurement. The sensor design was simplified based on a stepwise selection analysis. Long-term field experiments were conducted in Kansas and Georgia to evaluate the sensor performance, the effect of fouling, including bio-fouling, on sensor lenses, and the effect of temperature on the measurement. Methods of removing the fouling effect through data correction were developed. Results indicated that the designed optical SSC sensor was capable of providing rapid response to SSC fluctuations in water flow. Temperature of the water body has an insignificant impact on SSC measurement. In order to reduce fouling, an air-blast cleaning mechanism was integrated into the optical sediment sensor. Laboratory experiments in a manually created fouling environment were conducted to observe the fouling process on sensor cases made of different materials, and to verify the effectiveness of air-blast cleaning in reducing fouling. Results indicated that air-blast cleaning mechanism was capable of reducing clay/silt fouling on sensor signals. The duration and frequency of air-blast cleaning can be determined and adjusted depending on actual field conditions. An air pressure drop test was conducted on the hose carrying pressurized air. Results showed negligible pressure drop.A flow velocity measurement function based on the cross-correlation principle was integrated into the optical sediment sensor. An experiment was conducted in laboratory to examine the sensor performance on velocity measurement using a closed circulation system. A solution of blue colorant, Brilliant Blue FCF, was used as an artificial source to absorb light emitted by LEDs in the sensor and the signal variation patterns were measured. The results indicated that the cross-correlation-based velocity sensor was capable of measuring water flow velocity within in a certain velocity range using the dye injection method.

Master of Science / Department of Electrical and Computer Engineering / Balasubramaniam Natarajan / In this thesis, we first consider the problem of distributed estimation in an energy and rate-constrained wireless sensor network. To this end, we study three estimators namely - (1) Best Linear Unbiased Estimator (BLUE-1) that accounts for the variance of noise in measurement, uniform quantization and channel, and derive its variance and its lower bound; (2) Best Linear Unbiased Estimator (BLUE-2) that accounts for the variance of noise in measurement and uniform quantization, and derive lower and upper bounds for its variance; (3) Best Linear Unbiased Estima- tor (BLUE-3) that incorporates the effects of probabilistic quantization noise and measurement noise, and derive an upper bound for its variance. Then using BLUE-1, we analyze the tradeoff between estimation error (BLUE variance) at the fusion center and the total amount of resources utilized (power and rate) using three different system design approaches or optimization formulations. For all the formulations, we determine optimum quantization bits and transmission power per bit (or optimum actions) for all sensors jointly. Unlike prior efforts, we in- corporate the operating state (characterized by the amount of residual battery power) of the sensors in the optimization framework. We study the e®ect of channel quality, local measurement noise, and operating states of the sensors on their optimum choice for quantization bits and transmit power per bit. In the sequel, we consider a problem in distributed detection and signal processing in the context of biomedical wireless sensors and more specifically pulse- oximeter devices that record photoplethysmographic data. We propose an automated, two-stage PPG data processing method to minimize the effect of motion artifact. Regarding stage one, we present novel and consistent techniques to detect the presence of motion artifact in photoplethysmograms given higher order statistical information present in the data.For stage two, we propose an effective motion artifact reduction method that involves enhanced PPG data preprocessing followed by frequency domain Independent Component Analysis (FD-ICA). Experimental results are presented to demonstrate the efficacy of the overall motion artifact reduction method. Finally, we analyze a wireless ad hoc/sensor network where nodes are connected via random channels and information is transported in the network in a cooperative multihop fashion using amplify and forward relay strategy.

Master of Science / Department of Mechanical and Nuclear Engineering / Akira T. Tokuhiro / Robots will eventually become common everyday items. However before this becomes a reality, robots would need to learn be socially interactive. Since humans communicate much more information through expression than through actual spoken words, expression recognition is an important aspect in the development of social robots. Automatic recognition of emotional expressions has a number of potential applications other than just social robots. It can be used in systems that make sure the operator is alert at all times, or it can be used for psycho-analysis or cognitive studies. Emotional expressions are not always deliberate and can also occur without the person being aware of them. Recognizing these involuntary expressions provide an insight into the persons thought, state of mind and could be used as indicators for a hidden intent. In this research we developed an initial multi-modal emotion recognition system using cues from emotional expressions in face and voice. This is achieved by extracting features from each of the modalities using signal processing techniques, and then classifying these features with the help of artificial neural networks. The features extracted from the face are the eyes, eyebrows, mouth and nose; this is done using image processing techniques such as seeded region growing algorithm, particle swarm optimization and general properties of the feature being extracted. In contrast features of interest in speech are pitch, formant frequencies and mel spectrum along with some statistical properties such as mean and median and also the rate of change of these properties. These features are extracted using techniques such as Fourier transform and linear predictive coding. We have developed a toolbox that can read an audio and/or video file and perform emotion recognition on the face in the video and speech in the audio channel. The features extracted from the face and voices are independently classified into emotions using two separate feed forward type of artificial neural networks. This toolbox then presents the output of the artificial neural networks from one/both the modalities on a synchronized time scale. Some interesting results from this research is consistent misclassification of facial expressions between two databases, suggesting a cultural basis for this confusion. Addition of voice component has been shown to partially help in better classification.