Thesis (M.S.)--University of Wyoming, 2006. / Title from PDF title page (viewed on Nov. 30, 2006). Includes bibliographical references (p. 42-47).

The reliability of electronic component plays an important role in proper functioning of the electronic devices. The manufacturer tests electronic components before they are used by end users. Still at times electronic devices fail due to undue stresses existing inside the microelectronic components such as memory chips, microcontrollers, resistors etc. The stresses can be caused by variation in the operating voltage, variation in the usage frequency of the particular chip and other factors. This variation leads to variation in chip temperature, which can be made evident from thermal profiles of these chips. In this thesis, effort was made to study two different kind of stress existing in the electronic board, namely signal stress based on variation in duty cycle/frequency of chip usage and the voltage stress. Memory chips were stressed using these stresses causing change in heating rates, which were captured by infrared camera. This data was then extracted and plotted to obtain different curves for the heating rate. The same experiment was done time and again for a large number of chips to get heating rate data. This data consisting of average heating rate for large number of chips was used to build Neural Network model (NN). Back Propagation algorithm was used for modeling because of its advantage of converging to solution faster compared to other algorithms. To develop a prediction model, data sets were divided into two-third and one-third parts. This two-thirds of the data was used to build the prediction model and remaining one third was used to evaluate the model. The designed model would predict the stress levels existing in the chips based on the heating rates of the chips. Results obtained suggested 1. There is difference in heating rate for chips stressed at different stress levels. 2. Accuracy of the model to predict the stress is high (greater than 90 %). 3. Model is robust enough that is it can yield efficient results even if there is presence of noise in the data. 4. Generic methodology can be proposed based on the experiments. This work is a progress in direction of making predictive model, for a complete electronic device, which can predict the stress level existing on any component in the device and will provide an opportunity to either protect the data or removal of the defected components timely before it even fails.

This dissertation investigates three issues on business failure causality and prediction. First, a nonlinear model for mathematical programming based discriminant analysis is studied. This study proposes a nonlinear model that builds on the existing linear and quadratic models and allows for a more flexible degree of nonlinearity through a set of power parameters. The proposed nonlinear model is solved using a genetic algorithm and is tested against linear and quadratic models using real financial data. The results show that each model is better in certain cases, but the nonlinear model turns out to be the best overall among the three. Better performance of this nonlinear model appears likely, but a more robust solver would be required. Second, the relationship between aggregate business failures and macroeconomic conditions is studied from a causality perspective. A structural Vector Autoregression (VAR) is used while incorporating the recently developed causal inference method Directed Acyclic Graph (DAG). Particularly, DAG is used to provide a contemporaneous causal structure and the VAR results are summarized using innovation accounting techniques. The results show that during the period from 1980 to 2004 in the U.S., aggregate business failures were influenced by interest rates, but overall these failures appear to be far more exogenous than was found previously. Third, the effect of incorporating macroeconomic variables into business failure prediction models is investigated with a focus on the U.S. airline industry from 1995 to 2005. The attention is placed on prediction accuracy, parameter stability, and the effect of particular macroeconomic variables. The results show that the stability of parameters in the prediction model is improved when macro variables are added. In terms of prediction accuracy, the model augmented with a macro variable performed better in a jackknife prediction, but not in out-of-sample predictions. The macroeconomic variable found to be significant is the change of interest rate, which is probably related to the high level of leverage common in this particular industry. Also, the results demonstrate that a probability score can be used as a more informative evaluation measure than the current one based on cutoff probabilities.

Smooth cordgrass (Spartina alterniflora) is one of the harmful quarantine weeds in China. Since its first introduction in China in 1979, this alien species has spread rapidly and damaged local ecological environments. Research to predict a suitable new area is an important step for management of the species and to prevent a further spread. In this study, Spartina alterniflora’s ecological niche was modeled using the application MAXENT. Analysis was based on species’ current distribution. The investigations of this study were two-fold. First, a large-scale global investigation (outside China) was conducted to predict suitable areas in China by comparing global and Chinese records of the species. In the second set, the combined records were used to predict suitable areas in the Jiangsu Province. The model’s accuracy was evaluated by Receiver Operator Characteristic (ROC) curve. The areas under the ROC curve (AUC value) were all over 0.95, which indicated high predictive ability of this model. In the large scale prediction, Shanghai, Zhejiang, Fujian, Guangzhou, Guangxi and southern part of Wuhan, Jiangsu and Anhui were all potentially endangered by S. alterniflora invasion. On the smaller scale, the prone to invasion areas were mostly concentrated on southern part and some coastal areas of Jiangsu Province, where the precipitation and temperature were appropriate for this grass. Because of S. alterniflora has high dispersal ability and human induced history, the potential distribution areas in China are considerable and it may invade more areas, in result spreading faster in the future. To prevent further invasion and spread, an early eradication program should be adopted in the newly invaded areas. Meanwhile, the monitoring programs should also need to be applied in potential survival areas, especially in coastal harbors, airports, and tourism areas which are highly vulnerable to S. alterniflora invasion.

Disulfide bonds play crucial roles to predict the three-dimensional structure and the function of a protein. This thesis develops two algorithms to predict the disulfide bonding state of each cysteine in a protein sequence. These methods are based on the multi-stage framework and the multi-classifier of the support vector machine (SVM). The first algorithm achieves 94.0% accuracy of cysteine state prediction for dataset PDB4136, but in some datasets the results are not as good as our expectation. Thus the second algorithm is designed to improve the predicting ability for the proteins which have oxidized and reduced cysteines simultaneously. In addition, a new training strategy is also developed to increase the prediction accuracy. It appends the probabilities which are obtained from the SVM to the existing features and then starts a new training procedure repeatedly to get better performance. The experiments are performed on the datasets derived from well-known databases, such as Protein Data Bank and SWISS-PROT. It gets 94.3% accuracy for predicting disulfide bonding state on dataset PDB4136, which gets improvement 3.6% compared with the previously best result 90.7%.

For understanding the function of a protein, the protein structure plays an important role. The prediction of protein structure from its primary sequence has significant assistance in bioinformatics. Generally, the real protein structures can be reconstructed by some costly techniques, but predicting the protein structures helps us guess the functional expression of a protein in advance. In this thesis, we develop three terms as the materials of the fitness function that can be successfully used in protein backbone structure prediction. In the result of this thesis, it shows that over 80% of good values calculated from our fitness function, which are generated by the genetic programming, are better than the average in the CASP8.

The trend in passenger vehicle internal combustion (IC) engines is to produce smaller, more fuel-efficient engines with power outputs comparable to those of large displacement engines. One way to accomplish this goal is through using turbochargers (TCs) supported on semi-floating ring bearings (SFRBs). The thesis presents progress on the nonlinear modeling of rotor-bearing systems (RBSs) by including engine-induced TC housing excitations. Test data collected from an engine-mounted TC unit operating to a top speed of 160 krpm (engine speed = 3,600 rpm) validates the nonlinear predictions of shaft motion. Engine-induced housing excitations are input into the nonlinear time transient rotor model as Fourier coefficients (and corresponding phase angles) derived from measured TC center housing accelerations. Analysis of recorded housing accelerations shows the IC engine induces TC motions with a broad range of subsynchronous frequencies, rich in engine (e) superharmonics. In particular, 2e and 4e vibration frequencies contribute greatly to housing motion. Most importantly, the analysis reveals TC center and compressor housings do not vibrate as a rigid body. Eigenvalue analysis of the TC system evidences four damped natural frequencies within the TC operating speed range. However, only the highest damped natural frequency (first elastic mode, f = 2,025 Hz, ξ = 0.14) is lightly-damped (critical speed = 150 krpm). Predicted linear and nonlinear imbalance response amplitudes increase with TC shaft speed, with linear predictions agreeing with test data at high shaft speeds. The differences between predictions and test data are attributed to an inaccurate knowledge of the actual TC rotor imbalance distribution. For the nonlinear analysis, predicted shaft motions not accounting for housing accelerations show the TC is stable (i.e. no subsynchronous whirl) at all but the lowest shaft speeds (<70 krpm). However, predicted shaft motions accounting for housing accelerations, as well as the test data, reveal TC motions rich in subsynchronous activity. Clearly, engine-induced housing accelerations have a significant impact on TC shaft motions. Predicted total shaft motions show good agreement with test data. Predicted nonlinear subsynchronous amplitudes as well as peak shaft amplitudes also agree well with test data. However, nonlinear predictions only show TC shaft vibrations attributed to engine order frequencies below 6e, whereas test data evidences TC vibrations are due to order frequencies greater than 6e. Overall, nonlinear predictions and test data illustrate the importance of accounting for engine-induced housing vibrations in the design and operation of TC systems. The good agreement between predictions and test data serve to validate the rotor model. The tools developed will aid a TC manufacturer in reducing development time and expenditures.

This dissertation investigates three issues on business failure causality and prediction. First, a nonlinear model for mathematical programming based discriminant analysis is studied. This study proposes a nonlinear model that builds on the existing linear and quadratic models and allows for a more flexible degree of nonlinearity through a set of power parameters. The proposed nonlinear model is solved using a genetic algorithm and is tested against linear and quadratic models using real financial data. The results show that each model is better in certain cases, but the nonlinear model turns out to be the best overall among the three. Better performance of this nonlinear model appears likely, but a more robust solver would be required. Second, the relationship between aggregate business failures and macroeconomic conditions is studied from a causality perspective. A structural Vector Autoregression (VAR) is used while incorporating the recently developed causal inference method Directed Acyclic Graph (DAG). Particularly, DAG is used to provide a contemporaneous causal structure and the VAR results are summarized using innovation accounting techniques. The results show that during the period from 1980 to 2004 in the U.S., aggregate business failures were influenced by interest rates, but overall these failures appear to be far more exogenous than was found previously. Third, the effect of incorporating macroeconomic variables into business failure prediction models is investigated with a focus on the U.S. airline industry from 1995 to 2005. The attention is placed on prediction accuracy, parameter stability, and the effect of particular macroeconomic variables. The results show that the stability of parameters in the prediction model is improved when macro variables are added. In terms of prediction accuracy, the model augmented with a macro variable performed better in a jackknife prediction, but not in out-of-sample predictions. The macroeconomic variable found to be significant is the change of interest rate, which is probably related to the high level of leverage common in this particular industry. Also, the results demonstrate that a probability score can be used as a more informative evaluation measure than the current one based on cutoff probabilities.

Thirty convenience stores in College Station, Texas, have been selected as the samples for an energy consumption prediction. The predicted models assist facility energy managers for making decisions of energy demand/supply plans. The models are applied to historical data for two years: 2001 and 2002. The approaches are (1) to analyze nonlinear regression models for long term forecasting of annual patterns compared with outdoor temperature, and (2) to analyze multiple regression models for the building type regardless of outdoor temperature. In the first approach, twenty four buildings are categorized as base load group and no base group. Average temperature, cooling efficiencies, and cooling knot temperature are estimated by nonlinear regression models: segment and parabola models. The adjusted r-square results in good performance up to ninety percent accuracy. In the second approach, the other selected six buildings are categorized as no trend group. This group does not respond to outdoor temperature. As the result, multiple a regression model is formed by combination of variables from the nonlinear models and physical building variables of cooling efficiency, cooling temperature, light bulbs, area, outdoor temperature, and orientation of fronts. This model explains up to sixty percent of all convenience stores' data. In conclusion, the accuracy of prediction models is measured by the adjusted r-square results. Among these three models, the multiple regression model shows the highest adjusted r-square (0.597) over the parabola (0.5419) and segment models (0.4806). When the three models come to the application, the multiple regression model is best fit for no trend data type. However, when it is used to predict the energy consumption with the buildings that relate to outdoor temperature, segment and parabola model provide a better prediction result.

The nature of user traffic in coming years will become increasingly multimediaoriented which has much more stringent Quality of Service (QoS) requirements. The current generation of the public Internet does not provide any strict QoS guarantees. Providing Quality of Service (QoS) for multimedia application has been a difficult and challenging problem. Developing predictive models for best-effort networks, like the Internet, would be beneficial for addressing a number of technical issues, such as network bandwidth provisioning, congestion avoidance/control to name a few. The immediate motivation for creating predictive models is to improve the QoS perceived by end-users in real-time applications, such as audio and video. This research aims at developing models for single-step-ahead and multi-stepahead prediction of end-to-end single flow characteristics in best-effort networks. The performance of path-independent predictors has also been studied in this research. Empirical predictors are developed using simulated traffic data obtained from ns-2 as well as for actual traffic data collected from PlanetLab. The linear system identification models Auto-Regressive (AR), Auto-Regressive Moving Average (ARMA) and the non-linear models Feed-forward Multi-layer Perceptron (FMLP) have been used to develop predictive models. In the present research, accumulation is chosen as a signal to model the end-to-end single flow characteristics. As the raw accumulation signal is extremely noisy, the moving average of the accumulation isused for the prediction. Developed predictors have been found to perform accurate single-step-ahead predictions. However, as the multi-step-ahead prediction horizon is increased, the models do not perform as accurately as in the single-step-ahead prediction case. Acceptable multi-step-ahead predictors for up to 240 msec prediction horizon have been obtained using actual traffic data.