Thesis (Ph. D.)--University of Texas at Austin, 2005. / Vita. Includes bibliographical references.

In recent years, hardware-in-the-loop (HIL) simulation has assumed a prominent role in the vehicle development process. A physical part, which may be a prototype at any stage of development, is tested, while the rest of the vehicle is represented by a mathematical model. Vehicle models used with hardware-in-the-loop must be capable of simulating an event in less time than it takes the event to occur in reality. Fast simulation necessitates a model that is represented by very efficient simulation code. This thesis presents a procedure for automatically generating this simulation code, given a description of the vehicle as input. For this work, a symbolic formulation procedure based on linear graph theory and the principle of orthogonality is used to generate governing equations for vehicle systems; this procedure forms the basis of the DynaFlexPro software package. In order to generate simulation code for vehicle dynamics studies, the DynaFlexPro component model template was extended to include rules for calculating intermediate variables and rules for calling external functions. These changes enabled the development of a tire component model, known as DynaFlexPro/Tire, that adds critically important (and computationally efficient) blocks to the overall vehicle simulation code. The combination of DynaFlexPro and DynaFlexPro/Tire allows analysts to construct a model for any vehicle topology and gives analysts great freedom to define how tire forces and moments will be calculated. Simulation code describing the vehicle model is automatically generated using symbolic computing techniques. The accuracy of the approach was validated by comparing results for DynaFlexPro vehicle models to results for equivalent models developed in a well-established tool for vehicle dynamics simulation (MSC.ADAMS). Two different vehicle models were constructed using DynaFlexPro and DynaFlex- Pro/Tire: a generic 4-wheeled vehicle with independent suspension and an articulated forestry skidder. Both models had an open-loop topology. When appropriate modeling variables were selected, each model was described by a minimal set of ordinary differential equations (ODEs) and the simulation code generated by DynaFlexPro was capable of being used for hardware-in-the-loop applications; the braking and handling behavior of the example models was simulated faster than real-time on a desktop PC with a 3.2 GHz Pentium 4 processor and 1 GB of RAM. For the same vehicle models, a different choice of modeling variables resulted in a mixed set of differential and algebraic equations (DAEs); in that case, HIL-capable simulation code could not be consistently generated. The approach works well for vehicle models described by ODEs, but more research is needed into the treatment of DAEs for real-time simulation of vehicle dynamics.

Multiobjective optimization is a branch of mathematical programming for modelling problems with multiple conflicting objectives. Multiobjective optimization problems can be solved using Pareto optimization techniques including evolutionary multiobjective optimization algorithms. Many real world applications involve multiple objective functions and can be addressed within a multiobjective optimization framework. Multiobjective optimization methods allow exploration of the attainable values of the objective functions and trade-offs between objective functions without soliciting preference information from the decision maker(s) before potential solutions are presented. In order to be sufficiently representative of the possibilities and trade-offs, the results of multiobjective optimization may be too numerous or complex in shape for decision makers to reasonably consider. Previous approaches to this problem have aimed to reduce the solution set to a smaller representative set. The methodology developed and evaluated in this thesis employs hierarchical cluster analysis to organize the solutions from multiobjective optimiation into a tree structure based on their objective function values. Unlike previous approaches none of the solutions are removed from consideration before being presented to the decision makers. A hierarchical cluster structure is desirable since it presents a nested organization of the plans which can be used in decision making as shown in an example decision. The resulting dendrogram is a tree of clusters that can be used to see the attainable trade-offs on the Pareto front. As well, it can be used to interactively reduce the set of solutions under consideration or consider several subsets of solutions that lie in different regions of the Pareto front. A land use change problem in an urban fringe area in Southern Ontario, Canada is used as motivation and as an example application to evaluate the proposed methodology. Relevant literature in planning support systems is reviewed in order to focus the methodology on the application. The multiobjective optimization problem for this application was formulated and analyzed by Roberts (2003); the optimization algorithm used to generate the approximation of the optimal solutions is the Non-dominated Sorting Genetic Algorithm II, NSGA-II, developed by Deb et al. (2002). Future work will link the resulting objective function-based tree to map visualizations of the landscape under consideration. Decision makers will be able to use the tree structure to explore different potential land use plans based on their performance on the objective functions representing the quality of those plans for natural and human uses. This approach is applicable to multiobjective problems with more than three objective functions and discrete decision variables or hierarchically clustered Pareto optimal sets. The suitability for reuse with other datasets or other applications is discussed as well as the potential for inclusion in a decision support system (DSS).

The goal of this research is to develop a model-based tracking framework with biomedical imaging applications. This is an interdisciplinary area of research with interests in machine vision, image processing, and biology. This thesis presents methods of image modeling, tracking, and data association applied to problems in multi-cellular image analysis, especially hematopoietic stem cell (HSC) images at the current stage. The focus of this research is on the development of a robust image analysis interface capable of detecting, locating, and tracking individual hematopoietic stem cells (HSCs), which proliferate and differentiate to different blood cell types continuously during their lifetime, and are of substantial interest in gene therapy, cancer, and stem-cell research. Such a system can be potentially employed in the future to track different groups of HSCs extracted from bone marrow and recognize the best candidates based on some biomedical-biological criteria. Selected candidates can further be used for bone marrow transplantation (BMT) which is a medical procedure for the treatment of various incurable diseases such as leukemia, lymphomas, aplastic anemia, immune deficiency disorders, multiple myeloma and some solid tumors. Tracking HSCs over time is a localization-based tracking problem which is one of the most challenging tracking problems to be solved. The proposed cell tracking system consists of three inter-related stages: i) Cell detection/localization, ii) The association of detected cells, iii) Background estimation/subtraction. that will be discussed in detail.

System design, parameter design, and tolerance design are the three stages of product or process development advocated by Genichi Taguchi. Parameter design, or robust parameter design (RPD), is the method to determine nominal parameter values of controllable variables such that the quality characteristics can meet the specifications and the variability transmitted from uncontrollable or noise variables is minimized for the process or product. Tolerance design is used to determine the best limits for the parameters to meet the variation and economical requirements of the design. In this thesis, response surface methodology (RSM) and nonlinear programming methods are adopted to integrate the parameter and tolerance design. The joint optimization method that conducts parameter design and tolerance design simultaneously is more effective than the traditional sequential process. While Taguchi proposed the crossed array design, the combined array design approach is more flexible and efficient since it combines controllable factors, internal noise factors, and external noise factors in a single array design. A combined array design and the dual response surface method can provide detailed information of the process through process mean and process variance obtained from the response model. Among a variety of cuboidal designs and spherical designs, standard or modified central composite designs (CCD) or face-centered cube (FCC) designs are ideal for fitting second-order response surface models, which are widely applied in manufacturing processes. Box-Behnken design (BBD), mixed resolution design (MRD), and small composite design (SCD) are also discussed as alternatives. After modeling the system, nonlinear programming can be used to solve the constrained optimization problem. Dual RSM, mean square error (MSE) loss criterion, generalized linear model, and desirability function approach can be selected to work with quality loss function and production cost function to formulate the object function for optimization. This research also extends robust design and RSM from single response to the study of multiple responses. It was shown that the RSM is superior to Taguchi approach and is a natural fit for robust design problems. Based on our study, we can conclude that dual RSM can work very well with ordinary least squares method or generalized linear model (GLM) to solve robust parameter design problems. In addition, desirability function approach is a good selection for multiple-response parameter design problems. It was confirmed that considering the internal noise factors (standard deviations of the control factors) will improve the regression model and have a more appropriate optimal solution. In addition, simulating the internal noise factors as control variables in the combined array design is an attractive alternative to the traditional method that models the internal noise factors as part of the noise variables. The purpose of this research is to develop the framework for robust design and the strategies for RSM. The practical objective is to obtain the optimal parameters and tolerances of the design variables in a system with single or multiple quality characteristics, and thereby achieve the goal of improving the quality of products and processes in a cost effective manner. It was demonstrated that the proposed methodology is appropriate for solving complex design problems in industry applications.

In recent years, hardware-in-the-loop (HIL) simulation has assumed a prominent role in the vehicle development process. A physical part, which may be a prototype at any stage of development, is tested, while the rest of the vehicle is represented by a mathematical model. Vehicle models used with hardware-in-the-loop must be capable of simulating an event in less time than it takes the event to occur in reality. Fast simulation necessitates a model that is represented by very efficient simulation code. This thesis presents a procedure for automatically generating this simulation code, given a description of the vehicle as input. For this work, a symbolic formulation procedure based on linear graph theory and the principle of orthogonality is used to generate governing equations for vehicle systems; this procedure forms the basis of the DynaFlexPro software package. In order to generate simulation code for vehicle dynamics studies, the DynaFlexPro component model template was extended to include rules for calculating intermediate variables and rules for calling external functions. These changes enabled the development of a tire component model, known as DynaFlexPro/Tire, that adds critically important (and computationally efficient) blocks to the overall vehicle simulation code. The combination of DynaFlexPro and DynaFlexPro/Tire allows analysts to construct a model for any vehicle topology and gives analysts great freedom to define how tire forces and moments will be calculated. Simulation code describing the vehicle model is automatically generated using symbolic computing techniques. The accuracy of the approach was validated by comparing results for DynaFlexPro vehicle models to results for equivalent models developed in a well-established tool for vehicle dynamics simulation (MSC.ADAMS). Two different vehicle models were constructed using DynaFlexPro and DynaFlex- Pro/Tire: a generic 4-wheeled vehicle with independent suspension and an articulated forestry skidder. Both models had an open-loop topology. When appropriate modeling variables were selected, each model was described by a minimal set of ordinary differential equations (ODEs) and the simulation code generated by DynaFlexPro was capable of being used for hardware-in-the-loop applications; the braking and handling behavior of the example models was simulated faster than real-time on a desktop PC with a 3.2 GHz Pentium 4 processor and 1 GB of RAM. For the same vehicle models, a different choice of modeling variables resulted in a mixed set of differential and algebraic equations (DAEs); in that case, HIL-capable simulation code could not be consistently generated. The approach works well for vehicle models described by ODEs, but more research is needed into the treatment of DAEs for real-time simulation of vehicle dynamics.

Multiobjective optimization is a branch of mathematical programming for modelling problems with multiple conflicting objectives. Multiobjective optimization problems can be solved using Pareto optimization techniques including evolutionary multiobjective optimization algorithms. Many real world applications involve multiple objective functions and can be addressed within a multiobjective optimization framework. Multiobjective optimization methods allow exploration of the attainable values of the objective functions and trade-offs between objective functions without soliciting preference information from the decision maker(s) before potential solutions are presented. In order to be sufficiently representative of the possibilities and trade-offs, the results of multiobjective optimization may be too numerous or complex in shape for decision makers to reasonably consider. Previous approaches to this problem have aimed to reduce the solution set to a smaller representative set. The methodology developed and evaluated in this thesis employs hierarchical cluster analysis to organize the solutions from multiobjective optimiation into a tree structure based on their objective function values. Unlike previous approaches none of the solutions are removed from consideration before being presented to the decision makers. A hierarchical cluster structure is desirable since it presents a nested organization of the plans which can be used in decision making as shown in an example decision. The resulting dendrogram is a tree of clusters that can be used to see the attainable trade-offs on the Pareto front. As well, it can be used to interactively reduce the set of solutions under consideration or consider several subsets of solutions that lie in different regions of the Pareto front. A land use change problem in an urban fringe area in Southern Ontario, Canada is used as motivation and as an example application to evaluate the proposed methodology. Relevant literature in planning support systems is reviewed in order to focus the methodology on the application. The multiobjective optimization problem for this application was formulated and analyzed by Roberts (2003); the optimization algorithm used to generate the approximation of the optimal solutions is the Non-dominated Sorting Genetic Algorithm II, NSGA-II, developed by Deb et al. (2002). Future work will link the resulting objective function-based tree to map visualizations of the landscape under consideration. Decision makers will be able to use the tree structure to explore different potential land use plans based on their performance on the objective functions representing the quality of those plans for natural and human uses. This approach is applicable to multiobjective problems with more than three objective functions and discrete decision variables or hierarchically clustered Pareto optimal sets. The suitability for reuse with other datasets or other applications is discussed as well as the potential for inclusion in a decision support system (DSS).

The goal of this research is to develop a model-based tracking framework with biomedical imaging applications. This is an interdisciplinary area of research with interests in machine vision, image processing, and biology. This thesis presents methods of image modeling, tracking, and data association applied to problems in multi-cellular image analysis, especially hematopoietic stem cell (HSC) images at the current stage. The focus of this research is on the development of a robust image analysis interface capable of detecting, locating, and tracking individual hematopoietic stem cells (HSCs), which proliferate and differentiate to different blood cell types continuously during their lifetime, and are of substantial interest in gene therapy, cancer, and stem-cell research. Such a system can be potentially employed in the future to track different groups of HSCs extracted from bone marrow and recognize the best candidates based on some biomedical-biological criteria. Selected candidates can further be used for bone marrow transplantation (BMT) which is a medical procedure for the treatment of various incurable diseases such as leukemia, lymphomas, aplastic anemia, immune deficiency disorders, multiple myeloma and some solid tumors. Tracking HSCs over time is a localization-based tracking problem which is one of the most challenging tracking problems to be solved. The proposed cell tracking system consists of three inter-related stages: i) Cell detection/localization, ii) The association of detected cells, iii) Background estimation/subtraction. that will be discussed in detail.

System design, parameter design, and tolerance design are the three stages of product or process development advocated by Genichi Taguchi. Parameter design, or robust parameter design (RPD), is the method to determine nominal parameter values of controllable variables such that the quality characteristics can meet the specifications and the variability transmitted from uncontrollable or noise variables is minimized for the process or product. Tolerance design is used to determine the best limits for the parameters to meet the variation and economical requirements of the design. In this thesis, response surface methodology (RSM) and nonlinear programming methods are adopted to integrate the parameter and tolerance design. The joint optimization method that conducts parameter design and tolerance design simultaneously is more effective than the traditional sequential process. While Taguchi proposed the crossed array design, the combined array design approach is more flexible and efficient since it combines controllable factors, internal noise factors, and external noise factors in a single array design. A combined array design and the dual response surface method can provide detailed information of the process through process mean and process variance obtained from the response model. Among a variety of cuboidal designs and spherical designs, standard or modified central composite designs (CCD) or face-centered cube (FCC) designs are ideal for fitting second-order response surface models, which are widely applied in manufacturing processes. Box-Behnken design (BBD), mixed resolution design (MRD), and small composite design (SCD) are also discussed as alternatives. After modeling the system, nonlinear programming can be used to solve the constrained optimization problem. Dual RSM, mean square error (MSE) loss criterion, generalized linear model, and desirability function approach can be selected to work with quality loss function and production cost function to formulate the object function for optimization. This research also extends robust design and RSM from single response to the study of multiple responses. It was shown that the RSM is superior to Taguchi approach and is a natural fit for robust design problems. Based on our study, we can conclude that dual RSM can work very well with ordinary least squares method or generalized linear model (GLM) to solve robust parameter design problems. In addition, desirability function approach is a good selection for multiple-response parameter design problems. It was confirmed that considering the internal noise factors (standard deviations of the control factors) will improve the regression model and have a more appropriate optimal solution. In addition, simulating the internal noise factors as control variables in the combined array design is an attractive alternative to the traditional method that models the internal noise factors as part of the noise variables. The purpose of this research is to develop the framework for robust design and the strategies for RSM. The practical objective is to obtain the optimal parameters and tolerances of the design variables in a system with single or multiple quality characteristics, and thereby achieve the goal of improving the quality of products and processes in a cost effective manner. It was demonstrated that the proposed methodology is appropriate for solving complex design problems in industry applications.

Vision and audition are the two best understood modalities which humans use to interact with the outside world. These modalities can provide highly precise spatial and temporal information. Thus, the field of human-computer interface design has focused much of their study and design on these modalities. On the other hand, the sense of touch has been largely ignored despite the fact that it is an essential part of human ability to interact with the environment. We are interested to identify key findings on how to use tactile technology effectively to design and fabricate a tactile interface. We intend to design a wearable tactile interface which can assist Unmanned Aerial Vehicles (UAV) operators in supervisory control and monitoring tasks. Tactile displays are usually comprised of vibratory stimulators which are arranged in specific formation based on the application of the display. Quantitative properties of a vibrating tactor which was used as the vibratory stimulator in our tactile interface were investigated and evaluated in this study. We executed a series of experiments to investigate the intensity of vibrations that the vibrating tactor can generate when it is being activated through different electrical signals. Driving signals were different in terms of waveform, frequency and amplitude. By applying the outcomes of our experiments, and using the available guidelines for the design of tactile displays, we proposed some methods for displaying flight dynamics (Roll, Pitch and Yaw) of a UAV through a tactile display which is structured in form of a vest. Due to the relative infancy of this branch of information presentation, and also the lack of thorough discussion within the scientific community we need to execute further experiments to evaluate the performance of the suggested tactile display.