leff@smu.UUCP (Laurence Leff) (01/15/89)
Subject: AI-Related Dissertations from SIGART No. 104, part 2 of 3 The following is a list of dissertation titles and abstracts related to Artificial Intelligence taken taken from the Dissertation Abstracts International (DAI) database. The list is assembled by Susanne Humphrey and myself and is published in the SIGART Newsletter (that list doesn't include the abstracts). The dissertation titles and abstracts contained here are published with the permission of University Microfilms International, publishers of the DAI database. University Microfilms has granted permission for this list to be redistributed electronically and for extracts and hardcopies to be made of it, provided that this notice is included and provided that the list is not sold. Copies of the dissertations may be obtained by addressing your request to: University Microfilms International Dissertation Copies Post Office Box 1764 Ann Arbor, Michigan 48106 or by telephoning (toll-free) 1-800-521-3042 (except for Michigan, Hawaii, and Alaska). In Canada: 1-800-268-6090. From SIGART Newsletter No. 104 File 2 of 3 Engineering to Linguistics ----------------------------------------------------------------------- AN University Microfilms Order Number ADG87-17025. AU PIETRANSKI, JOHN FRANK. IN Louisiana Tech University D.E. 1987, 282 pages. TI APPLICATION OF AN EXPERT FUZZY LOGIC CONTROLLER TO A ROTARY DRYING PROCESS. SO DAI v48(05), SecB, pp1440. DE Engineering, Chemical. AB An expert fuzzy logic controller has been developed that consists of both precise and imprecise state descriptions and decision rules. The expert fuzzy logic controller incorporates two contrasting types of controllers under an intelligent supervisory expert system which supervises control of a simulated industrial drying process. The rotary drying operation contains multiple process input and output variables. Some of the variables are nonlinear and exhibit interaction effects. Concurrently, the proposed control system uses "crisp logic" deterministic based controllers for typical and familiar control situations which are precise and well known, as well as the more complex "fuzzy logic" and fuzzy algorithms for situations that demand linguistic valued determinations. The control system structure is hierarchical in that the knowledge-based expert system determines the selection and sequence of the control rules utilized by the fuzzy logic controllers. The fuzzy logic controllers are themselves master controllers in a cascade configuration and drive the setpoint for their respective slave flowrate controllers. The slave controllers are represented by the crisp logic flow controllers. The control system configuration uses multiple predetermined rule sets for the fuzzy logic controllers. The selection of the appropriate rule set is determined on-line by the expert system which is designated the expert fuzzy supervisor. Supervisory control of the process is performed using a five task algorithm. The tasks include interpretation of the process data, evaluation of the current state attributes, comparison of the current state and goal state attributes, determination of requisite process setpoint changes, and selection and sequence of valid operations which are to be executed. A major advantage of the configuration used in this study is the control system's ability to automatically drive the process to selected steady-states as determined by the rules and facts within the knowledge-base expert-system supervisor. In addition, this work demonstrates that an overall objective for a typical industrial operation, such as the rotary drying process, can be designed and implemented using intelligent supervisory control in the decision-making process. In order to be successful in realizing the control objective, a standard decision-making procedure that incorporates the desired results and anticipated problem situations into an agreeable knowledge-base supervisory program was developed. The expert fuzzy logic controller responded well to changes in the requirements for the process attributes. AN University Microfilms Order Number ADG87-12265. AU SIRCAR, JAYANTA KUMAR. IN University of Maryland Ph.D 1986, 534 pages. TI COMPUTER AIDED WATERSHED SEGMENTATION FOR SPATIALLY DISTRIBUTED HYDROLOGIC MODELING. SO DAI v48(05), SecB, pp1457. DE Engineering, Civil. AB Physically based hydrologic models that simulate streamflow in terms of watershed characteristics are important tools in the water resource decision making process. The parameters for physically based models in current use are defined as averages for the watersheds involved. A logical step in improving the quality of the simulated streamflows would be the development of practical, spatially distributed models that are capable of simulating the hydrologic consequences of physical variations within watersheds. A key element in spatially distributed hydrologic simulation is the interdependence between runoff and a three-dimensional flow conveyance network defined by the topography of the watershed. While remote sensing can provide an adequate definition of land cover and, in some cases, even the soil type and soil moisture, an efficient means of defining elements of the topographic network and related catchment areas must be available if practical versions of spatially distributed hydrologic models are to be realized. If the models are to be available for general use on larger watersheds, this topographic analysis must be computer assisted using digital format data extracted from hard copy maps. The objective of the present dissertation is to use a binary image to: (1) define the elevation, slope and aspect of any point on a map surface; and (2) delineate those catchments that drain into any user defined channel segment or point along the channel system. The technique developed is a hardware/software system that uses a graph-theoretic approach to represent and manipulate raster scanned digital contours in a computer. Segmentation of the watersheds is accomplished through the development of a set of "expert heuristics" that simulate the manual operations of subbasin delineation on maps. The designed system includes components that: (1) label the digitized contour traces output from a scanner with an elevation attribute; (2) create a corresponding digital matrix of elevations that describe the spatial variation of topography; (3) register digitized drainage networks to the elevation database using a geographic information system framework; and (4) delineate the basin or subbasins that contribute runoff to user defined points or reaches along the channel network. (Abstract shortened with permission of author.). AN University Microfilms Order Number ADG87-23098. AU THURSTON, HOWARD MICHAEL. IN Stanford University Ph.D 1987, 251 pages. TI AN EXPERT-KNOWLEDGE-BASED APPROACH TO BUILDING SEISMIC PERFORMANCE EVALUATION. SO DAI v48(07), SecB. DE Engineering, Civil. AB Evaluating the seismic performance of buildings often requires substantial engineering expertise gained from technical training and accumulated practical experience. The research work described herein investigates the applicability of expert-knowledge-based techniques in evaluating the seismic performance of buildings. Expert-knowledge-based techniques are defined as those which are critically dependent upon expertise, expert opinion, or rules-of-thumb in order to achieve satisfactory results. There is considerable data available on the seismic performance of unreinforced masonry buildings. These buildings generally tend to perform very poorly when subjected to seismic forces. However, certain buildings in this category have characteristics which greatly improve their seismic resistance. By contrast, an important category of building which lacks meaningful seismic performance data is tilt-up construction. Furthermore, modern tilt-up buildings are being built for retail and commercial applications. There is concern among some members of the engineering community about the seismic performance of some new types of tilt-up buildings. Expert-knowledge-based damage evaluation methodologies are presented for multistory unreinforced masonry and tilt-up buildings. These methodologies are building-specific, and are tailored to best represent the damage data and evaluative expertise available for each building type. Finally, a general framework is developed to aid in designing an expert-knowledge-based methodology for evaluating any building. The methodology for unreinforced masonry buildings is valuable in making the characteristics which are critical to performance of these structures explicit. Although developed on the basis of damage data from Mainland China, the methodology could be modified to more closely reflect the damage data and construction practices in other countries or regions. The tilt-up methodologies show good agreement with United States data for such structures. Expert-knowledge-based methodologies have the potential to be useful tools in the hands of knowledgeable professionals in understanding the damage-ability of buildings subjected to seismic loads. In addition, these methodologies make the expertise used in the evaluation process explicit, thereby aiding in understanding how engineering experts make such evaluations. AN University Microfilms Order Number ADG87-25105. AU BABA, MUTASIM FUAD. IN Virginia Polytechnic Institute and State University Ph.D 1987 179 pages. TI INTELLIGENT AND INTEGRATED LOAD MANAGEMENT SYSTEM. SO DAI v48(08), SecB. DE Engineering, Electronics and Electrical. AB The design, simulation and evaluation of an intelligent and integrated load management system is presented in this dissertation. The objective of this research was to apply modern computer and communication technology to influence customer use of electricity in ways that would produce desired changes in the utility's load shape. Peak clipping (reduction of peak load) using direct load control is the primary application of this research. The prototype computerized communication and control package developed during this work has demonstrated the feasibility of this concept. The load management system consists of a network of computers, data and graphics terminals, controllers, modems and other communication hardware, and the necessary software. The network of interactive computers divide the responsibility of monitoring of meteorological data, electric load, and performing other functions. These functions include: data collection, processing and archiving, load forecasting, load modeling, information display and alarm processing. Each of these functions requires certain amount of intelligence depending on the sophistication and complication of that function. Also, a high level of reliability has been provided to each function to guarantee an uninterrupted operation of the system. A full scale simulation of this concept was carried out in the laboratory using five microcomputers and the necessary communication hardware. An important and integral part of the research effort is the development of the short-term load forecast, load models and the decision support system using rule-based algorithms and expert systems. Each of these functions has shown the ability to produce more accurate results compared to classical techniques while at the same time requiring much less computing time and historical data. Development of these functions has made the use of microcomputers for constructing an integrated load management system possible and practical. Also, these functions can be applied for other applications in the electric utility industry and maintain their importance and contribution. In addition to that, the use of rule-based algorithms and expert systems promises to yield significant benefits in using microcomputers in the load management area. AN This item is not available from University Microfilms International ADG05-61010. AU DA MOTA TENORIO, MANUEL FERNANDO. IN University of Southern California Ph.D 1987. TI PARALLEL PROCESSING TECHNIQUES FOR PRODUCTION SYSTEMS. SO DAI v48(07), SecB. DE Engineering, Electronics and Electrical. AB Production systems static and dynamic characteristics are modeled with the use of graph grammar, in order to create means to increase the processing efficiency and the use of parallel computation through compile time analysis. The model is used to explicate rule interaction, so that proofs of equivalence between knowledge bases can be attempted. Solely relying on program static characteristics shown by the model, a series of observations are made to determine the system dynamic characteristics and modifications to the original knowledge base are suggested as a means of increasing efficiency and decreasing overall search and computational effort. Dependences between the rules are analyzed and different approaches for automatic detection are presented. From rule dependences, tools for programming environments, logical evaluation of search spaces and Petri net models of production systems are shown. An algorithm for the allocation and partitioning of a production system into a multiprocessor system is also shown, and addresses the problems of communication and execution of these systems in parallel. Finally, the results of a simulator constructed to test several strategies, networks, and algorithms are presented. (Copies available exclusively from Micrographics Department, Doheny Library, USC, Los Angeles, CA 90089-0182.). AN University Microfilms Order Number ADG87-22633. AU GRIER, JAMES THOMAS. IN The Union for Experimenting Colleges and Universities PH.D 1986, 76 pages. TI A PROCEDURAL MODEL IN A KNOWLEDGE SYSTEM OF A GENERALIZED INTELLIGENT DECISION SUPPORT SYSTEM WHICH EMPLOYS PSYCHOLOGICAL AND BIOLOGICAL CHARACTERISTICS. SO DAI v48(07), SecB. DE Engineering, Electronics and Electrical. AB In the past decade there has been a notable increase in the use of computer-based information systems to support decision-making in organizations. Much of the research on models of Decision Support Systems (DSS) has been concerned with organizational characteristics, computer systems (hardware), or information structure. Very few models have been designed to integrate a persons psychological and biological characteristics of the decision maker into the decision-making process. This dissertation creates a structured example which could contain the above characteristics. The research describes a framework of a knowledge-based Decision Support system which has a Management Information System (MIS) as a subset. The knowledge system model would contain individual behavioral characteristics of the user (e.g. cognitive, personalities, attitudes, biorhythms (emotional, physical, intellectual), values, and other variables which might be predictions of what the user's decision patterns might be). These user characteristics might impinge upon the decision makers ability as an information processor and have the potential to affect the quality of decisions. The model is designed to integrate recent research done by behaviorists on the behavioral aspects of the user and his/her interface with information system and the system's interaction with the organizational context and decision environment. AN University Microfilms Order Number ADG87-18499. AU HABERSTROH, RICHARD. IN Polytechnic University Ph.D 1987, 183 pages. TI GRAECO-LATIN SQUARES AND ASSOCIATED METHODS FOR THE PROBLEM OF LINE DETECTION IN DIGITAL IMAGES. SO DAI v48(05), SecB, pp1464. DE Engineering, Electronics and Electrical. AB In this dissertation the four-way experimental design known as the Graeco-Latin square (GLS) is used as a basis for robust statistics for the detection of lines of elevated grey level intensity in a digital image with a noisy and/or structured background. All of the methods that are developed use a 5 x 5 pixel mask analyzed as a nonrandomized GLS. The statistical tests which comprise the detector algorithms are unbiased for certain families of narrow, straight lines of arbitrary orientation, although four natural directions produce the highest power of test. The fundamental problem of line detection by means of ANOVA methods is developed using one-way analysis of variance (ANOVA), while two-way ANOVA illuminates the general advantages of multiple-way ANOVA techniques when the background of the image contains some unknown type of structure. These techniques are extended to the GLS with the algorithms described in detail. The basic GLS line detectors are improved upon by a "reduced" GLS detector, in which the alternative subspace is restricted to more closely correspond to the line targets of interest. Special methods for the suppression of false alarms due to edges in the image are proposed, which include an implied shape test of the treatment means directly in the test statistic. An adaptive GLS procedure is also discussed, in which the number of "ways" of the final test is based upon the information in the basic statistics coming out of the GLS. The problem of the estimation and removal of background structure leads to the introduction of some modified analysis of covariance (ANCOVA) models and their application to the line detection problem. It is shown how such models can completely remove image background structure of the form of an intensity plane with an arbitrary gradient. The methods of edge suppression are adapted for use with this model, and the ANCOVA model is also applied to images containing correlated noise. It is demonstrated how some elements of the structure induced by the correlation can be estimated and removed by employing the ANCOVA model. AN University Microfilms Order Number ADG87-21406. AU HSYUNG, NIEY-BOR. IN The University of Iowa Ph.D 1987, 239 pages. TI MORPHOLOGICAL TEXTURE ANALYSIS. SO DAI v48(07), SecB. DE Engineering, Electronics and Electrical. AB A new strategy has been developed for the analysis of texture called morphological texture analysis. The morphological texture is defined as "the pattern of the gray level distribution within the boundary of the field of view". This new strategy was developed from synthesizing three distinct disciplines: statistical textural features, symmetry operations, and mathematical methods including variational principles and fabric tensors. This methodology is superior to currently favored methods of texture analysis--which are structural and statistical approaches--because it analyzes texture without assuming that the image is periodic and because it is independent of neighborhood gray levels. In addition in this new strategy, there is a major contribution in the mathematical representation of morphological texture. This representation is based on the gray level distribution G(r,$\theta$) and boundary conditions of the image. These have been derived using two different methods, each of which produces the same result in the form of a Bessel-Fourier function. Statistical textural features are calculated based on the variations of the gray level from point to point along the r and $\theta$ directions of the field of view. The symmetry operation, which is used to characterize the picture, can be classified as two different types: rotational symmetry and translational symmetry. These two symmetries can be obtained directly from the coefficients of the Bessel-Fourier function. Some training set (templates) were tested by employing the Shape Analyzer$\sp{\rm TM}$, which is a 68000 UNIX-based system programmable in "C" language. A restoration technique was used in this research in order to be assured that the scanning process is independent of the illumination system. Some applications have been reported to demonstrate the potential and utility of morphological texture analysis; these include classification of natural pictures, interpretation of images, and differentiation of biological cells. An experimental comparison of human vision and the computerized textural feature has been tested in this research. Classification of Brodatz's textures, which are based on morphological textural parameters has also been done. Results indicate that natural images can be recognized by this new method with 90% accuracy. AN University Microfilms Order Number ADG87-15777. AU JOHNSON, KENNETH. IN Clarkson University Ph.D 1987, 175 pages. TI THE BIOLOGICAL IMPLEMENTATION OF NONLINEAR SPACE-TIME CODING IN THE RETINA AND ITS APPLICATION TO THE LOW LEVEL REPRESENTATION OF VISUAL INFORMATION IN MACHINE VISION SYSTEMS. SO DAI v48(06), SecB, pp1771. DE Engineering, Electronics and Electrical. AB In this dissertation a model for the processing and representation of early visual information is proposed. The results fall into four categories. First, a new theory describing the functional nature of neural coding in the retina is proposed. Second, the surround delay is incorporated into the difference of Gaussian model. Third, a discrete version of the model is developed. Fourth, the implications of these findings on the structure and function of simple cells in the visual cortex are considered. A thorough review of the eye's projection characteristics and the sampling of the retinal image by the cone mosaic is presented. A theory of processing in the retina that relates the temporal behavior of the bipolar responses to the opponent coding of chromatic information in ganglion cells is introduced. The theory indicates that what are interpreted as X and Y type responses are in reality chromatically dependent responses. The proposed mechanisms are used to explain how color is perceived in the temporally modulated achromatic stimulus of Festinger et al. (1971). The importance of stimulus timing in generating the different colors is verified by the inner layer timing results of Sugawara (1985). The delayed surround of Werblin and Dowling (1969b) is incorporated into Rodieck's (1965) difference of Gaussian (DOG) model. The new model is analyzed in the image and frequency domains. The image domain representation shows that the receptive field is capable of representing both nonmoving and moving edges in its output. Frequency domain analysis shows that the behavior of the operator is identical to the recently documented spatio-temporally coupled responses of Dawis et al. (1984). A practical implementation of the receptive field model is developed for use on a digital imaging system. The response of the discrete operator to simple steps and ramps is demonstrated for linear and homomorphic processing schemes. Its relationship to optical flow and the responses of simple cells in the visual cortex is considered. It is shown that the output of the convolutional operator proposed here is more compatible with the responses of the simple cells than the proposals made by Marr (1982). (Abstract shortened with permission of author.). AN This item is not available from University Microfilms International ADG05-60749. AU LIU, ZHI-QIANG. IN University of Alberta (Canada) Ph.D 1987. TI ASPECTS OF IMAGE PROCESSING IN NOISY ENVIRONMENTS: RESTORATION, DETECTION, AND PATTERN RECOGNITION. SO DAI v48(05), SecB, pp1468. DE Engineering, Electronics and Electrical. AB In one sense, the aim of computer vision is to extract and interpret the organization of images in ways which are useful in specific problem domains. However, such images taken in realistic environments inevitably suffer from various degradations including noise, blurring, geometrical distortion, etc.. These degradations pose many problems for implementing computer vision techniques. An extensive literature review has identified the interesting areas requiring further study. Of these areas, nonstationary image restoration, and image signal detection in noisy environments are pursued extensively in this thesis. Four processing algorithms and techniques have been developed in the framework of stochastical estimation theory. In order to obtain perceptually more satisfying restored images, image fields are assumed to be nonstationary. Two restoration algorithms are derived, namely, a sequential Kalman filter and an adaptive iterative filter. The sequential filter is developed based on a causal state-space image model. The adaptive filter uses a modified received image model. Both algorithms include some local spatial activity measurements in the filter gains such that the restored images retain edge information. Simulations show that the visual quality of the restored images is significantly improved. Matched filters have long been used for signal detection. They are optimal only when the noise is stationary and white. In order to effectively apply the matched filters to images embedded in nonstationary, nonwhite noise backgrounds, a new adaptive postfiltering technique is derived. Superiority of this technique is shown by experiments. In the fourth algorithm a hierarchical approach for multi-object detection is presented. In this approach, the detection of object is divided into three steps: prefiltering, pattern recognition, and detection. The images are prefiltered to suppress noise which would otherwise affect the entire detecting operation. In pattern recognition, a linear least square mapping technique is applied for classification. In the detection part, a known object of a class is used to match the received image of the same class. It is shown by simulations that with this approach, the computation time is reduced by more than 50%. Finally, suggestions for further research in the related areas are also included. AN University Microfilms Order Number ADG87-24311. AU MEADOR, JACK LEWIS. IN Washington State University Ph.D 1987, 144 pages. TI PATTERN RECOGNITION AND INTERPRETATION BY A BINARY-TREE OF PROCESSORS. SO DAI v48(08), SecB. DE Engineering, Electronics and Electrical. AB High-level-language-architecture research is an area which focuses upon narrowing the so-called "semantic gap" between high-level programming languages and computer hardware. The goal is to improve processing efficiency by bringing hardware and software design closer together. This goal is typically achieved by considering software and hardware aspects concurrently as part of an overall design process. A variety of approaches exist within this area, ranging from machines having optimized instruction sets to direct execution architectures where high-level tokens are fetched from program memory like low-level instructions. A key aspect of any high-level-language architecture is that the execution algorithm can be modeled as language translation. Any high-level-language architecture is effectively a direct implementation of an interpreter. A large numer of multiprocessor organizations exist today. A fundamental problem of multiprocessing is becoming less one of how to physically organize the processor, and more one of how to program it. The difficulty associated with programming multiprocessors is characterized here as a "parallel semantic gap". The research described within is motivated by the direct interpretation model used to narrow the sequential semantic gap. The direct implementation of an interpreter on some multiprocessor organization is proposed. The specific approach is to study syntax-directed interpretation on a binary-tree multiprocessor organization. Any interpretation scheme must use some pattern recognition algorithm to discern the actions that programs are to carry out. This dissertation presents two new recognition algorithms for a binary-tree multiprocessor and studies the application of these algorithms to parallel interpretation. Language interpretation is not the only application which these algorithms have. Compelling research directions are suggested for architectures supporting expert systems and complex pattern analysis. Included among these are machines for information retrieval from a semantic-network knowledge base and ones which perform scene analysis by detecting graph isomorphisms. AN University Microfilms Order Number ADG87-25190. AU MIYAHARA, SHUNJI. IN University of Pennsylvania Ph.D 1987, 198 pages. TI AUTOMATED RADAR TARGET RECOGNITION BASED ON MODELS OF NEURAL NETS. SO DAI v48(08), SecB. DE Engineering, Electronics and Electrical. AB This dissertation describes a new approach to target recognition, using radar returns and parallel processing based on models of neural networks. Target recognition usually consists of three steps: data acquisition, data representation (generation of feature vectors) and data classification. Two methods of target recognition are proposed and several results from their study are discussed. The methods are: (1) The use of sinogram representations as learning set in associative memory, based on models of neural nets as classifier. Such memories are known to be robust and fault tolerant, and (2) use of polarization representation for use in neural net based associative memory as a classifier. The advantages of these methods are: (1) they represent a new approach to signal processing and target recognition, (2) they have the potential to identify targets from small fractions of the target data (robustness), (3) they are insensitive to slight degradation in system hardware (fault-tolerance), (4) they can identify targets automatically by generating identifying labels or symbols, and (5) they can be implemented efficiently using optical hardware, since optics provides the parallelism and massive interconnectivity required in neural net implementations of associative memory employing parallel processing. Using microwave scattering data of scaled model targets, the concepts for the target recognition were demonstrated by computer simulation of a 1024 (32 $\times$ 32) element neural net associative memory based on the so called outer product model. The simulations show that partial input, consisting of less than 10% of the total information can identify the targets. This result illustrates the robustness of associative memory and the potential usefulness of the approach. 2-D optical implementations of a neural net of 8 $\times$ 8 (= 64) binary neurons were studied. Fault tolerance and robustness are examined, using a four dimensional (8 $\times$ 8 $\times$ 8 $\times$ 8) clipped outer product ternary $T\sb{ijkl}$ mask to establish the weighted interconnections of the net and electronic feedback based on closed loop TV system. The performance was found to be in agreement with that of computer simulation, even though aberration of lenses and the defects of the system, were present. These results confirm the practical suitability of the opto-electronic approach to the neural net implementation and pave the way for the implementation of larger networks. AN University Microfilms Order Number ADG87-19949. AU REBER, WILLIAM LOUIS, JR. IN University of California, Los Angeles Ph.D 1987, 195 pages. TI ARTIFICIAL NEURAL SYSTEM DESIGN: ROTATION AND SCALE INVARIANT PATTERN RECOGNITION. SO DAI v48(06), SecB, pp1774. DE Engineering, Electronics and Electrical. AB The design of an artificial neural system (ANS) for generalizing the recognition of two dimensional patterns having variations in rotation and scale is demonstrated using alpha-numeric characters. An ANS is a multi-layer network of parallel computational structures. These structures are based upon large numbers of artificial neurons or processing elements organized into disjoint sets or slabs having the interconnections and behavior appropriate for performing specific computational tasks. They provide the ANS with a parallel processing capability for performing signal processing, image processing, and various adaptive learning algorithms. An ANS design methodology is defined and used for developing the experimental ANS design. It consists of the functional specification of the ANS, the design of the ANS and its parallel computational structures, and the verification of the ANS. The experimental ANS design consists of a hierarchy of slabs for display of input patterns, for pattern transformation, for pattern classification, and for display of the ANS output. A sequence of pattern transformations including the log, polar, and 1-dimensional Fourier transforms, are developed for producing a generalized pattern representation for rotation and scale changes. The pattern transformations are embedded within the ANS, and implemented by designing appropriate parallel computational structures for each transformation. All of the necessary computations are performed in parallel by the artificial neural system. Mapping functions and behavior equations are defined for specifying the necessary processing element interconnections and behavior. The generalized pattern representations for a typical set of alpha-numeric characters are used for training a nearest-neighbor classifier. Computer simulations are used to verify the performance of the ANS. Results demonstrate rotation, scale, and both rotation and scale invariant pattern classification. AN This item is not available from University Microfilms International ADG05-61056. AU WEBER, ALLAN GUY. IN University of Southern California Ph.D 1987. TI HIERARCHICAL TEXTURE SEGMENTATION USING MULTIPLE RESOLUTION OPERATORS. SO DAI v48(07), SecB. DE Engineering, Electronics and Electrical. AB Texture information can be used as a basis for segmenting an image into regions. Image texture is basically a local area phenomenon that is sensitive to the size of the area. What appears as a non-textured area at one resolution level can appear as a region with distinctive texture at a different resolution. The performance of texture segmentation schemes is often highly dependent on the size of the local area operator used to generate the classification features. The size of the operators has a major impact on the performance near the boundaries between texture regions. Features based on large operators perform better overall but are highly affected by the mixing of class statistics when the operator overlaps more than texture, such as near the texture boundaries. Features based on small operators show poorer overall performance but are more likely to maintain an acceptable performance level in the boundary areas. The trade-off is between statistical accuracy of the classification versus the final accuracy of the texture region boundaries. The problem being studied here is how to combine information from texture classifiers operating at different resolutions into a segmentation process that gives acceptable performance in all areas of an image. In this study, the nature of the mixing problem is examined and a solution is proposed based on using multiple resolution features in a hierarchical decision process. A key component of the solution is an analysis of the image data prior to performing any classification. From this analysis, we determine the expected location in the feature space of the mixture points. Three different methods of isolating the mixture points in the feature space are proposed and tested. During the initial classification phase, image points that are within the mixture areas are left unclassified. Spatial information is incorporated into the segmentation process by the use of a local area cohesion operation. The final segmentation is based on a hierarchical decision process that uses the classification choice at both resolutions and the spatial cohesion data. Several decision processes are tested that use the information in different ways. (Copies available exclusively from Micrographics Department, Doheny Library, USC, Los Angeles, CA 90089-0182.). AN University Microfilms Order Number ADG87-22405. AU HENNINGSEN, JACQUELINE R. IN The University of Nebraska - Lincoln Ph.D 1987, 190 pages. TI A CONCEPTUAL FRAMEWORK FOR DISTRIBUTED EXPERT SYSTEM USE IN TIME SENSITIVE HIERARCHICAL CONTROL. SO DAI v48(07), SecB. DE Engineering, Industrial. AB There are many problems faced by decision makers involved in complex, time sensitive hierarchical control systems. These may include maintaining knowledge of the functional status of the system components, forecasting the impact of past and future events, transferring information to a distant or poorly connected location, changing the requirements for an operation according to resources available, or creating a independent course of action when system connectivity falls. These problems are transdisciplinary in nature, so decision makers in a variety of organizations face them. This research develops a framework for the use of distributed expert systems in support of time sensitive hierarchical control systems. Attention is focused on determining ways to enhance the likelihood that a system will remain functional during a crisis in which one or more of the system nodes fail. Options in the use of distributed expert systems for this purpose are developed following investigation of related research in the areas of cooperative and distributed systems. A prototype under development of a generic system model called DES (distributed expert systems) is described. DES is a "trimular" form of support structure, where a trimule is defined to be a combination of a human decision agent, a component system model and an expert system. This concept is an extension of the domular theory of Tenney and Sandell (1981). AN University Microfilms Order Number ADG87-20509. AU KHAJENOORI, SOHEIL. IN University of Central Florida Ph.D 1986, 219 pages. TI COMPUTER-ASSISTED DESIGN AND ASSEMBLY OF STANDARDIZED MODULES USING A ROBOT MANIPULATOR. SO DAI v48(06), SecB, pp1777. DE Engineering, Industrial. AB Graphical Motion to Robot Motion Translator (GMRMT), a new system for the automatic generation of robot motion control programs, is presented. The GMRMT system is an interactive, menu-driven software package which allows the design and subsequent assembly of three-dimensional structures built from standardized component modules to be accomplished using a common database. The standardized component modules used as example building blocks in the project are rectangular solids of several sizes. Presented are an object placement-sequencer algorithm, a height specification and interference checking algorithm, and a balance-checking algorithm. To avoid the creation of dynamic obstacles in the assembly robot's motion paths, and possible collision and interference of the robot arm with these obstacles, the proper sequencing of the blocks in the design database is essential. The object placement-sequencer algorithm is responsible for proper sequencing of the blocks in the design database prior to the building of the designed structure by the robot. The height specification and interference checking algorithm automatically generates the proper positioning of a block in the design by performing a sequential search over the accumulated design structure. The stacking feasibility of the blocks in the design is verified by the balance-checking algorithm, prior to the acceptance of the block as a permanent part of the design. When the structure design has been completed, it may be visualized using interactive computer graphics techniques. Then, upon the user's request, the system will produce the "VAL" robot motion control language program necessary to construct the designed structure and download the program to the robot controller for execution. AN University Microfilms Order Number ADG87-24899. AU CRANE, CARL DAVID, III. IN The University of Florida Ph.D 1987, 193 pages. TI MOTION PLANNING AND CONTROL OF ROBOT MANIPULATORS VIA APPLICATION OF A COMPUTER GRAPHICS ANIMATED DISPLAY. SO DAI v48(08), SecB. DE Engineering, Mechanical. AB It is often necessary in a hazardous environment for an operator to effectively control the motion of a robot manipulator which cannot be observed directly. The manipulator may be either directly guided via use of a joystick or similar device, or it may be autonomously controlled, in which case it is desirable to preview and monitor robot motions. A computer graphics based system has been developed which provides an operator with an improved method of planning, evaluating, and directly controlling robot motions. During the direct control of a remote manipulator with a joystick device, the operator requires considerable sensory information in order to perform complex tasks. Visual feedback which shows the manipulator and surrounding workspace is clearly most important. A graphics program which operates on a Silicon Graphics IRIS workstation has been developed which provides this visual imagery. The graphics system is capable of generating a solid color representation of the manipulator at refresh rates in excess of 10 Hz. This rapid image generation rate is important in that it allows the user to zoom in, change the vantage point, or translate the image in real time. Each image of the manipulator is formed from joint angle datum that is supplied continuously to the graphics system. In addition, obstacle location datum is communicated to the graphics system so that the surrounding workspace can be accurately displayed. A unique obstacle collision warning feature has also been incorporated into the system. Obstacles are monitored to determine whether any part of the manipulator comes close or strikes the object. The collision warning algorithm utilizes custom graphics hardware in order to change the color of the obstacle and produce an audible sound as a warning if any part of the manipulator approaches closer than some established criterion. The obstacle warning calculations are performed continuously and in real time. The graphics system which has been developed has advanced man-machine interaction in that improved operator efficiency and confidence has resulted. Continued technological developments and system integration will result in much more advanced interface systems in the future. AN University Microfilms Order Number ADG87-19113. AU LI, CHING. IN The University of Wisconsin - Madison Ph.D 1987, 170 pages. TI ON-LINE BEARING CONDITION MONITORING BY PATTERN RECOGNITION AND MATCHED FILTERING. SO DAI v48(08), SecB. DE Engineering, Mechanical. AB This thesis presents an on-line bearing condition monitoring system which imparts the following three newly developed signal processing algorithms into a digital computer: bearing localized defect detection/diagnosis scheme, in-process defect induced resonance self-learning scheme and the matched filter. Thus, the capability of automatically detecting/diagnosing the presence of localized defects in a given bearing system as well as evaluating the extent of bearing damage may be built into a digital computer. For automatic detection/diagnosis of bearing localized defects, a pattern recognition analysis technique has been developed for analyzing vibration signatures carrying the necessary information about the defect induced vibration bursts. Extracting normalized and dimensionless features by short-time signal processing techniques, linear discriminant functions discriminatory information provided is 38% more accurate than the best analysis procedure presently available. In order to on-line identify the resonances excited by the impulse produced by a damaged bearing for a given bearing system without relying on its historical records and human intelligence, an in-process defect induced resonance self-learning scheme was developed. Pinpointing every instant that a roller strikes a localized defect, two segments of subsequences may be formed from just one piece of bearing vibration signature. A template learning ========================================================================= Date: 10 March 1988, 21:30:13 CST