Dana S. Nau - US grants

Hardware upgrades for LISP machines, and display equipment, will be provided for researchers at the University of Maryland for research in the Department of Computer Science. This equipment is provided under the Instrumentation Grants for Research in Computer and Information Science and Engineering program. The research for which the equipment is to be used will be in the areas of solid modeling, feature-based modeling, and artificial intelligence reasoning for process planning.

Efficient hierarchical planning, based on a task network, is the topic of this research. Goal and subgoal interactions will be investigated to achieve improved control strategies for planning. For example, it should be possible to improve both the quality of the plan and the efficiency of the planning process by developing a control strategy which selects among alternative task reductions in such a way as to avoid harmful interactions while encouraging helpful interactions. The following specific issues are central to this investigation: (1) how the various properties of interactions can guide the choice among alternative reductions in a task network; (2) how to improve the consistency of task network representations, in particular, how to identify situations where a seemingly unresolvable conflict can be resolved at a lower level in the task network; (3) how to take advantage of the fact that some planning tasks, while not independent in general, become independent once certain other plannng decisions have been made. The investigators anticipate that completion of the above research topics will result in significantly improved planning in a number of environments. To test this idea, they will implement a planning system incorporating the above ideas, and evaluate its performance on a number of domains.

Decisions made during the design of a machined part can significantly affect the product's cost, quality, and lead time. Thus, it is important to evaluate the machinability of the design, providing feedback so that the designer can change the design to improve its machinability. For machining purposes, a part is often considered to be a collection of machinable features. However, often there can be several different interpretations of the same part as different collections of machinable features. Each interpretation corresponds to a different set of machining operations, in a different order, with different machinability. To determine the machinability of the part, all of these alternatives should be generated and their machinability evaluated. This research will develop ways to do this automatically. For generating the alternative feature interpretations, the investigator's will use an algebraic approach, in which new interpretations will be produced via algebraic operations on old interpretations. For evaluating the machinability of each feature, the researchers will take into account the feature geometry, tolerance requirements, surface finish requirements, and statistical variations in the process capabilities. The information provided by such an analysis can be used in several ways: (1) To provide information to the manufacturing engineer about alternative ways in which the part might be machined. This information can be useful in developing process planning alternatives depending on machine tool availability. (2) To provide feedback to the designer identifying problems that may arise with the machining. By comparing the best achievable tolerances with the designer's tolerance requirements, investigator's should be able to suggest small changes in the design that will significantly improve the machinability.

9306580 Nau This is the first year funding of a three-year continuing award. A primary goal of the proposed work is to define, analyze, and explicate features of the design of HTN planning systems. The proposed work will have the following benefits: It will correctly characterize HTN planning and provide a framework which accounts for many features used in existing planning systems, but not yet completely formalized (including reduction, critics, commitment choices and filter conditions). It will provide a clear understanding of the efficiency and expressivity of HTN planning, and how these compare with the efficiency and expressivity of other approaches to planning (such as planning with STRIPS-style operators). It will provide ways to determine whether a given set of methods and critics correctly describe the planning problem they were meant to solve. It will describe the characteristics of planning domains in which HTN planners can operate efficiently, and develop efficient planning systems for those domains.

CDA 9529444 Nau, Dana Hendler, James University of Maryland at College Park CISE Research Instrumentation for AI Planning Research The Department of Computer Science and the Institute for Systems Research at the University of Maryland will purchase computing machining equipment, including a small multipurpose machine tool, a pentium-based personal computer for solid modeling and input to the tool, and a large disk drive for keeping large case-based planning libraries, all of which will be dedicated to support artificial intelligence research in the area of computer and information science and engineering. The equipment will be used for several research projects, including in particular: *AI-based planning for manufacturing, *High Performance Support for Large Knowledge Bases, *Combination of generative and case-based reuse techniques in AI planning, and *Low-cost robotics. In addition, a CAD tool will be donated by Bentley Systems Inc. to facilitate experimentation with the AI planning techniques in the manufacturing domain.

This grant provides funding to address two problem areas in the application of information technology to engineering. The first area to be addressed is the development of techniques for automatic classification and retrieval of designs using detailed product design attributes that are more meaningful and accurate than Group Technology codes and can be computed automatically from the designs stored in the database. The second area to be addressed is the development of a hybrid variant/generative approach to process planning that will combine the best characteristics of both variant and generative process planning, while avoiding the worst limitations of each. When a process plan is needed for a new design, the hybrid approach will use the above classification scheme to retrieve portions or `slices` of several plans that are relevant for corresponding portions of the design, and then will combine these slices and modify them to produce a plan for the new design. If successful, this work will have the following benefits. First, it will combine, in an innovative way, the strengths of both variant and generative process planning. Like traditional variant process planning, the proposed approach to process planning will construct process plans that the process engineer may need to improve. However, by automatically adapting the retrieved plans to the new design requirements, it will minimize the need for such improvements, thus producing more successful implementations and applications than previous methods. Second, the proposed work will develop techniques that help design and manufacturing engineers effectively use legacy data and adapt it to new problems. Since it is estimated that practicing engineers spend more than 80 percent of their time searching through legacy data, catalogs, and earlier engineering projects, this should have significant positive impact on engineering practice.

With support from the National Science Foundation's Innovation and Organizational Change Program and GOALI, this research focuses on the development of an intelligent product representation for evolvable designs that can determine when new innovation indicates that a change in a previous decision is warranted. Deciding when to redesign a product is one of the most important decisions made by any company. Incorrect redesign timing can lead to a variety of problems and poor market response. Current strategies for deciding when to redesign products are usually ad hoc in nature, and may result in a significant lag between the time a technological innovation occurs and the time when it is successfully incorporated into a product design.

The researchers are developing decision-making models that will work on intelligent product representations and generate suggestions for redesign These decision making models perform a detailed trade-off analysis whenever technological or business constraints change in the intelligent product representation. These models incorporate decision variables, design constraints, and evaluation criteria that have led to the decisions currently in use. The project will develop a multi-disciplinary course sequence that will incorporate the latest advances in the management of technology area into engineering and business curriculum.

EIA-9986012
Dana S. Nau
University of Maryland-College Park

CISE Research Instrumentation: A Specialized Computing Environment for Distributed and Virtual Design and Manufacturing.

The University of Maryland will create a specialized computing environment for distributed and virtual design and manufacturing. This facility will support research in computer and information science sand engineering. The University of Maryland will purchase two graphics workstations, two NT-based servers, and the necessary software for developing the facility. The computing environment will be dedicated to several on-going projects on distributed and virtual design and manufacturing. These projects will yield specific contributions in areas related to computing and information systems. With this facility, the investigators can install and integrate our approaches, share data across various projects effectively, and test our results more thoroughly. Thus we can study and solve the computational challenges that will occur as manufacturers develop advanced, information-based product realization processes

0243803 Gupta

This award funds a five-year Research Experience for Undergraduates (REU) Site at the University of Maryland for fifteen students each summer for twelve weeks for research opportunities at the university's Institute for Systems Research. Students at colleges, universities, and community colleges will be recruited nationwide through a process involving efforts to reach students who would otherwise not have access to a research experience. The program incorporates activities that will involve participants in the following research directions of the institute: global communications systems, sensor-actuator networks, next-generation product realization systems, societal infrastructure systems, and cross-disciplinary systems education. Through the program students will be able to (1) establish a basis for systems thinking by conducting research and thus understand systems engineering as a discipline; (2) acquire broader and deeper understanding of both the research process and the practice of engineering and how new knowledge is created and communicated; (3) develop multicultural understanding and team competence and become aware of the societal implications of research; and (5) successfully seek admission in a four-year program and/or graduate school.

This project has the goal of developing planner-generalization techniques that can be used to modify AI planning algorithms to remove some of the restrictive assumptions found in classical approaches to AI planning, such as: perfect knowledge about actions and objects and history of the planning environment, static planning environment, instantaneous actions, discrete time, determinism, black-and-white solution criteria. A starting point for the project will be the PI's preliminary results on a method to "non-determinize" forward-chaining planners. Using both theoretical and empirical techniques, the project will explore ways to systematically generalize planning algorithms to deal with nondeterminism, actions with probabilistic effects, and temporal issues. Results of the project will provide theoretical and experimental underpinnings necessary to enable AI planning to better address the needs of real-world planning applications such as manufacturing planning and ship movement planning. It is intended that implementations of algorithms developed in this project will be made available as open-source software.

The objective of this award is to develop feature-based simplification of computer-aided-design models, specifically to accelerate and automate downstream finite-element-analysis. In particular, the research will create algorithmic foundations for learning conservative feature suppression rules from demonstrations performed by human experts. The effect of simplification on simulation accuracy will be formally characterized and this understanding will be used to create robust algorithms for feature suppression within computer-aided design models. Research findings will be integrated into graduate and undergraduate curriculum. The research will ultimately lead to a framework to automatically learn, validate, and apply context dependent model simplification rules that can be audited by human experts, and deployed to automate the model simplification task.

If successful, the research will significantly speed up model simplification, and enhance the automated use of engineering analysis tools in the design process. Potential applications include design of heat exchangers, aircraft structures, and semi-conductor equipment. The planned research and education integration activities and outreach activities will familiarize graduate, undergraduate, and high school students with the use of model simplification technologies in challenging engineering design projects. This project will also increase awareness among practicing engineers about the potential usage of automated model simplification in complex engineering design projects. Students working on this project will also participate in Badger Camp and Engineering Expo events at the University of Wisconsin campus and Maryland Day at the University of Maryland to enhance the public understanding of the science and technology.