Multimodal Case-Based Reasoning in Modeling and Design

该项目将进一步开发基于案例的推理(CBR)技术，用于从线条图自动生成多模式绘图-结构-行为-功能(DSBF)目的论模型。这个项目的重点是概念性的--即初步的、定性的--涉及直线运动和旋转运动的设计。所采用的方法的动机来自于当前的计算机辅助设计(CAD)实践，该实践使用各种“模式”信息(例如，草图、图纸、2-D和3-D几何模型)，并侧重于感知信息。该项目的一个长期实际目标是促进新一代CAD工具的开发，这些工具可以支持将目的论模型与图纸结合使用。长期的理论目标是在目的论建模和设计中建立多模式CBR的计算理论。在本项目中，案例存储由已知图形的DSBF模型组成，案例按其图形及其功能进行索引。给定运动学设备的目标图纸，CBR系统检索被认为与新图纸相似的已知图纸的DSBF模型。然后，DSBF模型引导将检索到的绘图的DSBF模型转移和适配为新绘图的候选DSFB模型。这项拟议的研究正在发展多模式规律性模式的概念，称为一般视觉目的性机制(GVTM)。GVTM将用于帮助将新图纸和已知图纸之间的差异映射到已知图纸的DSBF模型中，并使系统能够推断图纸之间的差异(例如，不同的部件、相同部件的不同配置)。构建的DSBF模型的质量将通过在遗留系统中为不同的任务重用它们来评估，即为实现特定功能的设计自动生成结构。这项工作的一个更广泛的影响是对下一代CAD系统和CBR系统的总体影响。另一个贡献是将SBF方法用于目的论建模，作为教授和学习科学和工程中因果模型的一种技术，例如，佐治亚理工学院的研究团队在NSF SLC(学习中心科学)催化剂赠款上共同工作。

This project will further develop case-based reasoning (CBR) techniques for the automatic generation of multi-modal drawing-structure-behavior-function (DSBF) teleological models from line drawings. The focus in this project is on conceptual-that is, preliminary, qualitative-designs involving linear and rotational motion. The employed approach takes its motivations from current practice in computer-aided design (CAD), which uses a variety of "modal" information (e.g., sketches, drawings, 2-D and 3-D geometric models), and focuses on perceptual information. A long-term practical goal of the project is to contribute to a new generation of CAD tools that can support the use of teleological models in conjunction with drawings. The long-term theoretical goal is to build a computational theory of multi-modal CBR in teleological modeling and design. In this project, case memory consists of DSBF models of known drawings, and cases are indexed by their drawings as well as their functions. Given a target drawing of a kinematics device, the CBR system retrieves a DSBF model of a known drawing that is considered to be similar to the new drawing. The DSBF model then guides the transfer and adaptation of the DSBF model of the retrieved drawing into a candidate DSFB model for the new drawing. The proposed research is developing the idea of multi-modal patterns of regularity, called Generic Visual Teleological Mechanisms (GVTM). GVTM would be used to help map differences between the new and the known drawings into the DSBF model of the known drawing, and to enable the system to reason about the differences between drawings (e.g., different components, different configurations of the same components). The quality of the DSBF models constructed is to be evaluated by reusing them in a legacy system for a different ask, namely, automated generation of a structure for a design achieving a specified function. One of the broader impacts of this work is on the next generation of CAD systems and CBR systems in general. Another contribution would be the use of the SBF methodology for teleological modeling as a technique for teaching and learning causal models in science and engineering, for instance, by the research team at Georgia Tech working together on an NSF SLC (Science of Learning Centers) Catalyst grant.