Acquisition of Functional Models: Combining Adaptive Modeling and Model Composition

Functional models explicitly represent the functions of a system and enable teleological reasoning. Structurebehavior-function models of a physical device, for example, specify the device functions and also its internal causal behaviors that explain how the device structure achieves the functions. They enable reasoning about the device teleology in tasks such as adaptive design and redesign. An open issue in research on functional models concerns their origin and acquisition. We describe a computational technique that combines the methods of adaptive modeling and model composition. In the integrated method, the model of the new device is acquired by revising the model of a similar device, and, if the new device contains structural elements not in the similar device, then by consolidating the models of the new elements with the model of the similar device. This technique is one part of a general computational theory of conceptual design called model-based analogy. Background, Motivations, and Goals Functional models explicitly represent the functions of a system to enable teleological reasoning about function-related tasks (Sembugamoorthy and Chandrasekaran 1986). Structure-behavior-function (SBF) models of physical devices, for example, specify the structure, the functions, and the internal causal behaviors that explain how the device structure results in its functions (Goel 1991). In particular, the internal behaviors specify how the functions of the structural elements are hierarchically composed into the functions of the device. SBF models have proved to be quite useful for teleological reasoning about function-related tasks such as variant and adaptive design, design verification, and redesign (Goel, Bhatta, and Stroulia 1997). Since they explicitly represent the functions and use them to organize behavioral knowledge, they help define problem spaces for design adaptation, verification and redesign, and provide access to the knowledge relevant for searching the spaces. They also give rise to a vocabulary for indexing designs. But the origin, generation and acquisition of functional models remain open issues. Not only are these questions fundamental, but, in addition, their answers are likely to impose additional constraints on the models. We describe a computational strategy for acquiring SBF models that integrates the methods of adaptive modeling and model composition. Given the structure of a new device as input, the former method generates a model for the new device by revising the model of a similar device, while the latter method generates the device model by composing models of the device elements. The new strategy integrates Bylander’s (1991) consolidation method of model composition with our method of adaptive modeling (Goel 1991, 1996). In the new strategy, the model of the new device is generated by revising the model of a similar device, and, if the new device contains structural elements not in the similar device, then by consolidating models of the new elements with the model of the similar device. The choice of consolidation as the method of model composition is due to its ontological compatibility with our adaptive-modeling method. The integrated strategy combines the efficiency of adaptive modeling with the generality and power of model composition. The origin of this work lies in our research on analogybased innovative design. We have developed a computational theory of innovative design called model-based analogy (or MBA) (Bhatta 1995). We have also instantiated and evaluated MBA in an operational computer program called IDEAL. The system solves function! structure design problems autonomously, by model-based retrieval and adaptation of design cases, and by model-based transfer of design abstractions over known cases. Depending on its background knowledge relative to a given problem, IDEAL might fail to solve the problem. If and when it fails, the system interacts with an oracle. If the oracle supplies the correct design, then IDEAL generates an SBF model for the new design and stores it in its case memory for potential reuse. This raises the issue of generation and acquisition of SBF models, and leads to the new integrated strategy. Thus the new model-acquisition strategy is one part of the MBA theory, and is instantiated in IDEAL. In this paper, we use the simplest example from IDEAL to illustrate the new strategy, but the strategy is applicable to a large range of problems as demonstrated by IDEAL. Also, we are not claiming that generating a model for the simple device in this example requires the integrated method, but merely using the example for illustration. From: AAAI Technical Report WS-98-01. Compilation copyright © 1998, AAAI (www.aaai.org). All rights reserved.