Building Layered Models to Support Engineering Decision Making: A Machine Learning Approach

Abstract

This paper presents a new model building methodology which, given a detailed mechanistic model of a task, can optimally produce a set of models with layered abstraction according to the user’s specified modeling objectives. These layered models can be used to evaluate decisions made at different levels of abstraction during a typical problem-solving process such as engineering design and planning. In our research, the model building process is viewed as a learning activity and inductive machine learning techniques from AI are combined with traditional optimization methods to form our prototype model building system called AIMS (Adaptive and Interactive Modeling System). The layered analysis models built by AIMS offer several distinctive advantages over those traditional analysis models which can only provide evaluations at very detailed stages of decision making. These advantages include: early evaluation to avoid costly iterations, fast execution for interactive applications, more comprehensibility for human inspection, and deep roots in domain physics for higher accuracy. Case study results of building layered models for a process design task of an intermittent cutting process are presented as a demonstration of the potential use of our system. We also explain this model building research in the context of the knowledge processing technology as a new foundation for advanced engineering automation.