Automating input-output modeling of dynamic physical systems
Matthew Easley, Elizabeth A. Bradley · 2000
System identification is the process of deducing a mathematical model of the internal dynamics of a black-box system from observations of its outputs. Input-output modeling, where one applies test signals to the system and observes their effects on its outputs, has a critical role in this process. This procedure is extremely difficult to automate. Any artificial intelligence tool that takes an input-output approach to modeling nonlinear dynamic physical systems must represent and reason with many heterogeneous kinds of knowledge about the real world. Knowledge about a target system ranges from general mathematics that applies in all situations—for example, theorems about differential equations—to very specific knowledge that is only useful in limited circumstances, such as specific forms of friction in ball bearings. The challenges in automating the input-output modeling process are to smoothly incorporate these varying levels and types of knowledge and to apply the appropriate reasoning techniques at the right place and time. This thesis describes a knowledge representation and reasoning framework that solves these problems. This framework encapsulates knowledge that is critical to the generation phase of the model-building process: fundamental engineering tactics and representations for efficient model construction and testing, and a succinct and yet effective method for reasoning about the relationship between a system's inputs and its outputs. The model-building knowledge resides in a small, powerful meta-domain theory that tailors the space of candidate models to the problem at hand. The input-output knowledge is instantiated in an intelligent sensor/actuator control module, which uses a process termed qualitative bifurcation analysis to reason effectively about sensors and actuators and their interactions with the target system. These two paradigms work well together: meta-domains provide a compact and flexible way of representing domain knowledge; qualitative bifurcation analysis allows the modeler to generalize detailed information about specific instances of a target system into abstract classes of behaviors. The context in which I demonstrated the power and utility of these ideas was the program PRET, which automates the system identification process by building a layer of artificial intelligence techniques around a set of traditional formal engineering methods. Unlike other modeling tools—most of which use libraries to model small, well-posed problems in limited domains and rely on their users to supply detailed descriptions of the target system—P RET works with nonlinear systems in multiple domains and interacts directly with the real world via sensors and actuators. My knowledge representation and reasoning framework allows PRET to succeed in a variety of simulated and real applications, ranging from textbook systems to real-world engineering problems.