A Theory of Mapping from Structure to Function Applied to Engineering Thermodynamics

John O. Everett

Introduction

In the field of engineering, a fundamental challenge lies in bridging the gap between a system's structure—its physical arrangement of components like pipes, pumps, and turbines—and its function, or intended purpose. While an experienced engineer can discern the goal of a complex schematic, automating this reasoning is difficult due to functional ambiguity. For instance, a turbine's physical behavior is always fluid expansion, but its role might be to generate work in a power plant or to provide cooling in a liquefaction system.

This paper introduces CARNOT, a theoretical framework and computational system designed to automatically translate thermodynamic schematics into meaningful functional explanations. Unlike previous systems that relied on rigid template matching, CARNOT employs a generative approach based on Qualitative Process Theory (QPT) to reason from first principles.

The CARNOT Framework

The core of CARNOT is a hierarchical mapping process that translates low-level structural data into high-level functional descriptions. This process is managed using a logic-based Truth Maintenance System (TMS) and follows several distinct layers of representation:

  1. Views: These describe device-specific behaviors based on local conditions (e.g., a pump "compressing liquid").
  2. Roles: These define the intended function of a device within its specific context (e.g., a pump acting as a "flash-preventer").
  3. Processes: This layer includes local, boundary, and Aggregate Processes. Aggregate processes are a key innovation, allowing the system to group multiple physical components—such as a series of turbines—into a single functional stage.
  4. Plans: These represent high-level abstractions of recognized cycles, such as the Rankine or Refrigeration cycles.
The hierarchical mapping from system topology and data to functional explanation.

The system performs a topological parse of the schematic to identify "floops" (fluid loops), which serve as the primary units of analysis. By resolving inequalities in temperature and pressure across these loops, CARNOT can confirm the roles of ambiguous components like mixers and splitters.

Reasoning and the Rational Designer

CARNOT's reasoning is guided by the Rational Designer Premise: the assumption that every component in a valid design exists to serve a specific goal. These goals typically fall into three categories: achieving a desired state change in the environment, maximizing thermodynamic efficiency, or preserving the integrity of the system (e.g., preventing component damage).

By utilizing a TMS, CARNOT provides articulate explanations for its conclusions. If queried about the presence of a specific component, the system can backtrack through its inference chain to explain how that component contributes to the overall system goals.

Empirical Evaluation

The CARNOT system was evaluated against established engineering benchmarks to test its coverage and robustness. The system was tasked with explaining steady-flow cycles of varying complexity, ranging from 4 to 24 components.

A standard thermodynamic cycle schematic representing the structural input for the CARNOT system.

In tests using introductory texts (e.g., Whalley, 1992), CARNOT achieved a 100% success rate across all 8 steady-flow cycles. When tested against more advanced professional texts (e.g., Van Wylen & Sonntag, 1985), it successfully explained 24 out of 32 cycles (75%). The system demonstrated a sophisticated ability to identify non-obvious functional roles, such as recognizing a mixer as an "open heat-exchanger" based purely on its topological context.

Conclusion and Future Work

CARNOT demonstrates that a generative, first-principles approach to functional mapping is both feasible and effective for complex engineering domains. While currently limited to steady-flow systems, the framework provides a solid foundation for future expansions into non-steady systems (like internal combustion engines) and other fluid-flow domains like hydraulics. The ultimate vision is the integration of CARNOT into intelligent tutoring systems like "CyclePad," providing students with expert-level automated feedback on their thermodynamic designs.

Summary updated

BibTeX
@inproceedings{everett1995theory,
  author = {John O. Everett},
  title = {{A Theory of Mapping from Structure to Function Applied to Engineering Thermodynamics}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {1995},
}
Figure