Abstract
In modern chemical process simulators, a significant portion of computational resources is dedicated to calculating physicochemical properties. Selecting the optimal estimation route—comprising a primary method and its necessary intermediate calculations—is a complex task influenced by application domains, required accuracy, and data availability. This paper presents an advanced Physicochemical Property Estimation System (PPES) that automates this selection through a two-stage process: routing and sequencing. By employing priority rules and a modular architecture, the system ensures feasible, accurate property estimation with a modest 15% increase in execution time.
Introduction
Process simulators are indispensable tools in chemical engineering for design and optimization. However, these systems often spend up to 80% of their CPU time performing physicochemical property calculations. The primary challenge for users is not merely the computation itself, but the selection of the most appropriate mathematical models. Choosing an estimation route involves navigating a maze of dependencies; for instance, calculating enthalpy deviation might require an acentric factor, which itself can be estimated via multiple competing methods.
The Challenge of Manual Route Selection
Traditionally, human experts manually select estimation routes. This process is prone to errors as systems grow in complexity. A user must consider if a method is suitable for polar mixtures, if the required input data exists in the database, and if the chosen sequence of methods avoids circular dependencies (loops). As the number of available methods increases, manual selection becomes a combinatorial challenge that often leads to sub-optimal or mathematically impossible routes.
System Architecture: Routing and Sequencing
The proposed PPES addresses these challenges by treating property estimation as an automated graph-traversal problem, divided into two distinct procedures: Routing and Sequencing.
Routing and the Priority Rule
The routing phase identifies potential estimation paths. It utilizes a Priority Rule to rank methods based on their statistical reliability within specific application domains. For example, the UNIFAC method is prioritized for highly non-ideal mixtures involving polar components.
The system's modularity is achieved by splitting each property module into two parts:
- Control Section: Contains metadata regarding the method's applicability (e.g., phase, temperature range) and its required input parameters.
- Estimation Section: Implements the actual mathematical algorithm.
Sequencing and Loop Detection
Once a route is proposed, the sequencing module evaluates its integrative feasibility. Its critical function is to detect "loops"—circular dependencies where a method requires an input that eventually depends on its own output. If a loop is detected, the system backtracks to the routing phase to select the next best alternative from the priority list.
Performance and Validation
The system was validated using a Basic Physicochemical Property Database (PCPDB) across 26 typical properties, including critical temperature, latent heat of vaporization, and surface tension.
A key case study involved calculating the vapor-liquid equilibrium (VLE) for an acetone-n-hexane system at 55°C. The PPES automatically navigated a complex third-level dependency route, selecting appropriate models based on the system's non-ideal nature. The resulting K-value of 2.5973 was highly accurate, closely matching the experimental value of 2.5785.
Crucially, the computational "overhead"—the time spent by the system to determine the route—was approximately 15% of the total execution time for typical samples. This is considered a negligible cost for the benefit of guaranteed model suitability and mathematical consistency.
Conclusion
The automated PPES represents a significant advancement in process simulation. By separating the logic of method selection from the algorithms themselves, the system provides a maintainable and extensible framework. While future work may focus on reducing the need for manual "man-machine" intervention in edge cases and scaling to even larger method libraries, the current system successfully mitigates the "combinatorial nightmare" of property estimation.
Summary updated