Software Framework for Advanced Power Plant Simulations

John F. Widmann, Sorin Munteanu, Aseem Jain, Pankaj Gupta, Mark Moales, Erik Ferguson, Lewis Collins, David Sloan, W. A. Fiveland, Yidong Lang, Lorenz T. Biegler, Michael Locke, S. Lingard, Jay Yun2010 · DOI

TL;DR

This paper details the development of APECS, a software framework that integrates high-fidelity Computational Fluid Dynamics (CFD) with process modeling for advanced power plant simulations. It introduces Reduced Order Models (ROMs) derived from CFD to significantly speed up simulations, making complex power plant design and analysis more efficient.

What Problem Does This Paper Address?

Developing advanced power plant concepts requires sophisticated simulation tools that can accurately model both plant-level processes and detailed equipment behavior. Existing process simulation software often lacks the fidelity to capture crucial fluid dynamics and equipment-specific behaviors, while high-fidelity CFD simulations are computationally expensive and time-consuming. This creates a gap in the ability to efficiently evaluate and optimize new power plant designs, especially during early conceptual and detailed design phases. The paper addresses the need for a framework that bridges this gap by enabling seamless integration of CFD with process modeling and developing computationally efficient surrogate models (ROMs).

Methods

The core of the paper describes the development of the Advanced Process Engineering Co-Simulator (APECS). Key methodological advancements include:

  • Integration Controller: A robust controller was developed to mediate information exchange between CAPE-OPEN compliant process modeling environments (like Aspen Plus) and CFD software (like FLUENT). This ensures vendor-neutral integration.
  • Reduced Order Models (ROMs): To overcome the computational cost of CFD, ROMs were developed. These models are "trained" using data from high-fidelity CFD simulations and can then be used directly in flowsheets, offering significant speedups. The paper discusses two main approaches for ROM development:
  • Regression-based ROMs: These models are built using statistical techniques like regression analysis.
  • PCA-based ROMs: Principal Component Analysis is used for model-order reduction, particularly effective for capturing complex field data like temperature contours.
  • Parallel and Remote Execution: Strategies were implemented to run CFD models on remote clusters and to execute multiple CFD models simultaneously, further reducing simulation time.
  • Model Database: An advanced model database (ANSYS EKM) was incorporated for efficient storage, retrieval, version control, and sharing of unit operation models (both CFD and ROMs).
  • Virtual Reality Integration: APECS was integrated with a virtual reality tool for 3D visualization of power plant designs and simulation results, enabling virtual plant walkthroughs.
  • Dynamic Simulation Prototype: A prototype dynamic simulator was developed to integrate plant-level and equipment-level models for transient simulations.

Datasets

The paper does not explicitly list or describe distinct datasets in the traditional machine learning sense. Instead, the "datasets" are generated through:

  • CFD Simulations: High-fidelity FLUENT simulations of various equipment items (e.g., gasifier, combustor, heat recovery steam generator, radiant syngas cooler) under different operating conditions.
  • Design of Experiments (DOE): Techniques like Latin Hypercube Sampling (LHS) were used to generate a systematic set of input parameters for the CFD simulations, ensuring coverage of the operating range.
  • Existing Models: Leveraging pre-existing CFD models and process models (e.g., from previous projects or industry partners) was crucial for generating the data for ROM training.

Key Insights and Ideas

  • Bridging the Fidelity Gap: The core insight is that integrating high-fidelity CFD with process modeling is essential for accurate advanced power plant simulations, but the computational cost of CFD is a major bottleneck.
  • ROMs as a Solution: Reduced Order Models (ROMs) are a powerful solution to this bottleneck, offering a significant speedup while retaining acceptable accuracy for many applications.
  • PCA for Complex Data: Principal Component Analysis (PCA) proved to be a valuable technique for developing ROMs that can capture complex spatial data (like temperature contours), overcoming limitations of traditional POD methods.
  • CAPE-OPEN Standard's Importance: The reliance on the CAPE-OPEN standard was key to achieving vendor-neutral integration between different software packages, allowing APECS to work with various process simulators and CFD codes.
  • Iterative Development: The project highlights the iterative nature of software development, with continuous improvements and bug fixes leading to a more robust and user-friendly tool.

Key Results

  • Developed APECS Framework: A comprehensive software framework (APECS) was developed, enabling the integration of CFD models with process modeling environments.
  • Efficient ROMs: Computationally efficient ROMs (regression and PCA-based) were successfully developed and implemented, offering significant speedups compared to direct CFD simulations. For example, ROM execution was orders of magnitude faster than CFD.
  • Successful Co-Simulations: The framework facilitated successful co-simulations of various power plant components, including gasifiers, combustors, and heat recovery steam generators, demonstrating its practical applicability.
  • Virtual Plant Walkthrough: Integration with virtual reality tools allowed for 3D visualization and navigation of simulated power plants.
  • Prototype Dynamic Simulator: A prototype dynamic simulator was developed, demonstrating the potential for transient simulations.

Limitations and Future Work

  • 2D PCA Limitation: The PCA technology for ROM contour plots was initially limited to 2D cases, with 3D capabilities expected later.
  • ROM Robustness: While regression ROMs showed promise, further work is needed to provide better statistical indicators for users to select the optimal regression order.
  • User Interface Enhancements: Suggestions for improving the ROM Builder include:
  • Providing better statistical indicators for ROM selection.
  • Allowing users to view all simulation results simultaneously.
  • Saving convergence monitors for better analysis.
  • Automating the prescription of Min/Max ranges for inlet variables.
  • Steam-Water Property Calculator: A more robust and platform-independent steam-water property calculator is recommended for FLUENT® tube bank models.
  • 3D PCA Development: Developing 3D PCA transformations for APECS is a key area for future research.
  • Interactive Post-processing: Enhancing interactive capabilities for manipulating PCA-generated plots (e.g., changing scales, enlarging areas) is suggested.
  • Pollutant Model Integration: Developing more flexible CAPE-OPEN variables for toggling pollutant models and specifying iterations is recommended.
  • SOx Model Improvement: A robust and reliable SOx mechanism and submodel are needed.
  • Remote FLUENT® Interaction: Improving user interaction with FLUENT® when running remotely on Linux platforms is suggested.
  • Automatic Saving of Cases: Implementing an option to automatically save FLUENT® cases after each execution is recommended.
  • DPM Inlet Ports: Upgrading the flow-splitter capability to include DPM inlet ports is suggested.
  • Parallelization of UDFs: Providing templates for proper parallelization directives for UDFs monitoring particle trajectories is recommended.

Summary updated

Abstract

This report summarizes the work accomplished during the Phase II development effort of the Advanced Process Engineering Co-Simulator (APECS). The objective of the project is to develop the tools to efficiently combine high-fidelity computational fluid dynamics (CFD) models with process modeling software. During the course of the project, a robust integration controller was developed that can be used in any CAPE-OPEN compliant process modeling environment. The controller mediates the exchange of information between the process modeling software and the CFD software. Several approaches to reducing the time disparity between CFD simulations and process modeling have been investigated and implemented. These include enabling the CFD models to be run on a remote cluster and enabling multiple CFD models to be run simultaneously. Furthermore, computationally fast reduced-order models (ROMs) have been developed that can be 'trained' using the results from CFD simulations and then used directly within flowsheets. Unit operation models (both CFD and ROMs) can be uploaded to a model database and shared between multiple users.
BibTeX
@misc{widmann2010software,
  author = {John F. Widmann and Sorin Munteanu and Aseem Jain and Pankaj Gupta and Mark Moales and Erik Ferguson and Lewis Collins and David Sloan and W. A. Fiveland and Yidong Lang and Lorenz T. Biegler and Michael Locke and S. Lingard and Jay Yun},
  title = {{Software Framework for Advanced Power Plant Simulations}},
  year = {2010},
  doi = {10.2172/1004878},
}
Figure