National Energy Technology Laboratory Dirk Van Essendelft (PI) Terry Jordan, Philip Nicoletti,...

15
National Energy Technology Laboratory Dirk Van Essendelft (PI) Terry Jordan, Philip Nicoletti, Tingwen Li (Team Members) Multiphase Flow Team, CSED August 13, 2015 Recent Developments and Accomplishments in C3M

Transcript of National Energy Technology Laboratory Dirk Van Essendelft (PI) Terry Jordan, Philip Nicoletti,...

PowerPoint Presentation

Dirk Van Essendelft (PI)Terry Jordan, Philip Nicoletti, Tingwen Li (Team Members)Multiphase Flow Team, CSEDAugust 13, 2015Recent Developments and Accomplishments in C3M

National EnergyTechnology Laboratory1What Does C3M Bring to the User?Easy, Intuitive, Reliable, and Graphical User InterfaceComprehensive interface between reliable sources of kinetic data and reacting, multi-phase CFD modelsVirtual Kinetic Laboratory for quickly assessing the validity of a chemical equation sets before going to full scale, expensive modelsSeamless formatting and units management for code specific implementationAdvanced Chemistry Analysis and Development ToolsOpen source for collaboration and development

What is C3M?

C3M is Chemistry Support for the Computational Modeler

124+ User Downloads Since April 1st100+ Version 2015 Downloads2Recent Developments in C3M

Virtual Experimental Capability (TGA/Drop Tube)

Neural Network SurrogatesUser Defined Chemistry

1) Select Species2) Define ChemistryUser Defined Modules

The development team for C3M have recently implemented a number of significant improvements to C3M. These developments include a symbolic math capability based on Pythons Sympy module, a fully implemented user defined chemistry capability, and a fully implemented user defined module capability among many others. The symbolic math capability in C3M is significant because it eliminates large portion of potential coding errors, ensures that all defined equations are handled consistently for any CFD code, and lessens maintenance and development time for chemistry writers within C3M. Further, the symbolic math capability was foundational to the implementation of a solid user defined chemistry capability and that user defined capability formed the basis for the user defined module capability. Users can now build, store, organize, and share their own chemistry and be confident that it will be implemented consistently in any available chemistry writer within C3M. Furthermore, a significant amount of effort was given to verification of the C3M output both in the user defined modules and the gasification module. In addition, significant effort has been devoted to ensuring that the advanced features like neural network chemistry modeling are fully implemented for CFD codes that allow for user defined functionality. This capability allows users to implement extremely complex chemistry without sacrificing either computational speed or fidelity. Finally, development in C3M is shifting to new territory by expanding the number of writers to include more multiphase CFD codes and more advanced chemistry integration.

3Neural Network Pyrolysis TGA Demonstration

4Neural Network Gasification Demonstration

5Adaptive NN Training Next Generation Technology

6Adaptive NN Training Next Generation Technology

7User Defined Chemistry and Module Demonstration

8US-Canada Clean Energy Dialogue (CANMET Collaboration)Work with CANMET to compare the Neural Network surrogate model to their existing CFD models and dataWork with CANMET to integrate the ROM based model of their PWR style test reactor and do a UQ study with itDirect from Experiment to Modeling CapabilityCollaborate with Advance Combustion folksDirect from TGA to CFD capabilityLarge Model Reduction Using Neural NetworksApply our advanced Neural Network Training Ability to large scale chemistry problems like advanced hydrocarbon combustion (many hundred step mechanisms) using and reduce them to a Neural Network that can run in a large scale CFD simulationOpenFOAM SupportWrite an exporter for OpenFOAM (another popular multiphase, open source CFD code)Where Do We Go from Here?9

10

Move From an Ad Hoc Research Code to Finalized and PolishedCheck against all sources of information, unitsAnnotate EquationFinalize Export CodeActs as Benchmark Point to Ensure ConsistencySubtask 2.3:Verify and Finalize Existing Gasification Chemistry

11Subtask 2.3:Neural Network SurrogatesBasic/Traditional C3M

Limited Local InformationLow AccuracyNo Speed SacrificeUnknown ErrorVery Litd. Complexity12

Subtask 2.3:Neural Network SurrogatesLimited Domain Scale Information

Limited Local Information

Explicit Surrogate C3MMedium AccuracySmall Speed SacrificeKnown ErrorLimited Complexity13

Subtask 2.3:Neural Network SurrogatesUnlimited Local Information

Unlimited Domain Scale InformationReaction RatesNeural Network C3MHigh AccuracySmall Speed SacrificeKnown ErrorUnlimited Complexity14Subtask 2.3:Neural Network SurrogatesSimplified PSDF Riser Model

Proof of Concept for Surrogate Implementation

Functioning reacting model

Not Validated, still in alpha release form

Simplified model of the psdf. Say where recycle is coming in, coal AMD outlet

15