Multivariate Data Analysis of Post Combustion CO2 Capture
1
Multivariate Data Analysis of Post Combustion CO 2 Capture Kasturi Nagesh Pai 1 , Sai Gokul Subraveti 1 , Arvind Rajendran 1 , Vinay Prasad 1 , Zukui Li 1 PARTNERS 1 Dept of Chemicals and Materials Engineering , 116 th st 89 th Ave, Edmonton, Alberta. FES PROJECT OVERVIEW BACKGROUND RESULTS FUTURE DIRECTIONS • Explosive growth in ability to synthesize novel solid sorbents • Proven ability to reduce parasitic energy consumption • Process design and optimization challenging Adsorption for CO 2 Capture Optimizer • Multi Objective Optimization • Stochastic Output PSA Cycle Simulator • Stiff PDEs • Cyclic Process • Convergence Complexities ≥3000 single core hours for the evaluation of 1 material Material Characterization & Tailoring Properties Scale up and Industrial level simulations and Implementation Process Design Engineers Reduce Parasitic Energy Machine Learning Filters Materials Engineers Quick Performance Evaluation Process Design & Simulations Lab Scale Demonstration Single Column Experiments Systems Level Optimization AIMS AND OBJECTIVES Parallel Computing Evaluation and Optimization of One Solid Sorbent Single Core Computing Material Data Machine Learning Algorithm Evaluation and Optimization of One Solid Sorbent ≥120 hours to evaluate one material on multiple core processors ≥0.12 hours to evaluate one material on a single processor Previous Data about material performance 3-4 orders of magnitude computing time saved Filtering out materials Material Data PSA Cycle Program Hydrocarbons will continue to serve as an essential energy source while the world transitions to a lower-carbon energy economy, but can we prevent the use of those fuels from contributing to the accumulation of CO 2 in the atmosphere? Existing technologies can capture carbon, but these methods can be costly and energy-intensive. Extracting energy without burning fuels, improving CO 2 capture efficiencies if they are burned, and finding effective ways to store or reuse captured carbon may be essential to ensuring it does not enter the atmosphere. Power Generation CO 2 for Storage & Usage Air Coal 66000 TPD Flue Gas Clean Air CO 2 Capture Unit Adsorbent Material Process Cycle Parasitic Energy 500 MW Machine Learning Algorithm Training Solid Adsorbent 6 variables Isotherm 12 variables Feed CO 2 :N 2 15:85 Extract Product ADS Co-BLO Cn-BLO LPP Light Product 3 variables 4 variables Simulation based Data Dimensional Reduction (PCA) Supervised Machine Learning • Support Vector Machines • Ensemble Bagged Training R 2 Adj Purity - 0.87 Recovery - 0.92 Energy - 0.98 Productivity - 0.99 70 Materials for Evaluation Proxy Model 7 Variables Full Model Evaluation 150000 Core Hours • Accurate Purity, Recovery, Energy and Productivity values • 6 operational variables 12 material parameters Proxy Model Evaluation 1.5 Core Hours • Accurate Filtration of Purity and Recovery • Precise Productivity Predictions • Limitations- Energy Predictions Proxy Model Evaluation Pros: • Quick and Accurate Evaluations of Multiple Materials • Quick Optimization of Materials • Based on only 4 Material properties Cons: • Not as Accurate as Detailed Model • Experimental Validation pending
Transcript of Multivariate Data Analysis of Post Combustion CO2 Capture
Multivariate Data Analysis of Post Combustion CO2 Capture Kasturi Nagesh Pai1, Sai Gokul Subraveti1, Arvind Rajendran1, Vinay Prasad1, Zukui Li1
PARTNERS
1Dept of Chemicals and Materials Engineering , 116th st 89th Ave, Edmonton, Alberta.
FES PROJECT OVERVIEW
BACKGROUND
RESULTS
FUTURE DIRECTIONS
• Explosive growth in ability to synthesize novel solid sorbents
• Proven ability to reduce parasitic energy consumption
• Process design and optimization challenging
Adsorption for CO2 Capture
Optimizer• Multi Objective Optimization• Stochastic Output
PSA Cycle Simulator• Stiff PDEs• Cyclic Process• Convergence Complexities
≥3000 single core hours for the evaluation of 1 material
Material Characterization
& Tailoring Properties
Scale up and Industrial level simulations and
Implementation
Process Design Engineers
Reduce Parasitic Energy
Machine Learning Filters
Materials Engineers
Quick Performance
Evaluation
Process Design & Simulations
Lab Scale Demonstration
Single Column Experiments
Systems Level
Optimization
AIMS AND OBJECTIVES
Parallel Computing
Evaluation and Optimization of
One Solid Sorbent
Single Core Computing
Material Data
Machine Learning Algorithm
Evaluation and Optimization of
One Solid Sorbent
≥120 hours to evaluate one material on multiple core processors
≥0.12 hours to evaluate one material on a single processor
Previous Data about material performance
3-4 orders of magnitude computing time saved Filtering
out materials
Material Data
PSA Cycle Program
Hydrocarbons will continue to serve as an essential energy source while the world transitions toa lower-carbon energy economy, but can we prevent the use of those fuels from contributing tothe accumulation of CO2 in the atmosphere? Existing technologies can capture carbon, butthese methods can be costly and energy-intensive. Extracting energy without burning fuels,improving CO2 capture efficiencies if they are burned, and finding effective ways to store orreuse captured carbon may be essential to ensuring it does not enter the atmosphere.
Power Generation
CO2 for Storage& Usage
Air
Coal66000 TPD Flue Gas
Clean Air
CO
2C
aptu
re
Un
it
Adsorbent Material
Process Cycle
Parasitic Energy500 MW
Machine Learning Algorithm Training
Solid Adsorbent
6 variables
Isotherm
12 variables
FeedCO2:N2
15:85
Light Product
Extract Product
AD
S
Co
-BLO
Cn
-BLO
LPP
Light Product
3 variables
4 variables
Simulation based Data
Dimensional Reduction (PCA)
Supervised Machine Learning
• Support Vector Machines• Ensemble Bagged
Training
R2Adj
Purity - 0.87Recovery - 0.92
Energy - 0.98 Productivity - 0.99
70 Materials for Evaluation
Proxy Model
7 Variables
Full Model Evaluation150000 Core Hours• Accurate Purity, Recovery, Energy and Productivity values• 6 operational variables 12 material parameters
Proxy Model Evaluation 1.5 Core Hours• Accurate Filtration of Purity and Recovery• Precise Productivity Predictions • Limitations- Energy Predictions
Proxy Model Evaluation Pros:• Quick and Accurate Evaluations
of Multiple Materials• Quick Optimization of Materials• Based on only 4 Material
properties Cons:• Not as Accurate as Detailed
Model• Experimental Validation pending
