1 of 16

Systems Analysis

Leveraging Hybrid Domain Knowledge and Machine Learning Approaches for Predicting Plastic Pyrolysis Performance

​

​

Jiaze Ma�Pallavi Dubey�Victor Zavala�Mark Mba Wright

1

​

​

Sustainable Energy Systems Analysis (SESA) Group

2 of 16

Background and Motivation

Plastic Pyrolysis Data

Machine Learning Approach

Collaboration with Experimental Groups

Integration with Techno-Economic and Lifecycle Analysis

​

2

​

​

Sustainable Energy Systems Analysis (SESA) Group

3 of 16

Machine Learning for Biomass Fast Pyrolysis

Kinetics + Machine Learning

Predicted Biofuel Prices:

    • $2.62–$5.43 per gallon

Predicted GHG Emissions:

    • −13.62 to 145 kg of CO2 per MJ

3

​

​

Sustainable Energy Systems Analysis (SESA) Group

4 of 16

Plastic Pyrolysis Kinetics

​

Identified: 2087

Screened: 151

Machine Learning: 35

Quantum Mechanics: 15

Global Kinetics: 101

4

Moher et al. 2009; Armenise et al. 2021

​

​

Sustainable Energy Systems Analysis (SESA) Group

5 of 16

Plastic Pyrolysis Kinetics

5

Web of Science

​

​

Sustainable Energy Systems Analysis (SESA) Group

6 of 16

Plastic Pyrolysis Kinetics

6

  • Limitations:
  • Single/Binary Plastic Feed(s)
  • Limited Material Characterization
  • TGA/DSC
  • Limited Reactor Representation
  • Limited Range of Operating Conditions

​

​

​

Sustainable Energy Systems Analysis (SESA) Group

7 of 16

Plastic Pyrolysis Machine Learning

Literature-based Dataset of 570 Points

Trained ML with ~300 points

Decision tree (DT),

Artificial neuron network (ANN),

Support vector machine (SVM),

Gaussian process (GP)

7

​

​

Sustainable Energy Systems Analysis (SESA) Group

8 of 16

Plastic Pyrolysis Machine Learning

Literature-based Dataset of ~300 points

Features:

    • Temperature
    • Heating Rate
    • Particle Size
    • Catalyst
    • Reactor Type
    • Oil Yield

8

​

​

Sustainable Energy Systems Analysis (SESA) Group

9 of 16

Other Machine Learning Approaches

9

​

​

Sustainable Energy Systems Analysis (SESA) Group

10 of 16

Machine Learning Approach

10

  • Train a prediction model
  • Compare to experimental data
  • Identify high information regions
  • Identify high uncertainty regions

​

​

Sustainable Energy Systems Analysis (SESA) Group

11 of 16

Machine Learning Approach

11

​

​

Sustainable Energy Systems Analysis (SESA) Group

12 of 16

Machine Learning Approach

12

​

​

Sustainable Energy Systems Analysis (SESA) Group

13 of 16

Collaboration with Experimental Groups

13

​

​

Sustainable Energy Systems Analysis (SESA) Group

14 of 16

Integration with Techno-economic and Lifecycle Assessment

Is the Machine Learning Prediction sufficient?

    • Should we combine it with global kinetic models?
    • Should we identify the required or missing data (reactor configuration, catalyst type, equivalence ratio…)?
    • Should we define what data will be shared with the experimental groups (optimal conditions, high uncertainty conditions, feed compositions)?

​

14

​

​

Sustainable Energy Systems Analysis (SESA) Group

15 of 16

Integration with Techno-economic and Lifecycle Assessment

15

Olafasakin et al. 2021

​

​

Sustainable Energy Systems Analysis (SESA) Group

16 of 16

16

markmw@iastate.edu

pallavid@iastate.edu

​

​

Sustainable Energy Systems Analysis (SESA) Group