In progress – Group project, 12 months
Artificial intelligence for the co-pyrolysis of cassava stumps and oil palm fibre
The project develops artificial-intelligence models that predict the thermogravimetric, energy and exergy behaviour of the co-pyrolysis of two abundant agro-industrial residues, cassava stumps and oil palm fibre, from thermogravimetric tests with quantified uncertainty.
Project details
- Code
- INGEBIR-P01
- Status
- In progress
- Duration
- 12 months
- Type
- Applied research
- Lines
Title
Artificial-intelligence models to predict the thermogravimetric, energy and exergy behaviour of the co-pyrolysis of cassava stumps and oil palm fibre in the department of Sucre
Cassava stumps are usually left scattered in the field after harvest, and the palm fibre that is not burned in the boilers of oil mills ends up in low-value uses. Processed together by co-pyrolysis they can become biochar, bio-oil and syngas, because the lignin-rich fibre adds thermal stability and the fast-volatilising stumps favour the formation of volatiles.
For these residues there are no data on the most efficient blend ratios, on the quality of the energy obtained under the second law of thermodynamics or on the uncertainty of the kinetic parameters used to design reactors. The project provides that baseline and extends it to untested blends with machine learning.
Objectives
General objective
Develop artificial-intelligence models to predict the thermogravimetric, energy and exergy behaviour during the co-pyrolysis of cassava stumps and oil palm fibre.
Specific objectives
- Characterise cassava stumps and oil palm fibre physicochemically and energetically, and establish the thermogravimetric behaviour of each residue and their blends at five heating rates.
- Determine the kinetic parameters, reaction mechanism and thermodynamic and exergy performance of the co-pyrolysis, with the uncertainty of each parameter.
- Develop and validate interpretable artificial-intelligence models that predict the mass loss and the kinetic, thermodynamic and exergy parameters of the blends.
How it works
Three linked phases built on a single experimental dataset.
Characterisation
The biomasses are characterised in triplicate by proximate, ultimate, structural and calorimetric analysis. Five blend ratios are tested at five heating rates, from 5 to 25 °C/min, for a total of 75 thermogravimetric runs.
Kinetics, thermodynamics and exergy
The FWO, KAS and Starink isoconversional methods, together with Coats-Redfern, give the activation energy, pre-exponential factor and reaction mechanism. From them come the enthalpy, Gibbs free energy, entropy and exergy destruction, with Monte Carlo uncertainty propagation.
Machine learning
Regularised regression, support vector machines, tree ensembles and neural networks, validated by leaving out whole runs, predict untested blends and rank the variables that govern the conversion.
Expected results
- A thermokinetic, thermodynamic and exergy baseline for both residues and their blends, useful to design and scale up reactors.
- Four articles in indexed scientific journals.
- A working paper with the integrated analysis protocol, a conference talk and a final research report.
- TG and DTG curves and modelling code in an open-access repository.
- Two undergraduate theses and a student research club with two students.