Machine learning unveils the materials physical properties driving thermoelectric generators efficiency: half-Heuslers case
Fuente:
arXiv
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866913476552687616 |
|---|---|
| author | Tukmakova, Anastasiia Graziosi, Patrizio |
| author_facet | Tukmakova, Anastasiia Graziosi, Patrizio |
| contents | We report the machine learning (ML)-based approach allowing thermoelectric generator (TEG) efficiency evaluation directly from 5 parameters: 2 physical properties - carriers density and energy gap, and 3 engineering parameters - external load resistance, TEG hot side temperature and leg height. Then, we propose to use genetic algorithm to optimize the proposed parameters in a way to maximize TEG efficiency. To prepare data, physical properties of n- and p-type materials were computed by coupling Density Functional Theory to Boltzmann Transport, and used for Finite Elements simulations. TEG efficiency was evaluated from a finite elements model considering design, radiative heat loss, contacts, external load resistance and different combinations of materials, resulting in 5300 different scenarios. For ML model, physical properties and engineering parameters were used as input features, skipping transport coefficients, while TEG efficiency was a target. Model was built on gradient boosting algorithm, its performance was evaluated using the coefficient of determination that reached a value of 0.98 on test dataset. Features importance analysis revealed the most crucial features for Half-Heusler-based TEG efficiency: carriers density or Fermi level, indicating the predominant role of electrical conductivity and electronic part of electrical conductivity. Features that were less important, but able to increase model performance were: energy gap, lattice thermal conductivity, charge carrier relaxation time and carriers conductivity effective mass. Features showed no impact were: density of states effective mass, heat capacity, density, relative permittivity and leg width. The proposed approach can be applied for the identification of the most important physical properties and their optimal values, the optimization of TEG design and operation conditions in a way to maximize TEG efficiency. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2408_11949 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Machine learning unveils the materials physical properties driving thermoelectric generators efficiency: half-Heuslers case Tukmakova, Anastasiia Graziosi, Patrizio Materials Science We report the machine learning (ML)-based approach allowing thermoelectric generator (TEG) efficiency evaluation directly from 5 parameters: 2 physical properties - carriers density and energy gap, and 3 engineering parameters - external load resistance, TEG hot side temperature and leg height. Then, we propose to use genetic algorithm to optimize the proposed parameters in a way to maximize TEG efficiency. To prepare data, physical properties of n- and p-type materials were computed by coupling Density Functional Theory to Boltzmann Transport, and used for Finite Elements simulations. TEG efficiency was evaluated from a finite elements model considering design, radiative heat loss, contacts, external load resistance and different combinations of materials, resulting in 5300 different scenarios. For ML model, physical properties and engineering parameters were used as input features, skipping transport coefficients, while TEG efficiency was a target. Model was built on gradient boosting algorithm, its performance was evaluated using the coefficient of determination that reached a value of 0.98 on test dataset. Features importance analysis revealed the most crucial features for Half-Heusler-based TEG efficiency: carriers density or Fermi level, indicating the predominant role of electrical conductivity and electronic part of electrical conductivity. Features that were less important, but able to increase model performance were: energy gap, lattice thermal conductivity, charge carrier relaxation time and carriers conductivity effective mass. Features showed no impact were: density of states effective mass, heat capacity, density, relative permittivity and leg width. The proposed approach can be applied for the identification of the most important physical properties and their optimal values, the optimization of TEG design and operation conditions in a way to maximize TEG efficiency. |
| title | Machine learning unveils the materials physical properties driving thermoelectric generators efficiency: half-Heuslers case |
| topic | Materials Science |
| url | https://arxiv.org/abs/2408.11949 |