Using Artificial Intelligence to control the production of H2
Lead Research Organisation:
CRANFIELD UNIVERSITY
Department Name: School of Water, Energy and Environment
Abstract
This project will utilise machine learning techniques to develop the online, real-time control and operation of a sorption enhanced steam reforming process with the aim of
producing CO2-abated H2.
A new small continuous flow reactor will be constructed and utilised in this project, it will have the capability of continuous bed material replenishment.
A key outcome of this research will be a comparison of a tuned PID, model predictive control, and machine learning algorithms in the operation of the reactor, with the aim
of optimising and improving the reactors operation such that the process could become more economical, safer, and produce higher yields and purities of H2.
producing CO2-abated H2.
A new small continuous flow reactor will be constructed and utilised in this project, it will have the capability of continuous bed material replenishment.
A key outcome of this research will be a comparison of a tuned PID, model predictive control, and machine learning algorithms in the operation of the reactor, with the aim
of optimising and improving the reactors operation such that the process could become more economical, safer, and produce higher yields and purities of H2.
Studentship Projects
| Project Reference | Relationship | Related To | Start | End | Student Name |
|---|---|---|---|---|---|
| EP/N509450/1 | 01/10/2016 | 30/09/2021 | |||
| 2269085 | Studentship | EP/N509450/1 | 03/06/2019 | 21/02/2023 | Pauline Nkulikiyinka |
| EP/R513027/1 | 01/10/2018 | 30/09/2023 | |||
| 2269085 | Studentship | EP/R513027/1 | 03/06/2019 | 21/02/2023 | Pauline Nkulikiyinka |
| Description | I proposed a novel machine-learning approach for the development of new combined sorbent catalyst materials (CSCM) for the sorption-enhanced steam methane reforming process (SE-SMR). The application of machine learning had not been applied to the concept of CSCM prior to my work I introduced the process of quantitative structure-property relationship (QSPR) analysis to a field that it hadn't been used in before- SE-SMR. Alongside the use of QSPR, I applied the application of multitask learning, to predict the behaviour of CSCM in the use of SE-SMR. This was a novel and interesting approach as it combined two separate databases to predict the molecular properties of a combined molecule I employed the process of data mining to gather experimental data for machine learning, in a more time-efficient manner I proposed a novel machine learning approach to develop a surrogate model for the monitoring of gas concentration readings in a hydrogen reactor, for when hardware sensors aren't applicable/suitable, for SE-SMR allowing for increasing the safety of operation Worked as part of a team of 18 academics, from UK and international universities to publish a review publication on machine learning in CCUS in a high-impact journal Organised and planned a multidisciplinary project with a Canadian PhD group, to study the application of machine learning in the development of metal-organic frameworks or hydrogen storage. (Cranfield/Concordia University). This was funded by the NERC grant. Collaboration focussed on novel research on the application of machine learning for the development of metal-organic frameworks for hydrogen storage, in a more time-efficient method compared to the traditional grand canonical monte Carlo computational method This research lead me to identify an optimal Metal-organic framework based on computational chemistry, highlighting a more streamlined method to screen MOFs for gas storage via machine learning |
| Exploitation Route | The use of QSPR in the combined sorbent and catalytic research area is completely novel, and the paper I published proposed specific conditions, methodologies and quantities, that would result in the most optimal material for SE-SMR. I feel this is a possibility to be explored further, whether it be proven right or discredited it would be interesting further research. |
| Sectors | Chemicals,Energy,Environment |
| Description | I presented at the Soapbox Science 2021, based on my research, to encourage younger primary school-aged children to go into STEM subjects |
| First Year Of Impact | 2021 |
| Sector | Education |
| Description | Machine Learning for the Discovery of Metal-Organic Frameworks for Hydrogen Storage Applications |
| Organisation | Concordia University |
| Country | Canada |
| Sector | Academic/University |
| PI Contribution | Metal-organic frameworks (MOFs) are a class of crystalline materials with ultrahigh porosity and high surface areas. With these properties, along with variability for both the organic and inorganic components of their structures, MOFs are of interest for potential applications in clean energy, most significantly as storage media for gases such as hydrogen and methane, and as high-capacity adsorbents to meet various separation needs. The use of quantitative structure-property/activity relationships (QSPRs - a form of machine learning) is an emerging and helpful mathematical tool that allows the link between physical or chemical properties to predict the behaviour or desired characteristic of a molecule. The purpose of this study is to exploit both fields in order to obtain the optimal MOF for hydrogen storage, as well as investigating what parameters affect the hydrogen uptake capabilities. Following the screening of optimal MOFs, they will be synthesized in the lab to test performance and validate the methodology. This research will be conducted by a PhD student (Paula Nkulikiyinka, who is funded by an EPSRC DTP grant) in collaboration with Dr A Howarth's research group at Concordia University, Canada. The research proposal has been discussed and agreed by both Dr Clough and Dr Howarth. |
| Collaborator Contribution | Collaboration focussed on novel research on the application of machine learning for the development of metal-organic frameworks for hydrogen storage, in a more time-efficient method compared to the traditional grand canonical monte Carlo computational method This research lead me to identify an optimal Metal-organic framework based on computational chemistry, highlighting a more streamlined method to screen MOFs for gas storage via machine learning |
| Impact | A short communication was written on the worked developed from this collaboration. |
| Start Year | 2022 |