2026-09-18T18:42:25Zhttps://dspace.mit.edu/server/oai/requestoai:dspace.mit.edu:1721.1/1450622022-08-30T03:11:20Zcom_1721.1_7582com_1721.1_7581col_1721.1_131022
Green, William H.
Forsuelo, Michael
Massachusetts Institute of Technology. Department of Chemical Engineering
2022-08-29T16:30:24Z
2022-08-29T16:30:24Z
2022-05
2022-05-19T15:34:55.624Z
https://hdl.handle.net/1721.1/145062
This thesis covers the diverse investigations of high-fidelity molecular property prediction with Message Passing Neural Networks (MPNNs) and nanoscale polymer fouling experimentation with Quartz Crystal Microbalances (QCMs).
Message Passing Neural Networks are promising deep learning architectures for chemical property prediction. The state-of-the-art MPNN known as Chemprop is adapted in this thesis for the prediction of infrared spectra. A novel loss function and architectural extensions are identified which increase the predictive performance of spectra relative to the original architecture. These extensions are collected into the software package Chemprop-IR. The Chemprop-IR architecture holds promise in the prediction of experimental infrared spectra with little to no peak shifts within the fingerprint region. Uncertainty estimation frameworks within Chemprop are also extended in this thesis. A pretraining procedure and expanded readout structure are identified which improve the accuracy of predicted properties and estimated aleatoric uncertainties while maintaining scalable predictivity.
Process fouling is a pervasive problem in ethylene plants. Foulant, or the undesired accumulation of material, has been identified throughout process units of ethylene plants. Fouling can result in a multitude of challenges in plant production and operation. A few challenges include reduced thermal separation efficiencies and increased process safety concerns. Identifying the root causes to various forms of process fouling is essential to identifying the appropriate mitigation strategies. In particular, polymer fouling within ethylene plants is probed experimentally by Quartz Crystal Microbalance with Dissipation monitoring (QCM-D). QCM-D is used to assess the potential for ethylene plant products to form polymer foulant or support growth of pre-existing polymer foulant at approximate process conditions. The effects of molecular oxygen and antioxidant inhibitors on foulant growth dynamics are also addressed.
Ph.D.
Massachusetts Institute of Technology
In Copyright - Educational Use Permitted
Copyright MIT
http://rightsstatements.org/page/InC-EDU/1.0/
Investigations into Message Passing Neural Networks and Polymer Fouling
Thesis
application/pdf
Doctoral
Doctor of Philosophy
Publication
unknown
unknown
<Publication xmlns="https://www.openaire.eu/cerif-profile/1.1/" id="cbd350bc-2796-49e3-ac06-975ec5e50c70">
<Type xmlns="https://www.openaire.eu/cerif-profile/vocab/COAR_Publication_Types">http://purl.org/coar/resource_type/c_1843</Type>
<Title>Investigations into Message Passing Neural Networks and Polymer Fouling</Title>
<PublishedIn>
<Publication>
</Publication>
</PublishedIn>
<PublicationDate>2022-05</PublicationDate>
<Authors>
<Author>
<DisplayName>Forsuelo, Michael</DisplayName>
<Affiliation>
<OrgUnit>
</OrgUnit>
</Affiliation>
</Author>
</Authors>
<Editors>
</Editors>
<Publishers>
<Publisher>
<DisplayName>Massachusetts Institute of Technology</DisplayName>
<OrgUnit />
</Publisher>
</Publishers>
<License>http://rightsstatements.org/page/InC-EDU/1.0/</License>
<Abstract>This thesis covers the diverse investigations of high-fidelity molecular property prediction with Message Passing Neural Networks (MPNNs) and nanoscale polymer fouling experimentation with Quartz Crystal Microbalances (QCMs).
Message Passing Neural Networks are promising deep learning architectures for chemical property prediction. The state-of-the-art MPNN known as Chemprop is adapted in this thesis for the prediction of infrared spectra. A novel loss function and architectural extensions are identified which increase the predictive performance of spectra relative to the original architecture. These extensions are collected into the software package Chemprop-IR. The Chemprop-IR architecture holds promise in the prediction of experimental infrared spectra with little to no peak shifts within the fingerprint region. Uncertainty estimation frameworks within Chemprop are also extended in this thesis. A pretraining procedure and expanded readout structure are identified which improve the accuracy of predicted properties and estimated aleatoric uncertainties while maintaining scalable predictivity.
Process fouling is a pervasive problem in ethylene plants. Foulant, or the undesired accumulation of material, has been identified throughout process units of ethylene plants. Fouling can result in a multitude of challenges in plant production and operation. A few challenges include reduced thermal separation efficiencies and increased process safety concerns. Identifying the root causes to various forms of process fouling is essential to identifying the appropriate mitigation strategies. In particular, polymer fouling within ethylene plants is probed experimentally by Quartz Crystal Microbalance with Dissipation monitoring (QCM-D). QCM-D is used to assess the potential for ethylene plant products to form polymer foulant or support growth of pre-existing polymer foulant at approximate process conditions. The effects of molecular oxygen and antioxidant inhibitors on foulant growth dynamics are also addressed.</Abstract>
<Access xmlns="http://purl.org/coar/access_right"
>
</Access>
</Publication>