Physics-Informed Neural Networks with Group Contribution Methods
Thermophysical properties of organic compounds are used in countless scientific, engineering, and industrial settings in developing theories, designing new systems and devices, analyzing costs and risks, and improving existing infrastructure. Often, due to costs, safety, prior interest, or procedura...
Gespeichert in:
Veröffentlicht in: | Journal of chemical theory and computation 2023-07, Vol.19 (13), p.4163-4171 |
---|---|
Hauptverfasser: | , , , |
Format: | Artikel |
Sprache: | eng |
Schlagworte: | |
Online-Zugang: | Volltext |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
container_end_page | 4171 |
---|---|
container_issue | 13 |
container_start_page | 4163 |
container_title | Journal of chemical theory and computation |
container_volume | 19 |
creator | Babaei, Mohammad Reza Stone, Ryan IV, Thomas Allen Knotts Hedengren, John |
description | Thermophysical properties of organic compounds are used in countless scientific, engineering, and industrial settings in developing theories, designing new systems and devices, analyzing costs and risks, and improving existing infrastructure. Often, due to costs, safety, prior interest, or procedural difficulties, experimental values for desired properties are not available and must be predicted. The literature is filled with prediction techniques, but even the best traditional methods have significant errors compared to what is possible considering experimental uncertainty. Recently, machine learning and artificial intelligence techniques have been applied to the property prediction problem, but the examples to date do not extrapolate well outside the data set used for training the model. This work demonstrates a solution to this problem by combining chemistry and physics when training the model and builds upon prior traditional and machine learning methods. Two case studies are presented. The first is for parachor which is used for surface tension prediction. Surface tensions are needed to design distillation columns, adsorption processes, gas–liquid reactors, liquid–liquid extractors, improve oil reservoir recovery, and undertake environmental impact studies or remediation actions. A set of 277 compounds is divided into training, validation, and test sets, and a multilayered physics-informed neural network (PINN) is developed. The results demonstrate that better extrapolation by deep learning models can be developed by adding in physics-based constraints. Second, a set of 1600 compounds is utilized for training, validating, and testing a PINN to improve normal boiling point predictions based on group contribution methods and physics-based constraints. The results show that the PINN performs better than any other method with a normal boiling point mean absolute error of 6.95 °C on training and 11.2 °C on test data. Key observations are that (1) a balanced split by compound type is important to have representative compound families in each of the train, validation, and test sets and (2) constraining group contributions being positive improves predictions on the test set. While this work demonstrates improvements for only surface tension and normal boiling point, the results offer significant hope that PINNs can improve prediction of other relevant thermophysical properties over existing approaches. |
doi_str_mv | 10.1021/acs.jctc.3c00195 |
format | Article |
fullrecord | <record><control><sourceid>proquest_cross</sourceid><recordid>TN_cdi_proquest_miscellaneous_2824684854</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2824684854</sourcerecordid><originalsourceid>FETCH-LOGICAL-a364t-e3a4c0cbe13f7bfa6230d3debe4481ef570005157c5aee053975b1307e82966a3</originalsourceid><addsrcrecordid>eNp1kDtPwzAURi0EoqWwM6FILAykXL_y2EAVlErlMcBsOY6jpqRxsR1V_fe49DEgMd07nO-7VwehSwxDDATfSeWGc-XVkCoAnPMj1Mec5XGekOT4sOOsh86cmwNQygg9RT2akpzmKe-j-_fZ2tXKxZO2Mnahy-hVd1Y2YfiVsV8uWtV-Fo2t6ZbRyLTe1kXna9NGL9rPTOnO0UklG6cvdnOAPp8eP0bP8fRtPBk9TGNJE-ZjTSVToAqNaZUWlUwIhZKWutCMZVhXPAUAjnmquNQa-Oa7AlNIdUbyJJF0gG62vUtrvjvtvFjUTummka02nRMkIyzJWMZZQK__oHPT2TZ8FygGGQDL00DBllLWOGd1JZa2Xki7FhjExq4IdsXGrtjZDZGrXXFXBFWHwF5nAG63wG90f_Tfvh8PS4T-</addsrcrecordid><sourcetype>Aggregation Database</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype><pqid>2840800497</pqid></control><display><type>article</type><title>Physics-Informed Neural Networks with Group Contribution Methods</title><source>American Chemical Society Journals</source><creator>Babaei, Mohammad Reza ; Stone, Ryan ; IV, Thomas Allen Knotts ; Hedengren, John</creator><creatorcontrib>Babaei, Mohammad Reza ; Stone, Ryan ; IV, Thomas Allen Knotts ; Hedengren, John</creatorcontrib><description>Thermophysical properties of organic compounds are used in countless scientific, engineering, and industrial settings in developing theories, designing new systems and devices, analyzing costs and risks, and improving existing infrastructure. Often, due to costs, safety, prior interest, or procedural difficulties, experimental values for desired properties are not available and must be predicted. The literature is filled with prediction techniques, but even the best traditional methods have significant errors compared to what is possible considering experimental uncertainty. Recently, machine learning and artificial intelligence techniques have been applied to the property prediction problem, but the examples to date do not extrapolate well outside the data set used for training the model. This work demonstrates a solution to this problem by combining chemistry and physics when training the model and builds upon prior traditional and machine learning methods. Two case studies are presented. The first is for parachor which is used for surface tension prediction. Surface tensions are needed to design distillation columns, adsorption processes, gas–liquid reactors, liquid–liquid extractors, improve oil reservoir recovery, and undertake environmental impact studies or remediation actions. A set of 277 compounds is divided into training, validation, and test sets, and a multilayered physics-informed neural network (PINN) is developed. The results demonstrate that better extrapolation by deep learning models can be developed by adding in physics-based constraints. Second, a set of 1600 compounds is utilized for training, validating, and testing a PINN to improve normal boiling point predictions based on group contribution methods and physics-based constraints. The results show that the PINN performs better than any other method with a normal boiling point mean absolute error of 6.95 °C on training and 11.2 °C on test data. Key observations are that (1) a balanced split by compound type is important to have representative compound families in each of the train, validation, and test sets and (2) constraining group contributions being positive improves predictions on the test set. While this work demonstrates improvements for only surface tension and normal boiling point, the results offer significant hope that PINNs can improve prediction of other relevant thermophysical properties over existing approaches.</description><identifier>ISSN: 1549-9618</identifier><identifier>EISSN: 1549-9626</identifier><identifier>DOI: 10.1021/acs.jctc.3c00195</identifier><identifier>PMID: 37293975</identifier><language>eng</language><publisher>United States: American Chemical Society</publisher><subject>Artificial intelligence ; Boiling points ; Cost analysis ; Deep learning ; Distillation ; Environmental impact ; Machine learning ; Neural networks ; Nuclear safety ; Organic compounds ; Physics ; Predictions ; Surface tension ; Test sets ; Thermodynamics ; Thermophysical properties ; Training</subject><ispartof>Journal of chemical theory and computation, 2023-07, Vol.19 (13), p.4163-4171</ispartof><rights>2023 American Chemical Society</rights><rights>Copyright American Chemical Society Jul 11, 2023</rights><lds50>peer_reviewed</lds50><woscitedreferencessubscribed>false</woscitedreferencessubscribed><citedby>FETCH-LOGICAL-a364t-e3a4c0cbe13f7bfa6230d3debe4481ef570005157c5aee053975b1307e82966a3</citedby><cites>FETCH-LOGICAL-a364t-e3a4c0cbe13f7bfa6230d3debe4481ef570005157c5aee053975b1307e82966a3</cites><orcidid>0000-0002-5535-5277 ; 0000-0001-6248-4459 ; 0000-0003-0709-3599</orcidid></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><linktopdf>$$Uhttps://pubs.acs.org/doi/pdf/10.1021/acs.jctc.3c00195$$EPDF$$P50$$Gacs$$H</linktopdf><linktohtml>$$Uhttps://pubs.acs.org/doi/10.1021/acs.jctc.3c00195$$EHTML$$P50$$Gacs$$H</linktohtml><link.rule.ids>314,780,784,2765,27076,27924,27925,56738,56788</link.rule.ids><backlink>$$Uhttps://www.ncbi.nlm.nih.gov/pubmed/37293975$$D View this record in MEDLINE/PubMed$$Hfree_for_read</backlink></links><search><creatorcontrib>Babaei, Mohammad Reza</creatorcontrib><creatorcontrib>Stone, Ryan</creatorcontrib><creatorcontrib>IV, Thomas Allen Knotts</creatorcontrib><creatorcontrib>Hedengren, John</creatorcontrib><title>Physics-Informed Neural Networks with Group Contribution Methods</title><title>Journal of chemical theory and computation</title><addtitle>J. Chem. Theory Comput</addtitle><description>Thermophysical properties of organic compounds are used in countless scientific, engineering, and industrial settings in developing theories, designing new systems and devices, analyzing costs and risks, and improving existing infrastructure. Often, due to costs, safety, prior interest, or procedural difficulties, experimental values for desired properties are not available and must be predicted. The literature is filled with prediction techniques, but even the best traditional methods have significant errors compared to what is possible considering experimental uncertainty. Recently, machine learning and artificial intelligence techniques have been applied to the property prediction problem, but the examples to date do not extrapolate well outside the data set used for training the model. This work demonstrates a solution to this problem by combining chemistry and physics when training the model and builds upon prior traditional and machine learning methods. Two case studies are presented. The first is for parachor which is used for surface tension prediction. Surface tensions are needed to design distillation columns, adsorption processes, gas–liquid reactors, liquid–liquid extractors, improve oil reservoir recovery, and undertake environmental impact studies or remediation actions. A set of 277 compounds is divided into training, validation, and test sets, and a multilayered physics-informed neural network (PINN) is developed. The results demonstrate that better extrapolation by deep learning models can be developed by adding in physics-based constraints. Second, a set of 1600 compounds is utilized for training, validating, and testing a PINN to improve normal boiling point predictions based on group contribution methods and physics-based constraints. The results show that the PINN performs better than any other method with a normal boiling point mean absolute error of 6.95 °C on training and 11.2 °C on test data. Key observations are that (1) a balanced split by compound type is important to have representative compound families in each of the train, validation, and test sets and (2) constraining group contributions being positive improves predictions on the test set. While this work demonstrates improvements for only surface tension and normal boiling point, the results offer significant hope that PINNs can improve prediction of other relevant thermophysical properties over existing approaches.</description><subject>Artificial intelligence</subject><subject>Boiling points</subject><subject>Cost analysis</subject><subject>Deep learning</subject><subject>Distillation</subject><subject>Environmental impact</subject><subject>Machine learning</subject><subject>Neural networks</subject><subject>Nuclear safety</subject><subject>Organic compounds</subject><subject>Physics</subject><subject>Predictions</subject><subject>Surface tension</subject><subject>Test sets</subject><subject>Thermodynamics</subject><subject>Thermophysical properties</subject><subject>Training</subject><issn>1549-9618</issn><issn>1549-9626</issn><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2023</creationdate><recordtype>article</recordtype><recordid>eNp1kDtPwzAURi0EoqWwM6FILAykXL_y2EAVlErlMcBsOY6jpqRxsR1V_fe49DEgMd07nO-7VwehSwxDDATfSeWGc-XVkCoAnPMj1Mec5XGekOT4sOOsh86cmwNQygg9RT2akpzmKe-j-_fZ2tXKxZO2Mnahy-hVd1Y2YfiVsV8uWtV-Fo2t6ZbRyLTe1kXna9NGL9rPTOnO0UklG6cvdnOAPp8eP0bP8fRtPBk9TGNJE-ZjTSVToAqNaZUWlUwIhZKWutCMZVhXPAUAjnmquNQa-Oa7AlNIdUbyJJF0gG62vUtrvjvtvFjUTummka02nRMkIyzJWMZZQK__oHPT2TZ8FygGGQDL00DBllLWOGd1JZa2Xki7FhjExq4IdsXGrtjZDZGrXXFXBFWHwF5nAG63wG90f_Tfvh8PS4T-</recordid><startdate>20230711</startdate><enddate>20230711</enddate><creator>Babaei, Mohammad Reza</creator><creator>Stone, Ryan</creator><creator>IV, Thomas Allen Knotts</creator><creator>Hedengren, John</creator><general>American Chemical Society</general><scope>NPM</scope><scope>AAYXX</scope><scope>CITATION</scope><scope>7SC</scope><scope>7SR</scope><scope>7U5</scope><scope>8BQ</scope><scope>8FD</scope><scope>JG9</scope><scope>JQ2</scope><scope>L7M</scope><scope>L~C</scope><scope>L~D</scope><scope>7X8</scope><orcidid>https://orcid.org/0000-0002-5535-5277</orcidid><orcidid>https://orcid.org/0000-0001-6248-4459</orcidid><orcidid>https://orcid.org/0000-0003-0709-3599</orcidid></search><sort><creationdate>20230711</creationdate><title>Physics-Informed Neural Networks with Group Contribution Methods</title><author>Babaei, Mohammad Reza ; Stone, Ryan ; IV, Thomas Allen Knotts ; Hedengren, John</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a364t-e3a4c0cbe13f7bfa6230d3debe4481ef570005157c5aee053975b1307e82966a3</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2023</creationdate><topic>Artificial intelligence</topic><topic>Boiling points</topic><topic>Cost analysis</topic><topic>Deep learning</topic><topic>Distillation</topic><topic>Environmental impact</topic><topic>Machine learning</topic><topic>Neural networks</topic><topic>Nuclear safety</topic><topic>Organic compounds</topic><topic>Physics</topic><topic>Predictions</topic><topic>Surface tension</topic><topic>Test sets</topic><topic>Thermodynamics</topic><topic>Thermophysical properties</topic><topic>Training</topic><toplevel>peer_reviewed</toplevel><toplevel>online_resources</toplevel><creatorcontrib>Babaei, Mohammad Reza</creatorcontrib><creatorcontrib>Stone, Ryan</creatorcontrib><creatorcontrib>IV, Thomas Allen Knotts</creatorcontrib><creatorcontrib>Hedengren, John</creatorcontrib><collection>PubMed</collection><collection>CrossRef</collection><collection>Computer and Information Systems Abstracts</collection><collection>Engineered Materials Abstracts</collection><collection>Solid State and Superconductivity Abstracts</collection><collection>METADEX</collection><collection>Technology Research Database</collection><collection>Materials Research Database</collection><collection>ProQuest Computer Science Collection</collection><collection>Advanced Technologies Database with Aerospace</collection><collection>Computer and Information Systems Abstracts Academic</collection><collection>Computer and Information Systems Abstracts Professional</collection><collection>MEDLINE - Academic</collection><jtitle>Journal of chemical theory and computation</jtitle></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext</fulltext></delivery><addata><au>Babaei, Mohammad Reza</au><au>Stone, Ryan</au><au>IV, Thomas Allen Knotts</au><au>Hedengren, John</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Physics-Informed Neural Networks with Group Contribution Methods</atitle><jtitle>Journal of chemical theory and computation</jtitle><addtitle>J. Chem. Theory Comput</addtitle><date>2023-07-11</date><risdate>2023</risdate><volume>19</volume><issue>13</issue><spage>4163</spage><epage>4171</epage><pages>4163-4171</pages><issn>1549-9618</issn><eissn>1549-9626</eissn><abstract>Thermophysical properties of organic compounds are used in countless scientific, engineering, and industrial settings in developing theories, designing new systems and devices, analyzing costs and risks, and improving existing infrastructure. Often, due to costs, safety, prior interest, or procedural difficulties, experimental values for desired properties are not available and must be predicted. The literature is filled with prediction techniques, but even the best traditional methods have significant errors compared to what is possible considering experimental uncertainty. Recently, machine learning and artificial intelligence techniques have been applied to the property prediction problem, but the examples to date do not extrapolate well outside the data set used for training the model. This work demonstrates a solution to this problem by combining chemistry and physics when training the model and builds upon prior traditional and machine learning methods. Two case studies are presented. The first is for parachor which is used for surface tension prediction. Surface tensions are needed to design distillation columns, adsorption processes, gas–liquid reactors, liquid–liquid extractors, improve oil reservoir recovery, and undertake environmental impact studies or remediation actions. A set of 277 compounds is divided into training, validation, and test sets, and a multilayered physics-informed neural network (PINN) is developed. The results demonstrate that better extrapolation by deep learning models can be developed by adding in physics-based constraints. Second, a set of 1600 compounds is utilized for training, validating, and testing a PINN to improve normal boiling point predictions based on group contribution methods and physics-based constraints. The results show that the PINN performs better than any other method with a normal boiling point mean absolute error of 6.95 °C on training and 11.2 °C on test data. Key observations are that (1) a balanced split by compound type is important to have representative compound families in each of the train, validation, and test sets and (2) constraining group contributions being positive improves predictions on the test set. While this work demonstrates improvements for only surface tension and normal boiling point, the results offer significant hope that PINNs can improve prediction of other relevant thermophysical properties over existing approaches.</abstract><cop>United States</cop><pub>American Chemical Society</pub><pmid>37293975</pmid><doi>10.1021/acs.jctc.3c00195</doi><tpages>9</tpages><orcidid>https://orcid.org/0000-0002-5535-5277</orcidid><orcidid>https://orcid.org/0000-0001-6248-4459</orcidid><orcidid>https://orcid.org/0000-0003-0709-3599</orcidid></addata></record> |
fulltext | fulltext |
identifier | ISSN: 1549-9618 |
ispartof | Journal of chemical theory and computation, 2023-07, Vol.19 (13), p.4163-4171 |
issn | 1549-9618 1549-9626 |
language | eng |
recordid | cdi_proquest_miscellaneous_2824684854 |
source | American Chemical Society Journals |
subjects | Artificial intelligence Boiling points Cost analysis Deep learning Distillation Environmental impact Machine learning Neural networks Nuclear safety Organic compounds Physics Predictions Surface tension Test sets Thermodynamics Thermophysical properties Training |
title | Physics-Informed Neural Networks with Group Contribution Methods |
url | https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2024-12-24T18%3A10%3A38IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-proquest_cross&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=Physics-Informed%20Neural%20Networks%20with%20Group%20Contribution%20Methods&rft.jtitle=Journal%20of%20chemical%20theory%20and%20computation&rft.au=Babaei,%20Mohammad%20Reza&rft.date=2023-07-11&rft.volume=19&rft.issue=13&rft.spage=4163&rft.epage=4171&rft.pages=4163-4171&rft.issn=1549-9618&rft.eissn=1549-9626&rft_id=info:doi/10.1021/acs.jctc.3c00195&rft_dat=%3Cproquest_cross%3E2824684854%3C/proquest_cross%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_pqid=2840800497&rft_id=info:pmid/37293975&rfr_iscdi=true |