Machine learning approach for predicting bridge components’ condition ratings
Information on bridge condition rating is critical to make decisions regarding rehabilitation or replacement of bridges. Currently, bridge components’ condition ratings are evaluated manually using inspection reports. Markov chain and Petri net models are most commonly used for predicting future val...
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Frontiers Media S.A.
2023-10-01
|
| Series: | Frontiers in Built Environment |
| Subjects: | |
| Online Access: | https://www.frontiersin.org/articles/10.3389/fbuil.2023.1254269/full |
| _version_ | 1797663339851022336 |
|---|---|
| author | Md. Manik Mia Sabarethinam Kameshwar |
| author_facet | Md. Manik Mia Sabarethinam Kameshwar |
| author_sort | Md. Manik Mia |
| collection | DOAJ |
| description | Information on bridge condition rating is critical to make decisions regarding rehabilitation or replacement of bridges. Currently, bridge components’ condition ratings are evaluated manually using inspection reports. Markov chain and Petri net models are most commonly used for predicting future values of bridge parameters, however, applicability of these models for a regional or statewide portfolio of bridges may be limited. The existing data based models have low prediction accuracy. Hence, a data and machine learning based approach is presented herein for predicting the future condition values of major components—deck, superstructure and substructure—in a portfolio of bridges with an objective to develop a more accurate approach. National Bridge Inventory (NBI) was used to get information on current and past bridge components’ condition from year 1992–2019 along with other parameters such as ownership, maintenance responsibility and age. After selecting important parameters, this data was used to train three RUSBoost based random forest models for predicting future values of deck, superstructure, and substructure conditions, respectively. The prediction accuracy of the developed models were found above 93%, thereby addressing the limitation of poor prediction accuracy of the existing studies. Additionally, the uncertainties associated with the random forest based predictions were quantified at the regional level and for individual bridges. On-system concrete pre-cast slab units and steel I-beam bridges in Louisiana were selected to demonstrate the proposed approach and predict bridge components condition ratings for years 2020 and 2021. |
| first_indexed | 2024-03-11T19:13:05Z |
| format | Article |
| id | doaj.art-657383b107bc476ca5dc79b66b9803b7 |
| institution | Directory Open Access Journal |
| issn | 2297-3362 |
| language | English |
| last_indexed | 2024-03-11T19:13:05Z |
| publishDate | 2023-10-01 |
| publisher | Frontiers Media S.A. |
| record_format | Article |
| series | Frontiers in Built Environment |
| spelling | doaj.art-657383b107bc476ca5dc79b66b9803b72023-10-09T09:18:38ZengFrontiers Media S.A.Frontiers in Built Environment2297-33622023-10-01910.3389/fbuil.2023.12542691254269Machine learning approach for predicting bridge components’ condition ratingsMd. Manik MiaSabarethinam KameshwarInformation on bridge condition rating is critical to make decisions regarding rehabilitation or replacement of bridges. Currently, bridge components’ condition ratings are evaluated manually using inspection reports. Markov chain and Petri net models are most commonly used for predicting future values of bridge parameters, however, applicability of these models for a regional or statewide portfolio of bridges may be limited. The existing data based models have low prediction accuracy. Hence, a data and machine learning based approach is presented herein for predicting the future condition values of major components—deck, superstructure and substructure—in a portfolio of bridges with an objective to develop a more accurate approach. National Bridge Inventory (NBI) was used to get information on current and past bridge components’ condition from year 1992–2019 along with other parameters such as ownership, maintenance responsibility and age. After selecting important parameters, this data was used to train three RUSBoost based random forest models for predicting future values of deck, superstructure, and substructure conditions, respectively. The prediction accuracy of the developed models were found above 93%, thereby addressing the limitation of poor prediction accuracy of the existing studies. Additionally, the uncertainties associated with the random forest based predictions were quantified at the regional level and for individual bridges. On-system concrete pre-cast slab units and steel I-beam bridges in Louisiana were selected to demonstrate the proposed approach and predict bridge components condition ratings for years 2020 and 2021.https://www.frontiersin.org/articles/10.3389/fbuil.2023.1254269/fullcondition ratingsrandom forestmachine learningLouisiana bridgesbootstrap uncertainties |
| spellingShingle | Md. Manik Mia Sabarethinam Kameshwar Machine learning approach for predicting bridge components’ condition ratings Frontiers in Built Environment condition ratings random forest machine learning Louisiana bridges bootstrap uncertainties |
| title | Machine learning approach for predicting bridge components’ condition ratings |
| title_full | Machine learning approach for predicting bridge components’ condition ratings |
| title_fullStr | Machine learning approach for predicting bridge components’ condition ratings |
| title_full_unstemmed | Machine learning approach for predicting bridge components’ condition ratings |
| title_short | Machine learning approach for predicting bridge components’ condition ratings |
| title_sort | machine learning approach for predicting bridge components condition ratings |
| topic | condition ratings random forest machine learning Louisiana bridges bootstrap uncertainties |
| url | https://www.frontiersin.org/articles/10.3389/fbuil.2023.1254269/full |
| work_keys_str_mv | AT mdmanikmia machinelearningapproachforpredictingbridgecomponentsconditionratings AT sabarethinamkameshwar machinelearningapproachforpredictingbridgecomponentsconditionratings |