Files in this item



application/pdfLIU-DISSERTATION-2019.pdf (5MB)Restricted Access
(no description provided)PDF


Title:Machine learning-based analytics of structured and unstructured data for enhanced bridge deterioration prediction
Author(s):Liu, Kaijian
Director of Research:El-Gohary, Nora
Doctoral Committee Chair(s):El-Gohary, Nora
Doctoral Committee Member(s):El-Rayes, Khaled; Zhai, ChengXiang; Liu, Liang Y; Golparvar-Fard, Mani
Department / Program:Civil & Environmental Eng
Discipline:Civil Engineering
Degree Granting Institution:University of Illinois at Urbana-Champaign
Subject(s):Bridge deterioration prediction
Data analytics
Machine learning
Natural language processing
Information extraction
Dependency parsing
Data linking
Data fusion
Abstract:The increasing availability of heterogeneous bridge data from multiple sources opens unprecedented opportunities for data analytics to better predict bridge deterioration for supporting enhanced bridge maintenance decision making. Such data include structured National Bridge Inventory (NBI) and National Bridge Elements (NBE) data, structured traffic and weather data, and unstructured textual bridge inspection reports. However, despite the availability of the data, existing data-driven prediction methods mostly learn from abstract inventory data (e.g., the NBI data which describe bridge conditions by condition ratings) from a single source – missing the opportunity of leveraging the wealth of unstructured textual inspection reports and the diverseness of the multi-source data for enhanced deterioration prediction. To capitalize on this opportunity, a novel bridge data analytics framework is proposed. The proposed framework is composed of six primary components: (1) a bridge deterioration knowledge ontology for facilitating semantic information and relation extraction from textual bridge inspection reports based on content and domain-specific meaning; (2) a semi-supervised machine learning-based semantic information extraction method for extracting information entities that describe bridge conditions and maintenance actions from the reports; (3) a supervised machine learning-based semantic relation extraction method for extracting dependency relations from the reports to link the extracted, yet isolated, information entities into concepts and to represent the semantically-low concepts in a semantically-rich structured way; (4) an unsupervised machine learning-based data linking method for linking the data records that are extracted from the reports and refer to the same entity; (5) a hybrid data fusion method for fusing the linked data records into a unified representation and for, subsequently, integrating the fused data with the other types of structured data (i.e., NBI and NBE data, as well as traffic and weather data); and (6) a data-driven, deep learning-based bridge deterioration prediction method for learning from the integrated bridge data to predict the condition ratings of the primary bridge components and to predict the quantities of specific bridge element-level deficiencies. The performance of the proposed framework was evaluated in predicting the deterioration of the state-owned bridges in Washington. It achieved a macro-precision and macro-recall of 89.9% and 85.8% when predicting the future condition ratings of the primary bridge components (i.e., decks, superstructures, and substructures), and achieved a root mean square error, coefficient of variation, and coefficient of determination of 1.3, 27.6%, and 0.89, respectively, when predicting the future quantities of specific bridge element-level deficiencies. The experimental results demonstrated the promise of the proposed framework.
Issue Date:2019-10-03
Rights Information:Copyright 2019 Kaijian Liu
Date Available in IDEALS:2020-03-02
Date Deposited:2019-12

This item appears in the following Collection(s)

Item Statistics