Bridge-CDE: A Physics-Informed Neural Controlled Differential Equation with Bridge-Specific Covariates for Irregularly Sampled Bridge Condition Index Prediction

The bridge condition index (BCI) is a widely used indicator for highway bridge maintenance and rehabilitation decisions. However, BCI records are collected at irregular inspection intervals, and bridges differ substantially in type, geometry, age, and load-carrying history. Imposing an equally spaced representation through interpolation can introduce artificial observations and distort the timing information that is central to deterioration modeling. This study presents Bridge-CDE, a physics-informed neural controlled differential equation (CDE) with bridge-specific covariates for irregularly sampled BCI prediction. Construction year, span, bridge length, width, standard and measured frequencies, technical condition, and load grade are encoded as a covariate vector. A bounded exponential baseline provides a physics-informed continuous path, and the CDE integrates along the path derivative from the latest inspection to the target age. A residual readout anchored to the recent BCI and a linear deterioration trend stabilizes learning on short inspection sequences. Experiments use 3308 BCI observations from 735 bridges in the Henan Province of China, grouped by bridge for model evaluation. Bridge-CDE achieves an MAE of 1.066, RMSE of 1.643, R2 of 0.955, and MAPE of 1.213%. It is competitive with the strong tabular baselines and records the lowest MAE, RMSE, and MAPE, whereas random forest has a marginally higher reported R2 (0.957). Permutation-based sensitivity analysis shows that bridge type, deck condition, measured frequency, and construction year are among the most influential external inputs.

Authors

Institutions

Publication Details

Journal
Materials
Published
2026-11-12
DOI
https://doi.org/10.3390/ma19224805
Primary Topic
Infrastructure Maintenance and Monitoring
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Bridge-CDE: A Physics-Informed Neural Controlled Differential Equation with Bridge-Specific Covariates for Irregularly Sampled Bridge Condition Index Prediction

Zhiju Chen, Bo Xu, Wentao Yang, Jie Li
Materials
Infrastructure Maintenance and Monitoring
article

Bridge-CDE: A Physics-Informed Neural Controlled Differential Equation with Bridge-Specific Covariates for Irregularly Sampled Bridge Condition Index Prediction

Zhiju Chen, Bo Xu, Wentao Yang, Jie Li
article en

Abstract

The bridge condition index (BCI) is a widely used indicator for highway bridge maintenance and rehabilitation decisions. However, BCI records are collected at irregular inspection intervals, and bridges differ substantially in type, geometry, age, and load-carrying history. Imposing an equally spaced representation through interpolation can introduce artificial observations and distort the timing information that is central to deterioration modeling. This study presents Bridge-CDE, a physics-informed neural controlled differential equation (CDE) with bridge-specific covariates for irregularly sampled BCI prediction. Construction year, span, bridge length, width, standard and measured frequencies, technical condition, and load grade are encoded as a covariate vector. A bounded exponential baseline provides a physics-informed continuous path, and the CDE integrates along the path derivative from the latest inspection to the target age. A residual readout anchored to the recent BCI and a linear deterioration trend stabilizes learning on short inspection sequences. Experiments use 3308 BCI observations from 735 bridges in the Henan Province of China, grouped by bridge for model evaluation. Bridge-CDE achieves an MAE of 1.066, RMSE of 1.643, R2 of 0.955, and MAPE of 1.213%. It is competitive with the strong tabular baselines and records the lowest MAE, RMSE, and MAPE, whereas random forest has a marginally higher reported R2 (0.957). Permutation-based sensitivity analysis shows that bridge type, deck condition, measured frequency, and construction year are among the most influential external inputs.

MaterialsVol. 19(22)
Zhengzhou University (CN), University of Glasgow (GB)
National Natural Science Foundation of China, Education Department of Henan Province
Industry, innovation and infrastructure
Infrastructure Maintenance and Monitoring
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon