TL;DR
This paper introduces a novel network-level assessment framework and predictive model for inland waterway infrastructure maintenance. By adapting traditional time-series analysis to a unique dataset of waterway structures, the study provides a more efficient and proactive approach to optimizing maintenance budgets and resource allocation, a critical need currently underserved in this sector.
What Problem Does This Paper Address?
Maintaining inland waterway infrastructure, such as dikes and revetments, is crucial for transportation and economic development. However, these structures are subject to degradation from natural forces and usage, leading to performance decline. Traditional maintenance approaches are often reactive, costly, and inefficient, especially given limited funding. While network-level assessment has been successfully applied to road infrastructure for decades, its adoption and research in waterway systems have lagged significantly. This paper aims to bridge this gap by developing a predictive model for waterway improvement buildings that considers their condition at a network level, moving beyond individual structure assessments to optimize maintenance decisions and fund allocation across entire waterway systems.
Methods
The core of this research is a proposed network-level assessment framework comprising three modules: data collection, maintenance prediction, and maintenance decision. For the prediction module, the authors adapted the moving average method to create a "jurisdiction moving average" (JMA) method, specifically designed for their dataset's characteristics. They also explored linear regression (LR) and quadratic curve regression (QCR) models. The JMA method calculates the average proportion of buildings requiring maintenance within a jurisdiction over time, incorporating confidence intervals for increased reliability. The models were trained and tested using historical data, and their performance was evaluated using error metrics like RMSE, R², MAE, MSE, and p-values from t-tests. The final maintenance quantity for each jurisdiction was then calculated based on the predicted proportions.
Datasets
The study utilized a dataset of technical condition evaluation results for nearly 500 improvement buildings along the Yangtze River trunk waterway, collected over a five-year period (2017-2022). This dataset was organized by 11 jurisdictional units. The data was categorized into five technical condition levels, with the "third category" (requiring repair) being the focus for prediction. The data for 2017-2021 was used for model training and evaluation, while the 2022 data was reserved for model testing.
Key Insights and Ideas
- Network-level assessment is transferable: The study demonstrates that the principles of network-level assessment, successful in road infrastructure, can be effectively adapted to inland waterway systems.
- JMA offers enhanced credibility: The inclusion of confidence intervals in the Jurisdiction Moving Average (JMA) model significantly increases the credibility of its predictions, making it a preferred choice.
- Simplicity can be effective: Despite the availability of more complex models, simpler methods like JMA and LR proved to be more practical and accurate for this specific application, especially with limited data.
- Proactive maintenance is key: Shifting from reactive to predictive maintenance at a network level allows for better resource allocation and more efficient use of limited funds.
Key Results
The research found that both the Jurisdiction Moving Average (JMA) and Linear Regression (LR) models performed well and are recommended for use. The JMA model, in particular, was highlighted for its confidence intervals, which enhance the reliability of its predictions. While the Quadratic Curve Regression (QCR) model showed a higher goodness-of-fit, it exhibited significant errors, leading to lower predictive accuracy and generalization ability, thus making it less suitable. The study also indicated that the proposed network-level assessment technology for inland waterway maintenance is more efficient than existing methods. The JMA model's predictions for future maintenance quantities closely aligned with actual measurements, with confidence intervals encompassing most actual values.
Limitations and Future Work
The authors acknowledge that the predictive accuracy of the proposed model is limited by the size of the time-series sample. They suggest that continuously adding new evaluation data can refine model parameters and improve accuracy over time. The model's premise relies on the assumption of relatively stable waterway conditions within jurisdictions; it may not be suitable for scenarios with sudden and dramatic changes. Future work could involve incorporating soft computing techniques like Markov models, random fields, rough sets, and neural networks for enhanced predictive accuracy and expanding the scope to minimize overall maintenance costs and maximize network performance indicators. The study also emphasizes that while the model provides valuable predictive insights, comprehensive maintenance decision-making should also consider other relevant factors.
Summary updated