Comparative study of corrosion-based service life prediction of reinforced concrete structures using traditional and machine learning approach

Abstract

Purpose

Accurate predictions of the steady-state corrosion phase and service life to achieve specific safety limits are crucial for assessing the service of reinforced concrete (RC) structures. Forecasting the service life (SL) of structures is imperative for devising maintenance and repair strategy plans. The optimization of maintenance strategies serves to prolong asset life, mitigate asset failures, minimize repair costs and enhance health and safety standards for society.

Design/methodology/approach

The well-known empirical conventional (traditional) approaches and machine learning (ML)-based SL prediction models were presented and compared. A comprehensive parametric study was conducted on existing models, considering real-world conditions as reported in the literature. The analysis of traditional and ML models underscored their respective limitations.

Findings

Empirical models have been developed by considering simplified assumptions and relying on factors such as corrosion rate, steel reinforcement diameter and concrete cover depth, utilizing fundamental mathematical formulas. The growth of ML in the structural domain has been identified and highlighted. The ML can capture complex relationships between input and output variables. The performance of ML in corrosion and service life evaluation has been satisfactory. The limitations of ML techniques are discussed, and its open challenges are identified, along with insights into the future direction to develop more accurate and reliable models.

Practical implications

To enhance the traditional modeling of service life, key areas for future research have been highlighted. These include addressing the heterogeneous properties of concrete, the permeability of concrete and incorporating the interaction between temperature and bond-slip effect, which has been overlooked in existing models. Though the performance of the ML model in service life assessment is satisfactory, models overlooked some parameters, such as the material characterization and chemical composition of individual parameters, which play a significant role. As a recommendation, further research should take these factors into account as input parameters and strive to develop models with superior predictive capabilities.

Originality/value

Recent deployment has revealed that ML algorithms can grasp complex relationships among key factors impacting deterioration and offer precise evaluations of remaining SL without relying on traditional models. Incorporation of more comprehensive and diverse data sources toward potential future directions in the RC structural domain can provide valuable insights to decision-makers, guiding their efforts toward the creation of even more resilient, reliable, cost-efficient and eco-friendly RC structures.

Citation

Rajender, A., Samanta, A.K. and Paral, A. (2024), "Comparative study of corrosion-based service life prediction of reinforced concrete structures using traditional and machine learning approach", International Journal of Structural Integrity, Vol. ahead-of-print No. ahead-of-print. https://doi.org/10.1108/IJSI-02-2024-0018