WP1 - Condition assessment and damage modelling
Main researcher: ir. Constantijn Martens
Affiliation: KU Leuven
Title PhD dissertation: Condition assessment and damage modelling of corroding reinforced concrete structures through Bayesian coupled learning model
Supervisors: Prof. dr. ir.-arch. Els Verstrynge, prof. dr. ir. Robby Caspeele
Publications: link
Work package description
As inspection data may concern data on specific point locations (e.g. core drilling, cover thickness, chloride content), grid data with a specific mesh size (e.g. potential mapping ) as well as zonal data (visual damage mapping), techniques are needed that allow for the integration of data sets with different spatial density. This will be achieved by modelling the damage by a random field. This results in zones with specific damage levels and related uncertainties.
A scheme for hierarchical classification of input data will be set up, dependent on their importance in the modelling and decision process, to enable dealing with lack of data and related uncertainties as encountered in real cases.
The identified damage levels will be linked to mechanical effects (rebar section reduction, bond loss, and concrete cracking). In this project, a practical approach is sought with the development of a Bayesian framework, using empirical relations and accounting for uncertainties (material, model, and measurement). In the adopted Bayesian methodology, the prior uncertainty of the damage model is updated to a posterior uncertainty on the basis of inspections.
Model uncertainties will be derived on the basis of comparisons with a limited number of experimental tests which are performed under controlled conditions (see WP4) and selected on-site case studies which might exhibit combined effects of mechanical and environment loading (cf. WP7). The latter will allow to assess model uncertainties in real conditions, by means of periodic inspection and vibration-based monitoring (WP3). The linking of the corrosion damage to mechanical effects leads to a parametrization of damage which is used as input parameters for performance prediction (WP2).
Research results
As part of WP1, work was carried out on an advanced, data-driven approach to estimate corrosion damage based on easily measurable damage parameters, including crack width.
Within this framework, the KUL-edCCRC database was established — an open-source database compiling all relevant experimental results from KU Leuven regarding the relationship between surface cracking in concrete and corrosion of the steel reinforcement. To study this relationship in more detail, an innovative experimental setup was developed, allowing concrete specimens to be exposed to natural corrosion conditions through wet-dry cycles, similar to typical outdoor environments (see photo). Both cracked and uncracked specimens were sprayed weekly with a salt solution (de-icing salts) over a period of two years, enabling a detailed analysis of the influence of cracks on the corrosion process.
The evolution of common parameters — including crack width, corrosion rate, potential, and electrical resistivity — was monitored over time for the various specimens. A key finding is that the corrosion rate is significantly higher in the vicinity of cracks, a situation frequently encountered in real structures. This information forms the basis for developing practical models capable of predicting corrosion rates based on visual damage parameters — something that has been hardly achievable in practice until now.
To make these predictions robust and reliable, a Bayesian approach was adopted. This is a probabilistic modeling method that not only allows for predictions but also explicitly incorporates uncertainties. An important feature is that the models can update themselves based on new information. For example, when new crack measurements (or data from similar conditions) become available through inspections, the model can automatically adjust the corrosion predictions to the specific structure. In this way, it evolves into a self-learning system that becomes increasingly reliable over time. This approach was applied to the relationship between crack width and corrosion level (total section loss of the reinforcement bar), where existing relations from the literature were refined and adapted through Bayesian updating. The proposed methodology therefore not only provides new predictive models but also enables realistic calibration of existing models using additional experimental data. This makes them more suitable for inspection and maintenance decision-making. An additional advantage is that the methodology is fully compatible with existing inspection practices. Visual damage inspections are already routinely carried out by many public authorities and infrastructure managers. What this research adds is the ability to directly link such inspection data to corrosion risks, something that is rarely or never done today. Moreover, by leveraging new robotic and drone technologies, this approach can be applied efficiently and cost-effectively on a large scale.
In addition to the laboratory research, several case studies were also examined, in which existing concrete structures were analyzed under real-world exposure conditions (see photo). These cases provided a valuable complement to the experimental program and offered further insight into damage development under natural exposure.
main objectives
This workpackage aims for the following Deliverables (D) and Milestones (M):
- D1.1 – Report of an overview of available corrosion propagation models, related inspection techniques and their suitability for incorporation into the developed methodology
- D1.2 – Report which describes the hierarchical classification of input data and related uncertainties to deal with lack of inspection data in practical situations
- D1.3 – Algorithm for the Bayesian updating of damage processes on the basis of additional data with different spatial distribution and density
- M1.1 – Implemented techniques for integration of data sets with different spatial distribution and data density into the updating procedure (required input for WPs 2, 3 and 5)
- M1.2 – Implemented procedure to predict spatial distribution of structural cross-sectional properties for the performance prediction models based on the inspection data and developed degradation/damage models (required input for WP2)