Service-life prediction of concrete bridge decks using case-based reasoning

READ THESE TERMS AND CONDITIONS CAREFULLY BEFORE USING THIS WEBSITE. https://nrc-publications.canada.ca/eng/copyright

Vous avez des questions? Nous pouvons vous aider. Pour communiquer directement avec un auteur, consultez la première page de la revue dans laquelle son article a été publié afin de trouver ses coordonnées. Si vous n’arrivez pas à les repérer, communiquez avec nous à PublicationsArchive-ArchivesPublications@nrc-cnrc.gc.ca.

Questions? Contact the NRC Publications Archive team at

PublicationsArchive-ArchivesPublications@nrc-cnrc.gc.ca. If you wish to email the authors directly, please see the first page of the publication for their contact information.

NRC Publications Archive

Archives des publications du CNRC

This publication could be one of several versions: author’s original, accepted manuscript or the publisher’s version. / La version de cette publication peut être l’une des suivantes : la version prépublication de l’auteur, la version acceptée du manuscrit ou la version de l’éditeur.

Access and use of this website and the material on it are subject to the Terms and Conditions set forth at

Service-life prediction of concrete bridge decks using case-based

reasoning

Morcous, G.; Lounis, Z.; Mirza, M. S.

https://publications-cnrc.canada.ca/fra/droits

L’accès à ce site Web et l’utilisation de son contenu sont assujettis aux conditions présentées dans le site LISEZ CES CONDITIONS ATTENTIVEMENT AVANT D’UTILISER CE SITE WEB.

NRC Publications Record / Notice d'Archives des publications de CNRC:

https://nrc-publications.canada.ca/eng/view/object/?id=4584ff5d-f30a-4bb4-ba97-8919e29f963e https://publications-cnrc.canada.ca/fra/voir/objet/?id=4584ff5d-f30a-4bb4-ba97-8919e29f963e

Service-life prediction of concrete bridge decks using

case-based reasoning

Morcous, G.; Lounis, Z.; Mirza, M.S.

A version of this document is published in / Une version de ce document se trouve dans :

Proceedings of 6th International Conference on Short and Medium Span Bridges, Vancouver, July 31-Aug. 2, 2002, v. II, pp. 769-776

1/8

SERVICE-LIFE PREDICTION OF CONCRETE BRIDGE DECKS

USING CASE-BASED REASONING

G. Morcous

McGill University, Montreal, Canada Z. Lounis

National Research Council of Canada, Ottawa, Canada M. S. Mirza

McGill University, Montreal, Canada

Abstract

Bridge management systems (BMSs) are developed to assist decision-makers in optimizing the allocation of their limited budget on maintenance needs of bridge networks. Reliable deterioration models are essential constituents of BMSs that are used to predict the remaining service life of bridge components. The deterioration models incorporated in the recent BMSs have limitations that can be accepted for the analysis at the network level but not at the component level. Moreover, the current mechanistic deterioration models developed for the component level analysis are neither versatile nor adequately extensible to predict the service life of a large number of bridge components, or to incorporate additional deterioration parameters. Therefore, an artificial intelligence approach “Case-Based Reasoning (CBR)” is proposed to develop extensible, reliable, and generic deterioration models for the analysis at the component level. The CBR approach is utilized to predict the time to corrosion initiation of the reinforcing steel in concrete bridge decks. Data obtained from the Dickson Bridge in Montreal are used to generate the cases that populate the case library of the CBR model by applying the Monte Carlo Simulation techniques. Parameters that significantly affect the deterioration rate of concrete decks, such as the concrete cover thickness, apparent diffusion coefficient, and surface chloride concentration, are considered.

1. Introduction

Highway bridges constitute critical and vital links in any roadway network, because even a partial failure of these bridges may lead to serious catastrophes. In North America, highway bridges are characterized by their growing deterioration rate as a result of

aging, increased traffic volumes and loads, exposure to aggressive environments, and inadequate inspection and maintenance. Maintenance, rehabilitation, and replacement (MR&R) of highway bridges require a tremendous amount of funds that too often exceed the available budgets of the transportation agencies. For example, MR&R costs for bridges in Canada were estimated to be $10 billion1.

Bridge Management Systems (BMSs) have been developed since the mid-1980’s to assist decision-makers in optimizing the allocation of their limited budgets on MR&R needs of bridge networks. The success of a BMS to provide the most cost-effective maintenance strategy is highly dependent on the accuracy and efficiency of the technique used for modeling bridge deterioration2. That is why the American Association of State Highway and Transportation Officials (AASHTO) has prescribed bridge deterioration models as a minimum requirement for any BMS3. The main function of bridge deterioration models is to predict the remaining service life of different bridge components. By definition, the service life of a bridge component is the period of time during which the component can fulfil its performance requirements4. Service life predictions enable decision makers to evaluate different maintenance alternatives based on their long-term costs and to estimate future funding requirements. Current BMSs employ different techniques for modeling bridge deterioration such as regression models and Markovian models. A detailed evaluation of these models was carried out by Morcous et al.5 This evaluation has revealed that these models ignore some of the essential factors that affect bridge deterioration, assume independent deterioration mechanisms of interacting components, and neglect the effects of the condition history on the future condition. These limitations may be acceptable for the analysis at the network level, where bridges are only prioritized for eligibility for maintenance funds. However, these deterioration models cannot be used effectively for the analysis at the component level (i.e. project level), where detailed information regarding certain deterioration mechanisms in specific bridge components is required. The deterioration of concrete bridge decks is one of the most serious problems of the transportation infrastructure in North America. For the last few decades, the service life of concrete bridge decks was found to be much shorter than that for the other bridge components (i.e. bridge foundation, substructure, and superstructure). Many bridge decks require replacement every 15 to 20 years while other components last for 40 years or more6. Although several deterioration mechanisms can occur in concrete bridge decks, such as sulphate attack, alkali-aggregate reaction, freezing-thawing cycles, and creep and shrinkage, corrosion of steel reinforcement is considered to be the most important contributor to their progressive deterioration7. A recent study carried by the Strategic Highway Research Program (SHRP) has shown that the cost of corrosion damage in the United States was estimated at over $20 billion, and it is increasing at the rate of $500 million per year8. Therefore, the focus of this study is on predicting the progress of corrosion of reinforcing steel in concrete bridge decks. However, the same

3/8

procedures can be applied to other deterioration mechanisms and for different bridge components.

Many researchers have studied the process of reinforcement corrosion in concrete structures and several mechanistic models have been developed. The examples include models of corrosion initiation due to chloride penetration9,10 and due to carbonation11,12, and models of concrete cracking due to the expansion of the corroded reinforcement13,14. These models are characterized by their ability to predict the time to corrosion initiation and cracking, provided that certain parameters are known and specific assumptions are satisfied. However, these models may lead to erroneous predictions when one of the governing parameters is not available, or one of the assumptions is not satisfied. These models cannot be easily updated to incorporate any additional parameters, or to benefit from any new data that are frequently accumulated. Furthermore, these models cannot be used for the analysis of a large network of components, which restricts their practicality to many transportation agencies

On the other hand, many transportation agencies perform periodic condition surveys on their bridges and maintain detailed condition reports on bridge decks experiencing deterioration. These reports contain the results of several field and laboratory tests that describe the corrosion of reinforcing steel, such as the results of the half-cell potential, linear polarization, and electrical resistivity tests. This large amount of data, which is updated on a regular basis, contains valuable information on the deterioration of actual bridge decks. This information can be utilized easily in predicting the future condition of other decks. Therefore, this paper proposes the use of an artificial intelligence (AI) technique, called case-based reasoning (CBR), to make the best use of this information. The CBR technique is expected to eliminate the shortcomings of the current models and provide BMSs with extensible, reliable, and generic deterioration models.

This paper briefly introduces the CBR approach and presents its main characteristics, followed by the steel corrosion mechanism in concrete, its different stages, and the parameters that govern its progress. The development of a CBR application for modeling steel corrosion in concrete bridge decks is presented along with the test results.

2. Case-Based Reasoning

Case-based reasoning (CBR) helps solving a problem by reusing the solutions of previous cases that are similar to the current problem. These cases, which are stored in the “case library” include the problem definitions and the related solutions15. CBR differs in concept from the other AI techniques such as expert systems and artificial neural networks (ANN) because these techniques use the general knowledge of the problem domain, while CBR is able to benefit from the specific knowledge of the previously solved problems, which is usually more accurate16. Both CBR and ANN are categorized as machine learning approaches because of their ability to simulate the

learning capabilities of the human brain through the transformation of the raw data of any domain into usable knowledge that can be easily applied. However, an earlier comparison between the two approaches concluded that CBR models are easier to update and more efficient in manipulating both numeric and symbolic data17, which is one of the main reasons for the use of the CBR approach in solving the current problem. In general, CBR has four main aspects15: 1) case representation, which organizes the information describing the problem and its solution, 2) case retrieval, which searches the case library for the case(s) that best match the current problem (known by the query case), 3) case adaptation, which revises the retrieved case(s), to fit the current problem context (if needed), and 4) case accumulation, which is responsible for the storage of new cases and updating of existing cases, thus providing the system with its learning capability.

CBR has been successfully used in solving a wide range of problems, such as design, prediction, classification, and estimation. CBR has been first proposed for modeling bridge deterioration by Morcous et al.5. In this “proof-of-concept” investigation, the future conditions of bridge decks were predicted by using the recorded conditions of bridge decks that are stored in the database of the Ministry of Transportation of Quebec. This application showed the potential of the CBR approach in the service life prediction at the network level. It also showed the higher level of accuracy of the CBR approach than the traditional prediction models that are based on regression analysis.

3. Corrosion of Reinforcing Steel

Reinforcing steel embedded in concrete is protected against corrosion by the high alkalinity of the cementitious environment (i.e. the pH of the pore water in concrete can be greater than 12.5). This alkalinity oxidizes the steel and forms a stable passive film on the steel surface18. However, carbonation of the concrete or the penetration of chloride ions causes breakdown of the passive film and activates the electrochemical reactions that generate corrosion products, or the rust. These products absorb water, increase considerably in volume, and apply pressure on the surrounding concrete, which causes concrete cracking, spalling or delamination, and can ultimately result in failure of the element. Tuutti19 proposed a model that describes the corrosion of the reinforcing steel embedded in concrete as a two-stage process: 1) initiation stage, which is the depassivation of the steel surface due to the penetration of Cl- or CO2, and 2)

propagation stage, which is the electrochemical reaction that occurs with the existence of oxygen, water, and suitable temperature. It should be noted that the rate of corrosion propagation is non-uniform and it may accelerate or decelerate the corrosion process because of changes in the environmental and/or operational conditions20.

In North America, corrosion of reinforcing steel in concrete bridge decks is normally due to the penetration of chloride ions derived from the two commonly used road de-icing salts (CaCl2 and NaCl). In this case, the initiation period is defined as the time

5/8

from the initial exposure until when the concentration of chloride ions at the level of top reinforcement reaches the so-called “threshold concentration”21. The length of this period (Ti) depends mainly on: 1) the surface chloride concentration, 2) the thickness of concrete cover, 3) the threshold chloride concentration in concrete, and 4) the rate of chloride ingress. Although the ingress of chlorides into concrete is a complex process that combines several transport mechanisms such as diffusion, capillary sorption, and permeation, diffusion is recognized as the key governing mechanism especially for concrete decks that are subjected to periodic applications of chloride. The Fick’s second law of diffusion can be applied to obtain the distribution of chloride concentration C(x,t) at depth (x) and time (t) as follows21:

          − = Dt x erf C t x C _S 2 1 ) , ( (1)

where, Cs = chloride concentration at the surface

D = apparent diffusion coefficient

erf = error function

By substituting the depth (x) with the depth of concrete cover (dc) and the chloride concentration (C) with threshold concentration (Cth) in Eqn. 1, the time to corrosion initiation (Ti) can be calculated as follows:

2 1 2 1 4             ₋ = − S th c i C C erf D d T (2)

The initiation time is a key factor in the service life prediction of a concrete element, because the risk of steel corrosion is highly dependent on the quality of design and construction of the concrete cover, which represents the physical barrier against any external aggressive agents. Increasing the density and impermeability of the concrete cover by reducing the water-cementitious materials ratio and producing properly placed, compacted, and cured concrete, reduces the apparent chloride diffusion and consequently delays the onset of corrosion. Therefore, the focus of this paper is on predicting the time to corrosion initiation.

4. Model Development

To develop the case library for the proposed CBR model, cases representing a large population of concrete decks were generated using the Monte Carlo simulation techniques, because of inadequate data on the condition of bridge decks. The mean, the

standard deviation, and the probability distribution of the parameters in Eqn. 2 were obtained from the results of the detailed condition assessment carried out on the Dickson Bridge in Montreal, Canada, prior to its demolition in 19998,22. This bridge was constructed in 1959 with a total length of 366 m and width of 27 m. The bridge superstructure had deteriorated severely because of the poor quality control in construction. Since most of the parameters have lognormal distribution, random values were generated using the Minitab statistical analysis software by modeling these parameters as lognormal random variables. Table 1 shows the location and scale parameters of the lognormal distribution as calculated from the mean and standard deviation of the corresponding normal distribution.

Two groups of 100 cases each, were selected at random from the 1000 cases generated using the Monte Carlo simulation: the first group is termed the training group and is used to refine the parameter weights, while the second group is termed the testing group and is used to validate the refinement. Parameter weights represent the contribution of each parameter to the process of corrosion initiation. Initially, all parameter weights were assigned the same importance and therefore the same numeric values. These weights were subsequently refined in several iterations until the results obtained from the training group were considered satisfactory (Table 1). These weights indicate that the thickness of the concrete cover thickness is the most important parameter to service life prediction while the threshold chloride concentration is the least important one.

To validate the accuracy of the CBR model, the testing group was used to predict the time to corrosion initiation of 100 bridge decks. The time to corrosion initiation of the retrieved cases that have highest similarity with the cases of the testing group were obtained. Figure 1 shows the cumulative distribution of the difference between the predictions of the CBR model and those of the Monte Carlo simulation. This figure indicates that the predictions of the CBR model are within ± one year from the simulated values in about 85% of the cases. The potential of the CBR approach to accurately model the deterioration mechanisms, such as corrosion of steel reinforcement, was proved in addition to its ability to incorporate additional parameters and to be updated as new actual field data becomes available.

7/8

5. Conclusions

The case-based reasoning approach for predicting the time to corrosion initiation of reinforcing steel embedded in concrete bridge decks is presented. Current deterioration models are either developed to support the network level analysis or restricted in terms of their applicability and extensibility. The CBR approach is able to benefit from the data collected in bridge condition surveys to provide decision makers with deterioration models that are easy to update and to incorporate additional deterioration parameters. The case library of the developed CBR model was populated using bridge deck cases generated by the Monte Carlo simulation of the data obtained from Dickson Bridge in Montreal. Testing this model showed about 85% of the CBR predictions match the simulated values, which proves its accuracy besides its versatility and extensibility.

6. Acknowledgements

The authors wish to acknowledge the Natural Sciences and Engineering Research Council of Canada (NSERC) for financially supporting this research.

7. References

1. Lounis, Z., “Reliability-based life prediction of aging concrete bridge decks”, Proceedings of the International RILEM Workshop of Life Prediction and Aging Management of Concrete Structures, Cannes, France, 2000.

2. Madanat, S., Karlaftis, M. G., and McCarthy, P. S., “Probabilistic Infrastructure Deterioration Models with Panel Data”, J. of Infra. Sys., ASCE, 3 (1) (1997), 4-9.

0% 20% 40% 60% 80% 100% -2 -1 0 1 2 3 4 5 6 7 8 9 10 Difference in years Cumulative Percentage

3. AASHTO, “Guidelines For Bridge Management Systems”, American Association of State Highway and Transportation Officials, Publications, Washington, D.C, 1993. 4. Lacasse, M. A. and Vanier, D. J., "A Review of Service Life Durability Issues",

Proceedings of the 7th International Conference of the Durability of Building Materials and Components, Stockholm, Sweden, Vol. 2, May 1996, 867-866. 5. Morcous, G., Rivard, H., and Hanna, A. M., “Predicting the Condition of Bridge

Decks Using Case-Based Reasoning”, Proceedings of the CSCE 29th Annual Conference, D. Noakes (editor), Victoria, BC, Canada, June 2001.

6. Non-Destructive Evaluation Validation Center, “The problem of deteriorating bridge decks”, www.tfhrc.gov/hnr20/nde/problem.htm, 2001.

7. Enright, M. P., and Frangopol, D. M., “Service-life prediction of deterioration concrete bridges”, J. of Structural Engineering, ASCE, 124 (3) (1998), 309-317 8. Amleh, L., “Bond deterioration of reinforcing steel in concrete due to corrosion”,

Ph.D. thesis, McGill University, Montreal, Canada, 2000

9. Page, C. L., Short, N. R., and El Tarros, A., “Diffusion of chloride ions in hardened cement pastes”, Cement and Concrete Research, 11 (1981), 395-406.

10. Cady, P. D. and Weyers R. E., ”Deterioration Rates of Concrete Bridge Decks”, Journal of Transportation Engineering, ASCE, 110 (1) (1983), 34-44.

11. Bentur, A., and Jaegermann, C., “Effect of curing and composition on the properties of the outer skin of concrete”, J. of Mat. in Civil Eng., ASCE, 3 (4) (1991), 252-262. 12. Parrott, P. J., “Design for avoiding damage due to carbonation induced corrosion”,

Durability of Concrete, ACI SP-145, V. M. Malhotra, (ed.), Detroit, 1994, 283-298. 13. Molina, F. J., Alonso, C., and Andrade, C., “Cover cracking as a function of rebar

corrosion: Part II – Numerical method”, Mat. and Str., Paris, 26 (1993), 532-548. 14. Andrade, C., Alonso, C., and Molina, F. J., “Cover cracking as a function of rebar

corrosion: Part I – Experimental test”, Mat. and Str., Paris, 26 (1993), 453-464. 15. Kolodner, J., “Case-Based Reasoning”, Morgan Kaufmann Publishers, Inc. (1993). 16. Aamodt, A., and Plaza, E., “Case-Based Reasoning: Foundational Issues,

Methodological Variations, and System Approaches”, AICOM, 7 (1) (1994), 39-59. 17. Arditi, D. and Tokdemir, O., “Comparison of Case-Based Reasoning and Artificial

Neural Networks”, J. of Comp. in Civil Eng., ASCE, 13 (3) (1999), 162-169.

18. Rosenberg, A., Hanson, C. M., and Andrade, C., “Mechanisms of corrosion of steel in concrete”, Material Science of Concrete I, J. Skalny, ed., American Ceramic Society, Westerville, Ohio, 1989, 285-313.

19. Tuutti, K., “Corrosion of steel in concrete”, Swedish cement and concrete research institute, Stockholm, 1982

20. Bentur, A., Diamond, S., and Berke, N. S., “Steel corrosion in concrete: fundamentals and civil engineering practice”, E & FN Spon, NY., 1997

21. Lounis, Z., and Mirza, M. S., “Reliability-based service life prediction of deteriorating concrete structures”, Proceedings Third International Conference on Concrete Under Severe Conditions, N. Banthia, K. Sakai, and O. E. Gjorv (eds.), University of British Columbia, Vancouver, Canada, 2001

22. Fazio, R., “The assessment and Prediction of Reinforcing Steel Corrosion on the Dickson Bridge”, M.Eng. Thesis, McGill University, Montreal, Canada, 1999.