Chapter 1: Introduction

1.1 General Introduction

As time progresses, industries continually evolve and innovate to enhance efficiency and
comfort. With technological advancements and the rise of artificial intelligence, various sectors
have achieved improved productivity. Among these, the construction industry stands out due to
rising population demands, necessitating more durable and serviceable buildings and infrastructure
One of the key components in concrete structures is steel reinforcement bars (rebars).

However, increased use of rebars also brings a major challenge—corrosion. Corrosion

adversely affects the mechanical properties of rebars, thereby compromising the structural integrity
of reinforced concrete. Concrete structures are typically designed for service lives ranging from 50
to over 100 years. However, exposure to environmental elements leads to degradation over time
[Canadian Concrete Expo, 01-08-2019]. The question of which early reinforced concrete (RCC)
buildings were constructed in India often brings up the city of Chandigarh and the involvement of
Le Corbusier. However, Chandigarh's development began in the mid-1950s. Now, approximately
75 years later, many of the earlier known RCC structures are nearing the end of their lifespan, and
signs of distress are increasingly evident in these buildings [The Early use of Reinforced Concrete
in India, Stuart Tappin]. As a result, the load-bearing capacity of structures is compromised,
increasing the risk of bending, cracking, or even catastrophic failure. [Li Fan & Xiaohui (2021,
July 14)]. The rate of corrosion in steel reinforcement is significantly influenced by the thickness
of the concrete cover. A thicker cover enhances durability by maintaining alkalinity and reducing
chemical ingress. A thicker cover enhances durability by maintaining an alkaline environment,
which protects the steel from corrosion. It also reduces the likelihood of crack formation, improving
structural integrity, and provides better resistance against moisture penetration. Additionally, the
increased cover acts as a physical barrier, slowing down the ingress of aggressive chemicals such
as chlorides and sulfates, thereby prolonging the lifespan of the reinforcement and the structure
[Greene, John P., et al. ,2005].
 Figure 1.1: Corrosion Process [[Link] 12:27, 26-04-2024]

Corrosion is the gradual deterioration or destruction of materials, usually metals, through

chemical or electrochemical reactions with their environment. This process often involves the
interaction of the material with oxygen, moisture, acids, or other corrosive agents, leading to the
loss of material properties such as strength, appearance, and functionality [Fontana, M.G. 2005].

Corrosion significantly impacts the structural integrity of materials, leading to weakened

systems and potential failures. It imposes a substantial economic burden due to increased
maintenance, repairs, and replacements in various industries. The deterioration caused by corrosion
can pose serious safety risks, resulting in accidents and endangering lives. Additionally, it
contributes to environmental damage, especially when failures in critical systems lead to leaks or
contamination.

According to the National Association of Corrosion Engineers, “the documented set of

processes and procedures are required for planning, executing, and continually improving the
ability of an organization to manage the threat of corrosion for existing and future assets and asset
systems.”

Corrosion results in an annual loss of 4.2% of India’s GDP, amounting to approximately $70.3
billion as of 2013 and $110 billion as of 2023, according to the Economic Times. In NACE
International’s report, it’s estimated that if more of those solutions are used, the total savings across
the globe could be anywhere between $357 and $875 billion (USD) annually. That’s between 15
to 35% of the total cost of corrosion, which is no small feat.

2
 The modulus of elasticity (E) of steel rebars is a critical property in civil engineering,
influencing the structural behavior of reinforced concrete elements. Predicting the modulus of
elasticity accurately is essential for ensuring the reliability and safety of engineering designs. Since
the modulus of elasticity depends on a range of material properties and the time constraints,
integration of optimization and design is inevitable in this day and age. To tackle the problem of
large variable data for the calculation, certain computational methods are used. Traditionally, this
property is determined through experimental testing; however, as computational tools advance,
there is increasing interest in developing predictive models based on existing data [Khakurel,
Hrishabh., et al 2021].

Artificial Neural Networks (ANNs) and Bayesian Belief Networks (BBNs) are two powerful
computational methods from artificial intelligence and machine learning used to predict Modulus
of Elasticity based on different properties. ANNs, inspired by biological neural networks, excel in
learning complex patterns from data, while BBNs model probabilistic relationships between
variables. Combining these techniques offers a promising approach to predict the modulus of
elasticity based on various influencing factors like chemical composition, temperature, and
mechanical properties of steel [ Fernando, José 2016].

1.2 Relevance

In today's construction industry, the demand for faster, more accurate, and efficient design tools
is increasing as structures grow more complex and safety standards become more stringent.
Traditionally, the mechanical properties of steel, including the modulus of elasticity (E), are
determined through physical testing methods, which are both time-consuming and costly.
Additionally, the variability in steel's material properties, influenced by different manufacturing
processes, presents challenges for engineers who often rely on generalized values for E. The
strength of Fe550 steel, characterized by a minimum yield strength of 550 MPa, makes it ideal for
use in high-stress applications such as heavy-duty construction and infrastructure projects.

Advancements in artificial intelligence, particularly machine learning techniques such as

Artificial Neural Networks (ANN) and Bayesian Belief Networks (BBN), are revolutionizing this
process. These predictive models enable engineers to estimate material properties with accuracy
using data-driven approaches. By reducing reliance on extensive physical testing, they offer a cost-
effective, time-saving alternative, improving design precision and efficiency. This integration of
3
AI into engineering workflows not only enhances decision-making but also aligns with the
industry's need for innovative, reliable, and scalable solutions.

1.3 Problem statement

The corrosion of steel reinforcement bars (rebars) in concrete structures significantly

undermines durability and structural integrity, particularly in adverse environmental conditions. A
limited body of research exists concerning the long-term effects of corrosion on non-embedded
rebars and its influence on the flexural behaviors of reinforced concrete beams. Furthermore, there
is a pressing need to develop technology capable of predicting corrosion and its subsequent effects
on the flexural behaviors of reinforced cement concrete (RCC) members.

1.4 Methodology

First Part:

Methodology for Corrosion Induction and Testing of Steel Rebars:

1. Preparation of Steel Rebar Specimen:

 Steel rebars of standard dimensions were cleaned to remove any surface
impurities (e.g., rust, grease, or dirt) to ensure consistent corrosion initiation.
 Each rebar was measured and marked for identification before the corrosion
process began.
2. Induction of Corrosion:

 The prepared steel rebars were subjected to 2% NaCl solution and wrapped in
moist gunny bags to simulate a high-humidity environment conducive to
corrosion.

 The gunny bags were periodically wetted with the above saline solution to
accelerate the corrosion process and mimic marine or aggressive conditions.

 Rebars were kept in this environment for a predetermined period, allowing

controlled corrosion to develop. Corrosion levels were monitored by visual
inspection and weight loss measurements at regular intervals.

4
3. Mechanical Testing:

 After specific exposure intervals (e.g., 1 month, 3 months, 6 months), rebars

were removed from the gunny bags, and surface rust was cleaned off using wire
brushes.

 The rebars were tested for their mechanical properties using a Universal Testing
Machine (UTM).

 Tensile Strength Test: Each rebar was subjected to tensile loading to

determine its yield strength, ultimate tensile strength, and elongation.

 Data from the tensile test was recorded, including stress-strain behavior, to
evaluate the impact of corrosion on mechanical properties.

4. Data Analysis:

 The results were analyzed to determine the correlation between the duration of
corrosion exposure and the degradation in mechanical properties.

 Comparisons were made between corroded and uncorroded rebars to quantify

the effect of corrosion.

5
Second Part: Methodology for Data Modelling using ANN and BBN:

1. Data Collection & Preprocessing: Gather steel rebar property data and corresponding
modulus of elasticity values, then process and prepare it for both models.

2. BBN Model Building: Define influencing variables, establish relationships, and construct a
network to learn probability distributions.

3. ANN Model Building: Choose an architecture, define inputs based on steel rebar properties,
and train the network to learn relationships.

4. Integration & Prediction: Combine BBN and ANN (hybrid or ensemble approach) to
generate a final modulus of elasticity prediction.

5. Model Evaluation: Use separate data to assess prediction accuracy with metrics like mean
squared error or R-squared.

6. Refinement: Based on evaluation, refine the BBN structure, ANN architecture, or integration
technique for better predictions.

1.5 Importance of the model developed.

Developing a predictive model using ANN and BBN for the modulus of elasticity offers several
advantages:

● Cost and Time Efficiency: Predictive models significantly reduce the reliance on
physical testing, which is often time-consuming and expensive. By using
computational tools, engineers can input relevant data and quickly obtain reliable
estimates for properties like the modulus of elasticity. This eliminates the need for
preparing multiple samples, conducting labor-intensive tests, and analyzing results
manually. The time saved allows for faster project timelines, and the resources
conserved can be redirected to other critical areas of the project. In large-scale
operations, where testing every batch of material is impractical, predictive models
offer a scalable and cost-effective solution.

6
● Higher Accuracy: Data-driven predictive models can be fine-tuned to
accommodate specific variations in steel rebars, such as differences in chemical
composition, mechanical properties, or manufacturing processes. Unlike generic
formulas or approximations, these models provide tailored predictions by
leveraging extensive datasets and sophisticated algorithms. This precision ensures
that the modulus of elasticity is calculated with a high degree of reliability, reducing
the chances of design errors. Accurate predictions are particularly critical in safety-
sensitive projects, such as high-rise buildings or bridges, where underestimating
material properties can lead to catastrophic failures.

● Incorporation of Uncertainty: Bayesian Belief Networks (BBNs) are uniquely

suited to handle uncertainty in predictive modeling. Real-world data often include
variations and noise due to factors like measurement errors, environmental
influences, or inherent material inconsistencies. BBNs incorporate probabilistic
relationships to account for these uncertainties, producing predictions with
associated confidence levels. This robust approach ensures that engineers are not
working with rigid or overly deterministic values but rather with adaptable models
that reflect the variability of real-world conditions. Such capability is essential for
risk assessment and decision-making in complex engineering projects.

● Enhanced Understanding of Influencing Factors: ANN and BBN models not

only provide predictions but also offer insights into the relationships between input
variables and the modulus of elasticity. By analyzing the impact of factors like
temperature, carbon content, or the steel’s manufacturing process, these models help
engineers understand how each variable contributes to the material's behavior. This
deeper understanding enables optimization of steel properties to meet specific
engineering requirements. For instance, if carbon content is identified as a major
factor affecting elasticity, adjustments in composition during manufacturing can be
made to achieve desired performance standards.

● Automation and Scalability: Once trained, predictive models can be deployed

repeatedly for a wide range of steel grades without requiring additional manual
intervention. This makes the process not only faster but also more consistent

7
 compared to traditional testing methods. The scalability of these models means they
can handle large datasets or accommodate new steel grades with minimal
reconfiguration. Automated predictions allow industries to integrate these models
into production pipelines, enabling real-time quality checks and faster decision-
making. This capability is particularly advantageous for large construction projects
where diverse materials are used, and rapid assessments are essential.

1.6 Summary of Chapter

The integration of machine learning techniques such as ANN and BBN in predicting material
properties like the modulus of elasticity of steel rebars represents a significant step forward for the
civil engineering industry. The development of these models is crucial for improving efficiency,
accuracy, and understanding of material behavior in reinforced concrete structures. With growing
computational power and data availability, predictive models will continue to evolve, contributing
to more robust, safe, and cost-effective engineering designs.

As the construction industry moves toward more data-driven practices, predictive models for
material properties will become indispensable tools for engineers and designers. The benefits of
reduced testing costs, improved prediction accuracy, and the ability to handle uncertainties make
these models valuable in both research and practical applications.

8
 Chapter 2: Literature Review and Gap

2.1 Introduction

2.1.1 Lin, C.-C., & Wang, C.-X. (2005). Correlation between accelerated corrosion tests and
atmospheric corrosion tests on steel. Journal of Applied Electrochemistry

This study tackles the problem of subtropical areas' elevated air corrosion rates for steel, which
call for costly outside exposure testing. To hasten the process, the study establishes connections
between rates and corrosion parameters (temperatures, time of wetness, and fluxes of Cl
deposition). The study examines corrosion rates for three steel specimens in diverse environments,
such as constant temperature and humidity chambers and salt. The corrosion rates are predicted
and observed. The findings show that variables like wind and precipitation affect the fluxes of Cl
deposition. In atmospheric testing, A588 has the best anti-corrosion qualities. To forecast corrosion
rates in coastal locations, a link between corrosion theory and Arrhenius' law is inferred, taking
into account a number of independent variables. However, because atmospheric elements are not
taken into account, model simulations are understated. Weight loss data is confirmed by SEM
results, which also demonstrate considerable corrosion pitting for SS400 and SPHC but less pitting
for A588. Overall, the intricacy of corrosion prediction is highlighted by the comparison of
observed and anticipated thickness losses in air corrosion, which shows differing degrees of
inaccuracy for each material.

2.1.2 Klakočar-Ciepacz, M., Falewicz, P., Kuczkowska, S., & Zybura, A. (2011). Modelling of
Corrosion of reinforcement in contact with cracked concrete cover.

The study utilized polarization curves and simulated pore solutions to assess the impact of
corrosion on reinforcing steel embedded in fractured concrete. These electrochemical analyses
provided insights into the kinetics of corrosion processes under varying anode-to-cathode area
ratios (Aa/Ac), which play a crucial role in determining corrosion severity. The results showed that
corrosion potential and current density were significantly influenced by this ratio. When Aa/Ac
was less than 1, indicating a larger cathodic area relative to the anodic zone, mixed control
mechanisms were observed—where both anodic and cathodic reactions influenced the corrosion
process—with cathodic control being predominant. This condition typically results in higher
corrosion rates due to an increased availability of reduction reactions.
9
 Conversely, when Aa/Ac exceeded 5, the corrosion rate dropped significantly, highlighting the
protective influence of a larger anode area relative to the cathode. These findings align with real
concrete structures where Aa/Ac commonly ranges between 1:1 and 1:2, implying moderate
corrosion rates in field conditions. However, the study also acknowledged limitations in using
artificial pore solutions, which—though chemically representative—failed to account for physical
factors like cracking. Cracks in actual concrete alter the resistivity and oxygen transport pathways,
thereby accelerating corrosion processes beyond what is predicted in controlled environments. This
underlines the importance of incorporating both chemical and physical degradation mechanisms
when modeling real-world corrosion behavior in reinforced concrete structures.

2.1.3 François, R., Khan, I., & Dang, V. H. (2012). Impact of corrosion on mechanical properties
of steel embedded in 27-year-old corroded reinforced concrete beams. Materials and Structures.

Corrosion poses a danger to the reinforced concrete sector, causing problems like concrete
cracking and weakening the link between steel and concrete, which reduces the service life of
buildings. An extensive long-term experimental programme was carried out in 1984 by the
Laboratories Matériaux et Durabilité des Constructions (L.M.D.C.) at INSA-Toulouse. Thirty-six
reinforced concrete beams were exposed to a chloride environment while maintaining sustained
loading, and another thirty-six beams were used as control beams under standard laboratory
conditions. In order to speed up the corrosion process, this 27-year study included wetting-drying
cycles and exposure to temperatures between -5°C and 35°C. The investigation included a wide
range of topics, such as the mechanical behavior of the beams under service loads, chloride content,
and cracking maps. There were differences in the concrete cover between two beam groups, type
A and type B, with different reinforcements but identical ordinary steel. Notably, tensile testing on
corroded steel bars revealed heterogeneous patterns of corrosion caused by aggressive chloride
exposure in the form of salt fog. The plateau on the stress-strain curve vanished as the true ultimate
stress rose with corrosion levels but the true yield stress stayed constant. As corrosion increased,
the ratio of ultimate stress to yield stress increased as well, impacting the steel class based on
criteria. Crucially, as corrosion increased, final elongation significantly decreased, which had an
impact on the structural stability of reinforced concrete in settings containing chloride. The study
also examined the effect on the steel's elastic modulus and concluded that, when average mass loss
was considered during computations, the elastic modulus was steady at about 200 GPa in spite of

10
corrosion. One of the most important factors affecting the durability of corroded reinforced
concrete buildings was found to be the steel bars' residual ultimate elongation.

2.1.4 Vu, N. S., Yu, B., & Li, B. (2017). Stress-strain model for confined concrete with corroded
transverse reinforcement. Engineering Structures

To forecast the atmospheric corrosion rates of steel, the study sought to establish a link between
corrosion rates and relevant variables (temperatures, wetness duration, and fluxes of Cl-
deposition). In salt and constant temperature/humidity chambers, accelerated corrosion studies
were carried out with different Cl- deposition fluxes (0, 100, and 2500 mg/m2/day) and
temperatures (288, 298, and 308 K). Tests for atmospheric corrosion were conducted at a coastal
station. Results showed that in both accelerated and ambient testing, SPHC (steel plate, hot rolled
coil) steel showed the greatest rates of corrosion, followed by SS400 and A588. Higher
temperatures and fluxes of Cl-deposition were associated with higher rates of corrosion. When
compared to the findings of air tests, a correlation that was created from accelerated test data using
corrosion theory and Arrhenius' law was shown to have an average inaccuracy of 31.6%. measures
of weight reduction, Surface morphology and corrosion behavior were studied using Tafel plots
and SEM analysis. Nevertheless, the study failed to take into consideration the decrease in oxygen

diffusion and ignored other ambient corrosion variables, such as SO2, which might have

contributed to an underestimate of corrosion rates. Notwithstanding these shortcomings, the

research effectively connected corrosion rates to certain elements, supporting the prediction of steel
corrosion rates in the atmosphere.

2.1.5 Quraishi, M., Nayak, D., Kumar, R., & Kumar, V. (2017). Corrosion of reinforced steel in
concrete and its control: An overview.

This study investigated the effects of corrosion inhibitors on the corrosion of steel rebar and the
characteristics of concrete. Concrete strength is somewhat reduced by the inclusion of inhibitors,
although this impact is mitigated with extended curing durations. Moreover, the use of mineral
admixtures such as fly ash and silica fume improves the workability, sulfate resistance, reduced
heat generation, and enhanced water tightness of concrete. The study assessed the effects of three
inhibitors (hexamine, calcium nitrite, and sodium citrate) on the consistency, setting time, strength,
and soundness of concrete. The efficiency of the corrosion and inhibitor on steel rebar was

11
evaluated by weight loss tests. The study also clarified the mechanisms underlying various forms
of corrosion in concrete and the advantageous function of mineral admixtures in boosting corrosion
resistance and durability. On the other hand, precise information on the degree of strength loss
brought on by inhibitors at various curing times is absent from the study. Furthermore, it does not
go into detail on how inhibitors affect other important concrete characteristics like permeability
and shrinkage. Research has to be done on the durability and long-term performance of concrete
treated with various inhibitors, as well as the best doses to achieve maximum corrosion inhibition.

2.1.6 Recupero, Antonino, and Nino Spinella, "Experimental tests on corroded prestressed
concrete beams subjected to transverse load."(2019)

Research on corroded prestressed concrete beams under transverse load indicated that tendon
corrosion greatly affects the load-bearing capacity and stiffness of the beams. The mid-span
corroded beam's load capacity was around 45% that of the defect-free control beam. Corrosion had
less of an impact on unbonded tendons (T6a and T6b) than bonded tendons. The investigation came
to the conclusion that the structural performance of prestressed concrete beams can be seriously
harmed by corrosion in the prestressing tendons. The amount of stiffness and load-bearing capacity
loss depends on the location and severity of corrosion. The goal of the project is to better understand
the failure processes and damage progression caused by corrosion in prestressed concrete parts by
examining the link between tendon corrosion and the decline in structural performance in beams
made of prestressed concrete.

2.1.7 Alabduljabbar, H., Haido, J. H., Alyousef, R., Yousif, S. T., McConnell, J., Wakil, K., &
Jermsittiparsert, K. (2020). Prediction of the flexural behavior of corroded concrete beams using
a combined method. Structures

The goal of the project is to employ finite element analysis (FEA) and artificial neural networks
to forecast the flexural behavior of corroded concrete beams. The material characteristics, corrosion
current density, beam cross-sectional area, and corrosion time are among the factors taken into
account. Experimental data for training (90%) and testing (10%) makes up the proposed neural
network model. Various software packages, such as MATLAB and ABAQUS version 6.14-1, are
employed in conjunction with techniques like the gradient descent approach. The variance, mean,
median, standard deviation, and correlation between the data are computed. A 20-noded nonlinear
quadratic brick element with decreased integration is used to simulate concrete. The ranges utilized
12
in the ANN development correspond to the beam parameters. When taking into account four beam
toughness, there is a good degree of consistency between current FEA results and experimental
data.

The flexural behavior of concrete beams damaged by corrosion has not been examined in this
work utilizing a hybrid model constructed using computational intelligence and finite element
simulation. The process includes collecting experimental data, dividing it into training and testing
data, and using a neural network's multilayer feed-forward with back propagation algorithm. For
the purpose of training and testing the suggested neural network, a MATLAB code was created.

2.1.8 D. J. Mills, D. Nuttall & L. Atkin (2021): EUROCORR 2020: ‘Closing the gap between
industry and academia in corrosion science and prediction’, Corrosion Engineering, Science
and Technology,

The document offers comprehensive summaries of presentations from the EUROCORR 2020
conference, focusing on a wide range of topics such as cathodic protection, corrosion mechanisms,
and challenges in industries like oil and gas production, refinery operations, petrochemicals, and
electronics. These presentations addressed both theoretical and practical aspects, emphasizing
innovative solutions to mitigate corrosion and enhance material performance under harsh
conditions. The discussions explored corrosion reliability in electronic devices, highlighting the
effects of environmental factors and strategies for improving durability.

A significant portion of the conference emphasized modelling approaches to predict corrosion

patterns and assess service life, combined with experimental investigations to validate theoretical
findings. Real-world applications demonstrated the practical relevance of these studies, bridging
the gap between laboratory research and industrial needs. These diverse contributions showcased
advancements in understanding corrosion and developing protective measures for critical
infrastructure and materials.

However, the document primarily focuses on individual presentations and lacks a synthesis of
broader themes or recurring challenges. It does not analyze the connections between different
studies or explore how they collectively advance the field. For example, shared issues like
standardizing testing methods, leveraging analytics for corrosion monitoring, and addressing gaps
in corrosion-resistant materials could have been discussed.

13
 Including a holistic overview of the conference findings, such as identifying common hurdles
or proposing future research directions, would have added significant value. Highlighting recurring
themes, such as the integration of predictive models and experimental approaches, or the
convergence of findings across sectors, could guide strategic advancements in corrosion science
and its industrial applications. Such an analysis would enhance understanding of the current state
of research and its potential to address global challenges in material degradation.

2.1.9 Vanama, R.K. and Ramakrishnan, B., (2020). Experimental investigation on the
mechanical properties of naturally and artificially corroded steel rebars.

Tensile characteristics of both artificially corroded ribbed steel rebars and naturally corroded
round steel rebars were investigated in this work. It was discovered that corrosion had significantly
reduced the longitudinal, transverse, and shank diameters of the ribs. With increasing corrosion
level, elastic modulus, ductility, yield, and ultimate stresses all reduced. The reduction in elastic
modulus indicated a loss in the material's stiffness, affecting its ability to resist deformation under
stress. Ductility loss was particularly critical, as it limited the steel's ability to undergo plastic
deformation before failure, increasing the risk of brittle fracture. Yield stress degradation pointed
to an earlier onset of plastic behavior under load, compromising structural integrity. The ultimate
stress drop revealed that corroded steel could no longer sustain high load levels, reducing its safety
margin. Corrosion reduced the yield plateau zone, and severely corroded steel rebar hardened
beyond the yield threshold. The shank diameter decreased by 26.45% at a mass loss of 50.48%. In
order to predict the tensile characteristics of corroded steel rebar, the study developed revised
degradation equations. Tensile property degradation models were verified up to 79.64% mass loss
in corrosion levels. The impact of corrosion on several mechanical attributes, including fatigue life
and fracture toughness, as well as the impact of rib loss on the steel rebar-concrete connection were
also investigated in this work.

2.1.10 Tariq, F., & Bhargava, P. (2021). Stress-strain curves and mechanical properties of
corrosion-damaged super ductile reinforcing steel. Structures.

The study investigated the effects of rusting on the behavior of reinforced concrete columns. It
found that rusting does not significantly impact the initial stiffness of the concrete, but as rust
increases, the strength decreases at its peak point. Columns with circular spirals showed improved

14
ductility compared to those without spirals, highlighting the importance of confinement in
enhancing performance.

As the metal inside the concrete corrodes, the material becomes less flexible, reducing its
strength and strain capacity. Using two hoops to confine corroded concrete was shown to improve
both strength and stretchability, allowing the material to endure more deformation before breaking.
This suggests that adequate confinement can mitigate some of the adverse effects of corrosion.

The study identified the corrosion level and the volumetric ratio of transverse reinforcement as
the key factors affecting the stress-strain relationship. Higher levels of corrosion had a more
pronounced effect on strain-stress behavior, causing greater reductions in strength and ductility.

The experimental setup included 36 reinforced concrete columns, each with a thickness of 10
mm, to assess these effects. The findings emphasize the critical role of reinforcement configuration
and confinement in maintaining structural performance under corrosive conditions. This study
provides valuable insights into designing more resilient reinforced concrete structures in
environments prone to corrosion.

2.1.11 Fornasari, G., Capozzoli, L., & Rizzo, E. (2023). Combined GPR and Self-Potential
Techniques for Monitoring Steel Rebar Corrosion in Reinforced Concrete Structures: A
Laboratory Study.

This study focuses on using non-destructive testing (NDT) techniques—Ground Penetrating

Radar (GPR) and Self-Potential (SP)—to monitor steel rebar corrosion in reinforced concrete.
Laboratory tests employed a 5% sodium chloride solution to accelerate corrosion, simulating
chloride ingress typical in harsh environments. GPR techniques captured signal amplitude
variations over time, revealing correlations between corrosion levels and signal changes. As
corrosion advanced, the reflected radar signals from the rebars showed attenuation and time shifts,
indicating progressive deterioration and changes in the dielectric properties of the concrete. SP
measurements tracked electrochemical potential differences as corrosion progressed, providing
insight into anodic and cathodic zones across the reinforced surface. This method effectively
detected early-stage corrosion by identifying negative potential zones, often associated with active
corrosion sites. The combined interpretation of GPR and SP data allowed for a more
comprehensive assessment of corrosion activity, offering both physical and electrochemical

15
perspectives. The findings demonstrate the potential of integrating GPR and SP data for diagnosing
and quantifying corrosion-related damage in reinforced concrete structures, paving the way for
timely maintenance strategies and life-cycle assessment of aging infrastructure.

2.1.12 Covatti, L., Medeiros, A., Silva Filho, L. C. P., Baptista-Pereira, C., & Oliveira, M. P.
(2023). Laboratory Induction of Accelerated Corrosion in Rebars of Reinforced Concrete
Structures.

This research applied induced accelerated corrosion using a constant electric current to study
corrosion's effects on ten cylindrical reinforced concrete specimens. The experiment employed
Brazilian Portland CP-V ARI cement and 3% NaCl solution, mimicking aggressive environments
typically found in coastal or deicing salt-exposed regions. By using impressed current techniques,
corrosion was initiated and controlled, allowing the researchers to observe how reinforced concrete
reacts under prolonged corrosive stress within a condensed timeframe. Results highlighted visual
pathologies, such as longitudinal and transverse cracks, surface delamination, and rust stains, which
are symptomatic of expansive corrosion products exerting pressure on the surrounding concrete.
These defects not only compromise the structure's appearance but also its durability and mechanical
integrity. The study employed thermochemical baths to extract and quantify the actual mass loss
of corroded steel, which was then compared to theoretical predictions based on Faraday’s Law.
The findings revealed significant discrepancies between expected and observed corrosion levels,
pointing to factors like non-uniform current distribution, concrete resistivity, and steel-concrete
bond quality influencing corrosion progression. These insights are valuable for improving
reinforced concrete design, especially in terms of corrosion allowance, protective measures, and
predictive maintenance strategies.

2.1.13 Khayatazad, M., Honhon, M. & De Waele, W. (n.d.). Detection of corrosion on steel
structures using an Artificial Neural Network.

The paper discusses the challenges associated with traditional image processing techniques in
detecting corrosion, emphasizing their dependency on manual parameter tuning, which can be both
time-consuming and inefficient. These methods often lack the adaptability required to perform well
across diverse conditions, limiting their practicality in dynamic environments.

16
 To address these issues, the authors introduce an innovative approach leveraging artificial neural
networks (ANNs) for corrosion detection. Unlike conventional convolutional neural networks
(CNNs), which typically demand large datasets for effective training, the proposed ANNs are
designed to operate with a drastically reduced data requirement—just 0.1% of the training data
needed by CNNs. This significant reduction in data dependency makes the ANN-based approach
more accessible and cost-effective for real-world applications, where obtaining large labeled
datasets can be challenging. Additionally, the simplicity and efficiency of the ANN model enable
faster deployment and integration into various industrial workflows. The paper underscores the
potential of this method to revolutionize corrosion detection by providing a reliable and scalable
solution with minimal resource requirements.

2.1.14 Preethi Kumari and Mulky Lavanya, with Mulky Lavanya. "Optimization Strategies for
Corrosion Management in Industries with Artificial Neural Network and Response Surface
Technology: A Comprehensive Review".

The paper offers an in-depth examination of corrosion management strategies, stressing their
critical importance in safeguarding infrastructure in industries such as pipelines, refineries, and
power generation plants. It identifies industrialization as a major driver of increased corrosion
challenges, underscoring the urgency of implementing sustainable approaches to protect valuable
assets and ensure operational efficiency. A central theme of the review is the integration of
advanced technologies like Artificial Neural Networks (ANNs) and Response Surface
Methodology (RSM) into corrosion management. These techniques are highlighted for their ability
to model and predict corrosion behavior with remarkable precision by analyzing the intricate
relationships between multiple influencing factors, such as material properties, environmental
conditions, and chemical exposure.

The review also explores how ANNs leverage machine learning to process large datasets,
enabling accurate forecasting of corrosion trends and enhancing decision-making in maintenance
planning. Similarly, RSM is presented as a robust statistical tool that optimizes corrosion control
strategies by identifying the most influential parameters and their interactions. By combining these
innovative approaches with traditional corrosion management practices, the paper illustrates a
pathway toward more effective, predictive, and resource-efficient corrosion control solutions. The

17
insights provided aim to inspire further research and adoption of these methods to address the
growing demands of industrial and environmental sustainability.

2.2 Literature Gap

Limited studies exist on the corrosion of non-embedded steel rebars exposed for over three years.
This research focuses on determining the flexural behavior of corroded concrete beams using a
hybrid model that combines a Bayesian Belief Network with an Artificial Neural Network.

2.3 Research Objectives

1. To evaluate the extent of corrosion and corresponding mechanical properties of steel

rebars.
2. To establish the relationship between corrosion percentage and mechanical
properties of steel rebar.
3. To develop a BBN model to find the initial probabilities of the factors affecting
corrosion of steel rebars.
4. To develop an ANN model representing the correlation between the key mechanical
properties of the steel rebar and corrosion percentage and various other factors
affecting the corrosion.

2.4 Summary of Chapter

In summary, these objectives focus on assessing the extent of corrosion in steel rebars and its
impact on mechanical properties, establishing quantifiable relationships between corrosion and
mechanical behavior, and employing advanced modeling techniques. Specifically, a Bayesian
Belief Network (BBN) model will be developed to identify and evaluate the initial probabilities of
factors influencing corrosion. Additionally, an Artificial Neural Network (ANN) model will be
formulated to represent the intricate correlation between key mechanical properties, corrosion
percentages, and the influencing factors. These approaches aim to provide a comprehensive
understanding and predictive capabilities for the behavior of steel rebars under corrosion
conditions.

18
 Chapter 3: Material and Methods

3.1 Introduction

In this experiment, steel rebars of varying diameters were subjected to controlled corrosion to
examine how their mechanical properties changed over time. Corrosion was induced by applying
a 2% NaCl solution to the bars and wrapping them in gunny bags to simulate a corrosive
environment. The bars were periodically exposed to this solution for a set duration, and after
corrosion was induced, the mechanical properties of the bars, including yield strength and tensile
strength, were tested using a Universal Testing Machine (UTM). The data collected allowed for
the establishment of a relationship between the percentage of corrosion and the degradation in the
mechanical properties of the steel rebars, highlighting the impact of corrosion on structural
materials.

The second phase of the project focused on developing a predictive model for the modulus of
elasticity of steel rebars. Data collected from various research papers was used to estimate the
conditional probabilities of factors influencing the modulus of elasticity, such as corrosion level,
temperature, and material composition. This data was employed to train both Artificial Neural
Network (ANN) and Bayesian Belief Network (BBN) models, which were used to predict the
modulus of elasticity based on different input parameters. The trained models provided valuable
insights into how corrosion and other variables influence the mechanical properties of steel rebars,
offering a tool for more accurate predictions in material design and structural engineering.

3.2 Materials

3.2.1 Steel Rebars

● Reinforcing bars (rebars) are essential elements in reinforced concrete, providing

tensile strength to concrete structures.

● In this experiment, rebars of different diameters and grades were used to understand
how corrosion affects mechanical properties across a range of real-world
applications.

19
 ● Grades Used: Focus was on Fe550 grade.

 Fe550 refers to a high-strength rebar with a minimum yield strength of

550 MPa.

 It is commonly used in high-rise buildings, bridges, and heavy

infrastructure.

● Diameters: Multiple diameters were used (e.g., 8 mm, 12 mm, 16 mm, etc.) to
analyze the impact of bar size on:

 Surface area available for corrosion.

 Strength degradation patterns.

❖ Purpose:

● These rebars were subjected to corrosion under controlled conditions.

● After corrosion, they were tested for:

 Yield strength

 Ultimate tensile strength (UTS)

 Modulus of elasticity

 Elongation

● This helped understand the quantitative effect of corrosion on structural steel used
in reinforced concrete.

3.2.2 Corrosion Inducing Agent

To simulate real-world environmental exposure and accelerate corrosion in a lab setting.

20
❖ Materials Used:

● 2% Sodium Chloride (NaCl) Solution:

 Mimics saltwater or deicing salts commonly responsible for rebar

corrosion in marine or cold regions.

 A 2% concentration ensures consistent and measurable corrosion over

time.

 Regular application ensures continuous exposure, similar to coastal or

splash zones.

● Gunny Bags:

 Made from jute fibers, these bags are excellent at retaining moisture.

 Rebars were wrapped in these after applying NaCl solution.

 The moist environment inside the gunny bags:

 Keeps the bars damp for a long time.

 Promotes rusting (oxidation of iron) similar to concrete cracks exposed

to water.

❖ Purpose: To replicate field corrosion conditions in a controlled laboratory environment over

a shorter period.

3.2.3 Testing Equipment

❖ Purpose: To evaluate the mechanical properties of steel rebars before and after corrosion.

❖ Equipment Used:

1. Universal Testing Machine (UTM):

 Used for tensile testing.

21
  Applies a gradually increasing load to a steel bar until failure.

 Records:

i. Yield strength: Point where steel begins to deform permanently.

ii. Tensile strength: Maximum stress the material can withstand.

iii. Modulus of elasticity (E): Slope of the elastic portion of stress-strain

curve.

iv. Elongation: Measure of ductility.

2. Measuring Devices:

 Digital calipers / micrometers: Measure the bar diameter.

 Precision scales: Weigh bars before and after corrosion to calculate mass
loss (%).

 Formula used:

❖ Codal Provisions:

● IS 1608 (Part 1): 2018

 Governs the method for tensile testing of metallic materials.

 Ensures standardized testing conditions, sample preparation, and loading

rates.

● IS 1786: 2008

 Specifies minimum elongation, yield, and tensile strength requirements for

Fe415, Fe500, Fe550, etc.

 Used to compare test results against code-mandated limits.

22
3.2.4 Data Sources

❖ Description: Data was used for training and testing machine learning models.

❖ Sources:

1. Research Papers & Studies:

 Provided published datasets on mechanical properties of rebars under different

corrosion conditions.

 Helped in understanding global trends and comparing results.

 Examples:

 Studies showing effects of 5%, 10%, 15% corrosion on strength.

 Reports on bar performance under elevated temperatures or cyclic loading.

2. Web Sources:

 Included datasets from laboratory or field experiments conducted by:

 Universities

 Research organizations

 Government labs

 Used for data validation and filling gaps in in-house experiments.

3. Experimental Data:

 150 rebars were corroded and tested under controlled conditions.

 Measured properties:

a) Diameter

b) Mass loss (%)

23
 c) Yield strength

d) Elongation

e) Modulus of elasticity

❖ Purpose:

● Dataset served as a foundation for model training.

● Larger and more diverse datasets = more accurate predictions using AI models.

3.2.5 Software Tools

❖ Tools Used for Data Handling, Analysis, and Modeling

i. Artificial Neural Network (ANN) – Python:

● Libraries: Tensor Flow, Keras, NumPy, Pandas.

● ANN is a machine learning model that mimics brain neurons.

● Input Parameters:

 % corrosion

 Diameter

 Initial strength

 Environmental factors (e.g., temperature)

● Output: Predicted modulus of elasticity

● Use: Can model complex nonlinear relationships.

ii. Bayesian Belief Network (BBN) –

● BBN model probabilistic relationships between variables.

24
 ● Example: If corrosion > 10%, what is the probability modulus of elasticity
reduces by >20%?

● Use: Good for uncertain and interdependent parameters.

iii. Excel (or similar spreadsheet software):

● Used for:

 Raw data entry

 Sorting, filtering

 Creating training/test datasets

 Preliminary charts and summary tables

iv. Statistical Analysis Tools:

● Used for:

 Feature selection (important input variables)

 Data cleaning (handling missing or outlier data)

 Correlation analysis

● Could include: Python (Scikit-learn), R, or even SPSS.

3.3 Methods and Processes

3.3.1 Corrosion Process

● Wet-Dry Cycles:

 Rebars wrapped in gunny bags were sprayed periodically with NaCl to simulate
realistic exposure conditions.

25
  Duration: Not specified but likely ranged from days to weeks depending on
corrosion levels desired.

● Mass Loss Calculation:

Gravimetric Method: Corroded bars are often cleaned using thermochemical baths
(e.g., Clarke’s solution) to remove rust but in our case we also used wire brush and
weighed to calculate percentage loss.

3.3.2 Mechanical Testing

● Pre-Corrosion Testing: Baseline tensile properties recorded from uncorroded

specimens.

● Post-Corrosion Testing: Specimens are tested in UTM to obtain

i. Yield Strength (fy)

ii. Ultimate Strength (fu)

iii. Elongation at fracture

iv. Modulus of Elasticity (E) via stress-strain curve

● Comparison: Corrosion level (%) vs. Degradation (%) of each mechanical property.

3.3.3 Predictive Modeling

● Objective: Predict modulus of elasticity using:

i. Artificial Neural Network (ANN):

 Built in Python using TensorFlow/Keras.

 Inputs: Corrosion %, temperature, steel composition, bar diameter, etc.

 Output: Estimated modulus of elasticity.

26
 ii. Bayesian Belief Network (BBN):

 Implemented in R using packages like bnlearn.

 Shows probabilistic influence of each variable on modulus of elasticity.

iii. Data:
 Collected from experimental testing (150 rebars) and verified through
published datasets.
 Preprocessed in Excel or statistical tools for cleaning and formatting.
 Refinement of Dataset Parameters

Initially, data was gathered from a wide range of sources, including

experimental testing, research publications, and online repositories. However,
this raw dataset was highly unrefined. It contained inconsistent formats,
missing values, mixed units, duplicate records, and parameters that were not
directly related to the objective of predicting the modulus of elasticity of
corroded steel rebars.

The dataset included various unrelated or redundant entries such as concrete

compressive strength, beam and reinforcement configurations, and material
compositions from different regions and standards. In its original form, this
data was difficult to interpret and unsuitable for accurate modeling.

To address this, the entire dataset—including both experimental and

secondary data sources—was systematically refined through the following
steps:

Parameter Selection: Only relevant features such as bar diameter, corrosion

percentage (mass loss), yield strength, tensile strength, elongation,
temperature, and modulus of elasticity were retained. Irrelevant or noisy
features were excluded.

Data Cleaning: Missing values were either corrected, interpolated where

appropriate, or removed. Units were standardized (e.g., MPa for stress, mm
for dimensions), and extreme outliers were carefully handled.

Data Structuring: The cleaned data was reorganized into a consistent tabular
format, with clearly labeled rows and columns, making it suitable for machine
learning input.

27
 Validation: Where possible, values were verified or cross-checked across
sources to ensure reliability.

This refinement process significantly improved the quality, consistency, and

usability of the dataset. The final refined dataset, comprising both validated
experimental results and selected secondary data, was then used as the input
for developing accurate and robust predictive models using Artificial Neural
Networks (ANN) and Bayesian Belief Networks (BBN).

Parameters
Bar Type
Beam Type
Reinforcement Percentage (%)
Load (kN)
Displacement (mm)
Specimen ID
Concrete Strength (fc, MPa)
Av fy, Ase, Ac, Cc, s, d
Cu (Ultimate Corrosion Level)
Vexp (Experimental Capacity)
Tensile Strength
Elastic Modulus
Corrosion Rate (mm/year)
Temperature (°C)
Humidity (%)
Carbon (C), Manganese (Mn), Silicon (Si), Phosphorus (P), Sulphur (S), Nickel (Ni), Chromium
(Cr), Molybdenum (Mo), Copper (Cu), Vanadium (V), Aluminium (Al), Titanium (Ti), Cobalt
(Co), Niobium (Nb), Tungsten (W), Tin (Sn)
Characteristic Strength (2013–2015)
Percentage Elongation (2013–2015)
Strain Ratio (2013–2015)
Ceq, Nb+Ta
Yield Load / Stress (Multiple Entries)
Weight to Length Ratio (W/L)
Table 3.1- Parameters in Non-Refined Dataset

28
Refined Dataset Parameters
Specimen ID
Bar Diameter (mm)
Bar Length (mm)
Initial Weight (g)
Final Weight (g)
Mass Loss (%)
Yield Strength (MPa)
Ultimate Tensile Strength (MPa)
Elongation (%)
Temperature (°C)
Modulus of Elasticity (GPa)

Table 3.2: Parameters in Refined Dataset

29
 Chapter 4: Experimental Work

In this experimental work, steel reinforcement bars (rebars) with diameters of 10 mm, 12 mm,
and 20 mm, each having a length of 500 ± 5 mm, are used to study the effect of corrosion on their
mechanical properties.

The bars are first weighed, then wrapped in gunny bags, and a 2% NaCl solution is regularly
sprinkled onto the gunny bags to induce corrosion in a controlled, moist environment. After
sufficient corrosion has occurred, the bars are cleaned using wire brushes to remove rust and
weighed again to determine the percentage of weight loss, which serves as a measure of corrosion.

The rebars are then subjected to tensile testing in a Universal Testing Machine (UTM), where
stress vs. strain graphs are generated, allowing the calculation of various mechanical properties
such as yield strength, ultimate tensile strength, and modulus of elasticity. The elongation is also
measured by assessing the change in length after the test.

A non-corroded bar is tested as a reference, and comparisons are made to examine how
corrosion affects the mechanical properties of the rebars. The study aims to establish a relationship
between the percentage of corrosion and the degradation of mechanical properties.

4.1 Selection and Preparation of Rebars

● Three rebar diameters are selected: 10 mm, 12 mm, and 20 mm, each cut to a
uniform length of 500 ± 5 mm.

● The initial weight (W_initial) of each bar is precisely recorded using a digital
weighing balance.

4.2 Corrosion Induction

● The rebars are wrapped in moisture-retaining gunny bags.

● A 2% Sodium Chloride (NaCl) solution is regularly sprinkled over these bags to

replicate a corrosive environment, mimicking conditions such as marine exposure
or de-icing salts in real life.

30
 ● This process is carried out over a defined time period to allow corrosion to
develop progressively.

4.3 Post-Corrosion Handling

● Once corrosion is visibly developed or the set exposure period ends, the rebars are
removed.

● They are cleaned using wire brushes to remove surface rust and corrosion products.

● The rebars are weighed again (W_corroded), and the percentage of mass loss is
calculated using:

4.4 Mechanical Testing

● All corroded bars are tested using a Universal Testing Machine (UTM).

● A non-corroded (control) rebar of each diameter is also tested for comparison.

● The UTM generates stress vs. strain curves, from which the following are evaluated:

 Yield Strength (Fy)

 Ultimate Tensile Strength (Fu)

 Modulus of Elasticity (E)

 Elongation (% change in length)

4.5 Analysis

● Mechanical property data from corroded bars are compared to that of non-corroded
(control) bars.

● A relationship between mass loss (%) and degradation in mechanical properties is

established.

● This analysis helps predict how corrosion affects structural performance over time.

31
Specimen ID Bar Length Corrosion Initial Final Weight Mass Loss (%) Test
Diameter (mm) Exposure Weight (g) Conducted
(mm) (g)

NC-10 10 500 ± 5 None (Control) 310 310 0% Tensile Test

C1-10 10 500 ± 5 2% NaCl (3 Days) 310 309.86 0.04% Tensile Test

C2-10 10 500 ± 5 2% NaCl (7 Days) 310 309.5 0.16% Tensile Test

NC-12 12 500 ± 5 None (Control) 444 444 0% Tensile Test

C1-12 12 500 ± 5 2% NaCl (3 Days) 444 443.85 0.033% Tensile Test

C2-12 12 500 ± 5 2% NaCl (7 Days) 444 443.5 0.12% Tensile Test

NC-20 20 500 ± 5 None (Control) 1234 1234 0% Tensile Test

C1-20 20 500 ± 5 2% NaCl (3 Days) 1234 1233.5 0.04% Tensile Test

C2-20 20 500 ± 5 2% NaCl (7 Days) 1234 1232 0.16 Tensile Test

Table 4.1 – Details of Corroded and Control Specimens with Mass Loss and Testing
Information

Note: C1 and C2 indicate different levels of corrosion exposure (e.g., C1 = mild, C2 = severe).

32
Note on Testing Limitation:

Due to the significantly reduced diameter of the corroded steel bar collected from the site (Fig 7),
it was not possible to conduct tensile testing using the Universal Testing Machine (UTM). The
bar's cross-sectional area was insufficient to securely grip or withstand the applied load during
testing, leading to the exclusion of this sample from mechanical evaluation.

QR Code for Accessing Non-Refined Dataset Used in BBN and ANN Modeling

QR Code for Accessing Refined Dataset Used in BBN and ANN Modeling

(Note: A detailed explanation of Bayesian Belief Network (BBN) and Artificial Neural
Network (ANN) modeling approaches, along with their implementation, is provided in
Chapter 5.)

33