AI reinforced concrete:

Algorithms predict concrete strength

Artificial intelligence (AI) A new research project has been outputs for new, out-of-sample inputs,
initiated between the Civil Engineering which may be within the given domain
and machine learning (ML) departments at the University of (interpolation), or even outside the
Pretoria and Neapolis University in domain (extrapolation). Predicting
algorithms are involved in a Cyprus with the main objective of out of the limits of a given domain is
developing design formulae for the a hard and highly unstable problem
vast variety of scientific and
prediction of the maximum capacity without interesting results. However,
of reinforced concrete structural recently, a numerical solution
industrial projects, contributing
members through the use of software- with extended prediction horizons
to the solution of emerging generated data that will be used to was published (Bakas, 2019). This
train machine learning algorithms. This significant breakthrough is one of the
problems, from data modelling innovative approach aims to replace main reasons why the current research
the actual physical experiment with 3D project is now feasible through its
and analysis to automatic detailed state-of-the-art finite element provision of new mathematical tools
modelling that will be used to generate and insights for predictions.
literature reviews, face millions of results for different types
of reinforced concrete structural Major advancements in the 3D
recognition and self-driving cars.
members, assuming different concrete detailed modelling of reinforced
and rebar material properties, while concrete structures have been
In recent years, researchers
various geometries and reinforcement published in the last decade, and
and professionals have been configurations will be considered. The more recent work managed to
artificially generated data will then be present numerical solutions that
engaging with AI algorithms used to develop predictive models are able to capture the mechanical
and construct new design formulae, behaviour of full-scale reinforced
in an unusually wide range of in closed mathematical form, for the concrete structures under extreme
needs of modern reinforced concrete cyclic loading conditions. Through a
scientific, technological and structural design methodologies. recent breakthrough, the ability to
The research project also foresees capture the static cyclic and dynamic
business fields. Regardless
a validation phase to verify the non-linear mechanical behaviour of
ability of the developed numerical reinforced concrete structures is now
of the database, their basic
models to capture physically derived feasible. The proposed algorithm,
purpose is to construct a experimental data found in the integrated within Reconan FEA,
international literature. was excessively validated through
prediction algorithm that will be the use of experimental results found
The relationship between in the international literature. These
accomplished by a mathematical independent and dependent variables experiments foresaw the study of
is often highly non-linear, and reinforced concrete beams with
model describing the complex mathematical models are aiming to and without stirrups, shear walls
form a generalised relationship that under monotonic and cyclic loading
interactions among some input
will be able to objectively link them. conditions, piles embedded in soil
The mathematical models’ success and the investigation of full-scale
variables and a corresponding
depends on the formulation and the reinforced concrete structures that
response (Dimopoulos, Tyralis, underlying theory, the algorithmic undergo cyclic static and non-linear
application, the handling of the dynamic ultimate limited state loading.
Bakas and Hadjimitsis, 2018). prediction errors, as well as the
generalisation of the results. The proposed modelling approach
foresees the use of the isoparametric
The final aim is to develop a numerical hexahedral element for discretising
model, capable of predicting new the concrete domain, while the

8 I N N O V A T I O N F O C U S  I N N O V A T E 1 4 2 0 1 9
The innovative use of
artificial intelligence
and machine learning
established through
this project could
revolutionise the
way Civil Engineers
design their
structures.
  Figure 1: Reinforced concrete four-storey building discretised through the HYMOD approach. A finite element mesh of
concrete and rebars.
hexahedral elements and more than
2.7 million embedded rebars. This
nuclear reactor building had a 2 260 m2
plan view area and a total height of
40 m. Given the numerical superiority
of the developed algorithm,
Reconan FEA is computationally
efficient in handling large-scale
numerical problems. It is thus
currently the software that has
managed to solve the largest model
found in the international literature
in terms of the number of embedded
rebars.

 Figure 2: Nuclear reactor building. Crack patterns and deformed shape due
to the serviceability load combination that foresees soil pressure. This project aimed to utilise the ability
of Reconan FEA to generate a massive
reinforcement is modelled as conditions, where the deterioration data set that will be fed to the state-of-
embedded rebar rods or beam of the concrete and steel materials is the-art AI algorithms in order to train
elements within the hexahedral also accounted for through the newly models that will be able to predict
mesh. This is shown in Figure 1. The introduced material damage factors. the maximum capacity of reinforced
microcracking is also accounted These factors are directly connected concrete structural members. The
for during the non-linear analysis, to the number of opening and closing construction of new design formulae
whereas macrocracking is modelled cracks during the non-linear loading will also be performed based on the
through the smeared crack approach. time history. scheduled research activities of this
Furthermore, the approach has research work by using advanced
the ability to numerically handle Figure 2 shows the predicted crack ML codes. This will set the base of
the opening and closing of cracks patterns of a NuScale reactor building revolutionising the way that civil
for both monotonic and cyclic that comprised around 181 000 engineers design their structures.

For the needs of this research project, Reconan Multirun was also recently developed and References
is currently utilised to generate hundreds of thousands of finite element models that will be
Dimopoulos, T, Tyralis, H,
used to analyse reinforced concrete beams without stirrups through the use of Reconan FEA. Bakas, NP and Hadjimitsis, D.
This first data set generation phase will be followed by the construction of beam models with 2018. Accuracy measurement
stirrups. Thereafter, reinforced concrete structural columns and shear walls will be analysed of random forests and linear
regression for mass appraisal
in line with the framework of the research project. models that estimate the
prices of residential
Upon successful completion of this first critical step, a second stage will follow that will apartments in Nicosia, Cyprus.
Advanced Geosciences, 45:
foresee an ultimate objective in the development of an AI software that will be able to non- 377–382.
linearly analyse reinfornced concrete structures, among other tasks, under static and dynamic Bakas, NP. 2019. Numerical solution
ultimate limited state loading conditions for seismic assessment requirements and the for the extrapolation problem
development of fragility curves. of analytic functions. Research,
2019: 1–10.

10 I N N O V A T I O N F O C U S  I N N O V A T E 1 4 2 0 1 9