Prediction of mechanical properties of sustainable recycled aggregate concrete using artificial intelligence techniques
In the last few decades, there have been many applications of AI techniques in the civil engineering field, such as structural health monitoring, prediction of different properties of concrete, design optimization of structural elements, etc. Data-driven AI techniques have also proven to be
2025-06-28 16:28:51 - Adil Khan
Prediction of mechanical properties of sustainable recycled aggregate concrete using artificial intelligence techniques
Project Area of Specialization Artificial IntelligenceProject SummaryIn the last few decades, there have been many applications of AI techniques in the civil engineering
field, such as structural health monitoring, prediction of different properties of concrete, design
optimization of structural elements, etc. Data-driven AI techniques have also proven to be
successful in the prediction of RAC mechanical properties, including the modulus of elasticity and
compressive strength (Behnood et al., 2015; Deng et al., 2018). However, the small amount of data
analyzed in the existing research compromises the ability of these models to accurately generalize
the underlying phenomena involved to predict the behavior of new sets of input data. Therefore,
creating reliable and more comprehensive datasets to fill the gap in the literature is intended in
this project. Furthermore, novel AI methods will be applied for the first time to predict significant
properties of RAC in the present study.
Despite a large amount of research carried out to determine the engineering properties of RAC,
the need for more diverse datasets is critical to developing reliable knowledge on the effects of the
addition of RA in concrete. This report aims at creating AI models, which after learning from
certain training datasets, can forecast accurate predictions on unnoticed data related to RAC for
application in the concrete industry. Accordingly, the objectives of the present report are outlined
below:
i. Analyze previous studies on the application of AI methods to predict the
compressive strength of novel concrete technologies available in the open literature.
Accordingly, determine the advantages and disadvantages of the different algorithms and
summarize their achieved performance, highlighting their contributions to the development of
mainstream concrete mixtures.
ii. Develop a large and reliable dataset for predicting the compressive strength of RAC, ensuring
that the created AI models can generalize the underlying principles of the compressive
strength of RAC.
iii. Develop AI models to predict the compressive strength of RAC given the growing
recognition that the mechanical properties of concrete are affected by the inclusion of RA.
To achieve the stated objective, a dataset was to be created by collecting enough data from
previous research literature. A comprehensive review of published articles is the foundation of
this research, to complete this theoretical thesis, a sufficient amount of data needed to be
extracted. A total of 25 research papers of previous work have been used in this research; each of
the 25 papers were published and found from different sources The researchers must have
accomplished their results experimentally by including the components of the concrete mixtures,
and their samples must consist of recycled concrete aggregate. The testing of the mechanical
properties or compressive strength of concrete must have been done after 28 days of samples’
curing. Each of the samples must include the following:
? Cement (Kg/m3)
? Water (Kg/m3)
? Natural coarse aggregate (Kg/m3)
? Recycled coarse aggregate (Kg/m3)
? Fine aggregate (Kg/m3)
? Fly ash (Kg/m3)
All of the data that have been selected to be used from the papers were extracted and added to a
data matrix on Microsoft Excel, MATLAB is one accessible coding software to be used
regarding generating or even using a built-in code or an algorithm to generate a specific fitting
the linear or non-linear function between dependent variables and one or more independent
variables. To get started with MATLAB, and as a first step, the data matrix that was created
previously on Microsoft Excel needed to be exported to MATLAB. The software offers two or
more methods to do so, the first one is by converting the Microsoft Excel file into a text file and
transfer it to MATLAB which is a convoluted method in this research
Support Vector Machine is an ML classification model that aims to find an optimal hyperplane
separating two different classes, the target of this method is maximizing the margin, which
represents the distance from the hyperplane to the closest point of each class, to attain a better
classification performance on test data MATLAB was chosen to build the prediction function
and model, Support Vector Machine algorithm was coded and used because it uses both
regularization and generalization. Generalization is the task of creating a pattern between the
data assuming the data is nonlinear, and regularization diminishes any possible issue of data
overfitting. In this work, the model considered the six different properties as input parameters,
Amongst the dataset, the training data constituted the highest because
training always requires more data to generate a more accurate general model. In
In particular, the training data in this research study formed about 70% which was found to
be sufficient after several trials. The remaining data were used for model validation and
testing. In summary, the distribution of the data was as the follows:
- 70% of the data has been selected randomly to be trained
- 15% of the data has been selected to be validated
- 15% of the data has been selected to be tested
Though there have been various studies on the application of traditional techniques to predict the
compressive strength of RAC, the present study aims at creating a large and more comprehensive
dataset and applying state-of-the-art AI techniques that have not yet been explored for RAC in the
open literature. The models presented in the present study will be executed using a programming
language. To utilize these models, the user can simply apply the development steps along
with hyperparameters reported in this study. The proposed models can also be used as a reference
guideline for designing eco-friendlier and more economical RAC mixtures in practice.
i. Thorough analysis of the literature on the application of AI methods to predict the behavior
of concrete in general.
ii. Development of the comprehensive dataset for prediction of the performance of RAC using
generalized AI models.
iii. Development of AI models for prediction of significant properties of RAC.
Conduct an analysis of previous studies on the application of AI methods to predict the
compressive strength of novel concrete technologies available in the open literature.
Accordingly, determine the advantages and disadvantages of the different algorithms and
summarize their achieved performance, highlighting their contributions to the development of
mainstream concrete mixtures.
ii. Develop a large and reliable dataset for predicting the compressive strength of RAC, ensuring
that the created AI models can generalize the underlying principles of the compressive
strength of RAC.
iii. Develop AI models to predict the compressive strength of RAC in view of the growing
recognition that the mechanical properties of concrete are affected by the inclusion of RA.
| Item Name | Type | No. of Units | Per Unit Cost (in Rs) | Total (in Rs) |
|---|---|---|---|---|
| Total in (Rs) | 29520 | |||
| compressive strength test test | Equipment | 3 | 2000 | 6000 |
| UTM lab test | Equipment | 4 | 3000 | 12000 |
| cement | Equipment | 3 | 860 | 2580 |
| fine aggregate | Equipment | 2 | 660 | 1320 |
| coarse aggregate | Equipment | 4 | 780 | 3120 |
| fly ash | Equipment | 2 | 550 | 1100 |
| moulds | Equipment | 5 | 680 | 3400 |