Applications of Computational Intelligence in Concrete Technology

Sakshi Gupta, Parveen Sihag, Mohindra Singh Thakur, Utku Kose, Sakshi Gupta, Parveen Sihag, Mohindra Singh Thakur, Utku Kose

• Publisher: CRC Press
• Year: 2022
• Print ISBN: 9781032013022
• eBook ISBN: 9781000600582
• Topic: Tecnologia e ingegneria
• Subtopic: Intelligenza artificiale (IA) e semantica

Computational intelligence (CI) in concrete technology has not yet been fully explored worldwide because of some limitations in data sets. This book discusses the selection and separation of data sets, performance evaluation parameters for different types of concrete and related materials, and sensitivity analysis related to various CI techniques. Fundamental concepts and essential analysis for CI techniques such as artificial neural network, fuzzy system, support vector machine, and how they work together for resolving real-life problems, are explained.

Features:

• It is the first book on this fast-growing research field.
• It discusses the use of various computation intelligence techniques in concrete technology applications.
• It explains the effectiveness of the methods used and the wide range of available techniques.
• It integrates a wide range of disciplines from civil engineering, construction technology, and concrete technology to computation intelligence, soft computing, data science, computer science, and so on.
• It brings together the experiences of contributors from around the world who are doing research in this field and explores the different aspects of their research.

The technical content included is beneficial for researchers as well as practicing engineers in the concrete and construction industry.

1 Usage of Computational Intelligence Techniques in Concrete Technology

Utku Kose

Suleyman Demirel University

DOI: 10.1201/9781003184331-1

CONTENTS

1. 1.1 Introduction
2. 1.2 Computational Intelligence Models for Concrete Technology
   1. 1.2.1 Artificial Neural Networks (ANN)
   2. 1.2.2 Adaptive Neuro-Fuzzy Inference System (ANFIS)
   3. 1.2.3 Genetic Algorithm (GA)
   4. 1.2.4 Random Forest (RF)
   5. 1.2.5 Random Tree (RT)
   6. 1.2.6 Linear Regression (LR)
   7. 1.2.7 M5P Model
   8. 1.2.8 Support Vector Machine (SVM)
3. 1.3 Predictive Computational Intelligence in Concrete Technology
   1. 1.3.1 Prediction of Compressive Strength of the Concrete
   2. 1.3.2 Prediction of Ultrasonic Pulse Velocity of the Concrete
4. 1.4 Conclusions
5. References

1.1 Introduction

Today’s modern world is approaching rapid advancements thanks to many digitally transformed tools. However, such tools are highly supported with computational aspects forming analytical and algorithmic solutions. Here, the most important contribution is by the field of artificial intelligence (Dean et al., 1995; Neapolitan & Jiang, 2018). Artificial intelligence is known as the most innovative research area for not only the present world but also for the future. When artificial intelligence is examined, it is possible to observe that there is a wide variety of algorithmic architectures effectively used for developing different techniques. These techniques are generally used to ensure the following solution methods (Garnham, 2017; Lan, 2020; Neapolitan & Jiang, 2018):

• Iteratively finding the most optimum values for the target mathematical model of the problem,
• Iteratively processing data samples to find a hidden pattern, which could not be detected by using traditional analytical approaches.

The first solution method is generally associated with optimization-oriented efforts, which are examined under classical optimization by Mathematics. On the other hand, the second solution is related to statistically empowered iterative algorithms, which can optimize some parameters so that the used algorithm may be sensitive for further data flows. Briefly, these solutions are used to form different soft computing techniques under the umbrella of artificial intelligence. Since these techniques are associated with computational runs, a more formal name, computational intelligence, has been widely used to define them (Azar & Vaidyanathan, 2015; Eberhart & Shi, 2011). For today’s advanced problems, which are requiring more contributions by alternative data-processing algorithms, computational intelligence is often used for deriving alternative findings and shaping the way of the scientific literature.

When it is deeply examined, it is possible to see that computational intelligence techniques often use iterative data-processing so that simulating a learning mechanism, which seems to be a digitally transformed way of human-side learning capabilities. In the context of artificial intelligence, these techniques are collected under also machine learning. Machine learning is known as the most effective sub-area of artificial intelligence as machine learning techniques can learn from past data samples to ensure predictive or descriptive findings for newly encountered new data flows (Dean et al., 1995; Eberhart & Shi, 2011; Neapolitan & Jiang, 2018). Although descriptive solutions tend to be associated with instant connections among the observed data, predictive solutions are mostly for deriving new data, which are used for determining the future state of events, problems, and so on. When the usage of computational intelligence techniques is observed, it is possible to see that there is a great interest for predictive solutions since determining future state is among popular focuses, and even descriptive outcomes are highly connected with predictive further steps.

Early periods of computational intelligence techniques are associated with not daily life before. However, with more developments within the last quarter of the 20th century and the 20-year period of the 21st century so far, foundations by computational intelligence have been widely seen in our daily life. It is remarkable that for even today’s daily life solutions, there has always been an effort in terms of solving engineering problems. That situation is similar when different tools and objects, which have a critical role in human life, are discussed. From healthcare to education, or defensive technologies to communication, Computational intelligence is highly associated with all different fields of life (Andina & Pham, 2007; Kose & Koc, 2014; Yu et al., 2018). Here, construction technologies have been often supported by predictive outcomes of Computational Intelligence.

Concrete technology is a remarkable component where computational intelligence techniques have been widely employed. As associated with that, the literature needs reference works for the outcomes reported for the combinations of different computational intelligence techniques to deal with different applications of concrete technology. As an introductory approach, this chapter briefly gives pre-information about this book. The following sections discuss the use of different computational intelligence/soft computing solutions to find effective outcomes. However, this chapter uses original information by the authors and enables readers to have the necessary knowledge for understanding this book.

As associated with the aim of this chapter, the next section provides information about the models used in this book. After that, the third section is devoted to brief information about this book and the outcomes. This chapter ends with conclusions and some ideas about the future.

1.2 Computational Intelligence Models for Concrete Technology

Computational Intelligence is a wide family of soft computing techniques to deal with advanced problems. By focusing on the field of artificial intelligence, it is possible to examine hundreds of different techniques with even variations so that there is a great scope for the researchers to use alternative solutions for deriving findings. Figure 1.1 provides a general scheme for widely used Computational Intelligence models.

FIGURE 1.1 Computational intelligence models for concrete technology applications.

In this book, some remarkable techniques were used. However, the majority of the techniques were with the use of neural network-based models. That’s because of the advanced accuracy capabilities of neural networks and perceptron foundations. However, there are also some other strong competitive techniques and additional components to ensure hybrid formations for improved outcomes. The following sections include general information on these techniques as combined with the original content by the authors.

1.2.1 Artificial Neural Networks (ANN)

Artificial neural network (ANN) is the strongest and long-time used technique coming from the machine learning side of artificial intelligence. With its computational capabilities, it is among the widely used computational intelligence models to deal with advanced problems. Briefly, a typical ANN architecture had set of nodes which divided into layers (input layer, hidden layer can be multi-layers or zero, output layer), biases, weights between these nodes, and the ways of flowing the information on the neural network as feed-forward networks, the signal in this model transfer in one way from the input layer to the output layer. Furthermore, in feedback networks, the signal can travel in both directions by the loops between the hidden layers, and this type of connection has a memory for an internal state to process a sequence of inputs (Abraham, 2005). According to Benardos and Vosniakos (2007), there are four elements to clarify optimal ANN’s architecture: number of layers (especially hidden layers), number of nodes (neurons) in each layer, type of active function in each node/layer, and the algorithm used in training plays an essential role in determining the value of weight and biases of nodes. The backpropagation (BP) algorithm is a supervised learning algorithm; it is used to train a multi-layer neural network, which is done by iteratively updating the weights of nodes to minimize the value of the error function in weight space, and this technique is called as the gradient descent or delta rule (Ampazis et al., 1999; Zweiri et al., 2003). Figure 1.2 represents a general scheme for a typical ANN (Bre et al., 2018).

FIGURE 1.2 Artificial neural networks (Bre et al., 2018).

1.2.2 Adaptive Neuro-Fuzzy Inference System (ANFIS)

Adaptive neuro-fuzzy inference system (ANFIS) is a technique in which the number of hybrid combinations of FIS and ANN is used to find out the various aspects and, as a result, used to prepare the model forecast for the future responses. This inference system can combine the benefits of fiber length with neural network principles in a single framework to solve problems. ANFIS is a hybrid machine learning technique that transforms a given input into a target output by using a fuzzy interference system model. It also enables the analysis easier. The training and test data set is utilized to create the ANFIS model, and evaluate the built model’s prediction ability during the model building process, respectively. The concept behind using a test data set for ANFIS model validation is that the model starts overfitting the training data set after a certain point in the training. The parameters decide the performances of a fuzzy system which describes the membership function and the rule-based use in the system specifies how often a fuzzy system performs. The various parameters can be modified and can be combined with a fuzzy system and ANNs to build a neuro-fuzzy system (Ahmadi-Nedushan, 2012; Poddar et al., 2018; Kumar et al., 2020; Poddar et al., 2020). The benefits of both approaches are combined in neuro-fuzzy systems, which merge the natural language description of fuzzy systems with the learning properties of ANNs. ANFIS is a composite in which the parameters of the fuzzy system are fixed using an adaptive BP learning algorithm. By using the fuzzy sets, ANFIS combines the human-like reasoning structure of fuzzy systems. ANFIS is a multilayer feed-forward network where each node performs a specific role on incoming signals (node function). In this process two inputs x and y can be considered and one output z. During the ANFIS operation mainly five processing stages take place (Figure 1.3). The main objective of the ANFIS is to incorporate the best features of fuzzy systems and neural networks. ANFIS approach mainly uses the membership function parameters which can be adjusted using either al...

Table of contents

• Cover
• Half Title
• Series Page
• Title Page
• Copyright Page
• Table of Contents
• Preface
• Editors
• Contributors
• Chapter 1 Usage of Computational Intelligence Techniques in Concrete Technology
• Chapter 2 Developing Random Forest, Random Tree, and Linear Regression Models to Predict Compressive Strength of Concrete Using Glass Fiber
• Chapter 3 Prediction of Compressive Strength at Elevated Temperatures Using Machine Learning Methods
• Chapter 4 Implementation of Machine Learning Approaches to Evaluate Flexural Strength of Concrete with Glass Fiber
• Chapter 5 A Comparative Study Using ANFIS and ANN for Determining the Compressive Strength of Concrete
• Chapter 6 Prediction of Concrete Mix Compressive Strength Using Waste Marble Powder: A Comparison of ANN, RF, RT, and LR Models
• Chapter 7 Using GA to Predict the Compressive Strength of Concrete Containing Nano-Silica
• Chapter 8 Evaluation of Models by Soft Computing Techniques for the Prediction of Compressive Strength of Concrete Using Steel Fibre
• Chapter 9 Using Regression Model to Estimate the Splitting Tensile Strength for the Concrete with Basalt Fiber Reinforced Concrete
• Chapter 10 Prediction of Compressive Strength of Self-Compacting Concrete Containing Silica’s Using Soft Computing Techniques
• Chapter 11 Using Soft Computing Techniques to Predict the Values of Compressive Strength of Concrete with Basalt Fiber Reinforced Concrete
• Chapter 12 Soft Computing-Based Prediction of Compressive Strength of High Strength Concrete
• Chapter 13 Forecasting Compressive Strength of Concrete Containing Nano-Silica Using Particle Swarm Optimization Algorithm and Genetic Algorithm
• Chapter 14 Prediction of Ultrasonic Pulse Velocity of Concrete
• Chapter 15 Evaluation of ANN and Tree-Based Techniques for Predicting the Compressive Strength of Granite Powder Reinforced Concrete
• Chapter 16 Predicting Recycled Aggregates Compressive Strength in High-Performance Concrete Using Artificial Neural Networks
• Chapter 17 Compressive Strength Prediction and Analysis of Concrete Using Hybrid Artificial Neural Networks
• Index