DOI: https://doi.org/10.21203/rs.3.rs-2026550/v1
Every year, a massive amount of literature is published on many aspects of microwave applications. Over the past few years, the need for progressive methods for assessing different properties of concrete, as well as for maintenance and monitoring of structures, has led to the development of microwave non-destructive testing (MNDT) techniques for preserving the civil infrastructure. As the number of studies being published in this field is growing fast, it is important to categorize the studies at deeper levels. In this paper, a concrete cube of M20 grade has been created in ANSYS HFSS for simulation purposes. An artificial rectangular crack was developed in the concrete cube and the simulation was run at dielectric constants from 1–10 with 0.5 intervals, loss tangents from 0.001,0.1 &0.2, and the gap between sample and circular waveguide from 0-7cm with 1cm interval. The artificial neural network was developed in python for Mean square Error(MSE) analysis and prediction of Electromagnetic properties(Dielectric constant and Loss tangent) with inputs as the S-parameter(real and imaginary) data of the Homogenous sample and sample with Rectangular crack. The result analysis shows the variation in Mean squared error for crack and No-crack concrete samples.
Non-destructive techniques play a key role in crack determination and the need for progressive methods for assessing different properties of concrete, as well as for maintenance and monitoring of structures, has led to the development of microwave nondestructive testing techniques (MNDT) for preserving the civil infrastructure. The MNDT technique is an imperative scientific breakthrough that consists of the development of calibration methods, probes, and techniques to detect defects, complex permittivity, moisture leakages, voids, cracks, initial rebar detection, inhomogeneity, etc. in materials [2]. This nondestructive method provides faster determination of the strength of the concrete concerning the destructive methods. Determining epsilon parameters would provide predicting the strength of a concrete sample [12]. Karl Bois et al. proposed that the statistical distribution of multiple reflection property measurements is used as a tool to determine the water-to-cement (w/c) ratio as well as the coarse aggregate volumetric distributions in several concrete mixtures [7]. Electromagnetic waves at microwave frequency range 7–13 GHz using free-space microwave method was used for measuring dielectric properties of concrete. It is observed that both dielectric constant of concrete increase with increasing compressive strength of concrete, but loss factors property of concrete decrease with increasing compressive strength of concrete. Most of the dielectric materials including wood and concrete are nonmagnetic, making the permeability very close to the permeability of free space [10]. An open-ended coaxial probe method is used for the measurement of real and imaginary parts of the complex permittivity of concrete. The increased loss factor at a higher frequency or with higher moisture content, or both, reduces the penetration depth of the wave in concrete, which compromises the benefit of increased detectability [14]. The Electromagnetic wave (EMW) NDT method could be applied to identify the slit orientation on the CFRP composites, while attention must be paid to the evaluation criteria when using various incident angles [15]. R. zoughi et al. proposed that these microwaves and mile millimeter-wave techniques offer certain advantages when detecting fatigue cracks such as the probe may or may not be in contact with the surface under examination, the dimensions of a crack can be estimated, particularly its depth, orientation, edge, and tip locations can be determined [16]. The novel use of microwave measurements is demonstrated to distinguish between mortars containing alkali-silica reactive aggregate and non-reactive aggregate [9]. According to Hamse Abdillahi Haji omer et al. antenna resonance plays a vital role in this concept of structural health monitoring where the S-parameter plot shows the variation of return loss (in dB) over a range of frequencies. The designed antenna resonates at 1.8 and 2.4 GHz respectively [6]. The properties of an open-ended rectangular waveguide probe radiating into Portland cement-based materials at 5GHz (G-band) and 10GHz (X-band) these microwave near-field sensing techniques are also useful for quality control purposes of a mixture in its early stages of curing [8]. Alireza A.chiniforush et al. highlight a lack of precision in the measurement of concrete’s dielectric properties using the commonly used coaxial probe MNDT by comparing the co-axial measurement with those obtained through the two-port technique in the frequency range of 2.6–3.95 GHz [4]. For the answer to the challenge of aging infrastructure, popular terms such as Internet of things and smart structures were coined as a result of the intersection between advances in other engineering disciplines with civil engineering to produce the new field of structural health monitoring (SHM) [5]. As far as microwave nondestructive testing techniques artificial intelligence plays a key role these days in damage detection. Implementing artificial intelligence to classify cracked from non-cracked surfaces has appreciable impacts in terms of sensing sensitivity, cost, and automation. Furthermore, applying artificial intelligence for post-processing the data collected from microwave sensors for handheld test equipment can outperform rack equipment with large screens and sophisticated plotting features [3]. The application of deep architectures inspired by the fields of artificial intelligence and computer vision has made a significant impact on the task of crack detection. In addition to providing higher accuracy in terms of detection, the crack localization is done at the pixel level thereby providing higher localization accuracy [18]. The responses of the multi-crack structure subjected to transit load were determined using the fourth-order Runge-Kutta numerical method and Finite element analysis (FEA) executed using ANSYS software to authenticate the employed numerical method [17]. Artificial Neural Networks, usually called Neural Networks (NNs), are computational models which consist of an interconnected group of artificial neurons. These neurons process information using a connectionist approach to computation. The ANN technique generally provides an efficient tool to classify problems associated with nonlinearities, when they are well represented by input patterns, and to avoid the complexity introduced by conventional computational methods [1]. Having the ability to perform various tasks with outstanding performance, machine learning (ML) has become a popular technique in almost every field. By providing a sufficient amount of data, ML algorithms can automatically digest intrinsic knowledge of the data, such as hidden structures or relationships. Traditional ML techniques require a predefined feature extraction stage to reduce the complexity of the data and make patterns more visible to learning algorithms[19]. An experimental study detailing the evidence related to using an ultrasonic pulse velocity test for accessing the cracking condition between steel and concrete reinforcement is presented along with a multilayer feedforward backpropagation perceptron ANN that is successfully able to predict the crack width thus eradicating the need for simplification assumptions needed to model complex nonlinear stress distribution [11]. This advantage of ANN as an AI model can be employed in NDT to recognize and estimate harmful defects before causing the system failure as well as improve the safety criteria [13].
Much research has been conducted over the past decades on Microwave based Nondestructive testing techniques using various methods with varying frequency ranges. Most of the work was conducted in determining the compressive strength of the concrete for various water-cement ratios using open-ended rectangular and circular waveguide sensors. Artificial intelligence is the recently developing technique in the field of civil structures for Predicting the physical properties, damage detection, mixing ratios, etc., In the present study, Mean squared error analysis was done in PYTHON software.
An open-ended circular waveguide sensor and a concrete sample were modeled using ANSYS Electronics (HFSS). Solution setup and frequency sweep were set up for the simulation to run between the required frequency and with the desired number of points. Analysis of the concrete sample with crack (Rectangular slit) and No-crack (homogenous) for different dielectric constants and loss tangents was done. The resultant S-parameter data S11(dB, real and imaginary) was used to prepare the dataset to predict the Electromagnetic properties of the concrete sample. The flow chart below shows the step-to-step procedure followed.
The use of the Ansys Electronics solution suite minimizes testing costs, ensures regulatory compliance, improves reliability, and drastically reduces product development time.
In radio-frequency engineering and communications engineering, a waveguide is a hollow metal tube (rectangular or circular in cross-section) that transmits electromagnetic energy from one point to another. A waveguide can be defined as a geometrical structure capable of propagating electromagnetic energy in a preferred direction in space within a certain frequency range. Whenever high-frequency electromagnetic wave propagation is present, waveguides are used. Waveguides are commonly used as transmission lines at microwave frequencies, where they connect microwave transmitters and receivers to their antennas. A waveguide with a circular cross-section is referred to as Circular Waveguide. An open-ended circular waveguide sensor was designed in Ansys HFSS. Once the object was drawn a properties window opens up where we can give the position of the waveguide as well as the material will be assigned as “air”.Then a dielectric probe was designed which helps for the collection of S-parameter data from the concrete cube. Then the excitations and boundaries were assigned to the sensor and further steps were carried out i.e. Concrete cube designing and simulation steps.
Concrete is commonly referred to as "man-made rock." Concrete is a widely used construction material made up of cement, water, fine aggregate (sand), coarse aggregate (rocks), air, and in some cases admixtures. When aggregate is mixed with water and cement, the outcome provides a fluid sludge that is conveniently cast and molded into virtually any shape and so that strengthens over a few hours to produce a concrete specimen that unites the components into a durable rock specimen that has so many uses. A concrete cube was modeled by drawing a box and then properties were assigned such as dimensions and the position of the concrete cube as right above the circular waveguide. A radiation box was modeled to restrict the computational area.
Artificial neural networks (ANN) are based on the concept of the human brain neural network since the human brain is a highly complex, nonlinear parallel computer consisting of neurons responsible for undertaking simultaneous multiple tasks. ANNs are also composed of similar structures with several neurons that are excellent for pattern recognition, prediction, classification, and categorization. The fundamental benefit of using ANN over traditional analytical modeling is the lack of need to make simplification assumptions. By adopting ANN, researchers have been able to recognize patterns in complex nonlinear problems. In the present study, a neural network with a regression layer for the output was used.Frequency(GHz), and S11(real and imaginary) values were the inputs to the neural network whereas Dielectric constant and Loss tangent are the outputs. A standard scaler is used for the standardization of the data as it contains various varied values. The total data was divided randomly as 80% for training and 20% for testing. A scaler fit model was used for training the data. 3 layers of tangential activation functions were used with 50 neurons in each layer and then the mean squared error loss function was determined for a various number of epochs. A comparison of true and predicted values for the homogenous concrete sample for all the gaps was determined. Further, the Mean squared error for the sample with rectangular crack was determined. The architecture of the neural network was as shown below.
3.1 Mean Squared Error
The mean squared error for the homogenous sample with a 0cm gap between Open ended circular waveguide sensor and the Concrete sample is shown in fig.3 and the mean squared error for the concrete sample with a rectangular crack was as shown in fig.4.
Table 1
Mean Squared Error for Homogenous concrete sample and rectangular crack sample
|
S.No. |
GAP(between concrete sample and sensor in cm) |
MEAN SQUARED ERROR @Epochs |
|
|
HOMOGENOUS SAMPLE |
RECTANGULAR CRACK SAMPLE |
||
|
1 |
0 |
0.17 @ 280000 |
0.24 @ 240000 |
|
2 |
1 |
0.17 @ 260000 |
0.46 @ 280000 |
|
3 |
2 |
0.23 @ 220000 |
0.44 @ 280000 |
|
4 |
3 |
0.17 @ 280000 |
0.40 @ 300000 |
|
5 |
4 |
0.17 @ 240000 |
0.36 @ 300000 |
|
6 |
5 |
0.14 @ 280000 |
0.32 @ 300000 |
|
7 |
6 |
0.18 @ 280000 |
0.38 @ 300000 |
|
8 |
7 |
0.27 @ 280000 |
0.64 @ 300000 |
The Least Mean Squared error occurs approximately between 260000 and 280000 epoch for all gaps for a homogenous sample and rectangular concrete sample it is mostly at 300000epoch. The mean squared error values for 0cm,1cm,3cm,4cm, and 6cm gap are the same which is 0.17 for a homogenous concrete sample whereas it varies for rectangular crack concrete sample.
Further study could be done for detecting the crack in the concrete sample as well as the crack properties and the variation in Dielectric and Loss tangent values for Homogenous concrete samples and concrete samples with crack.
Flow chart is available in the Supplementary Files section