A review on implementation of AI/ML in MNDT

DOI : 10.17577/IJERTV12IS070042

Download Full-Text PDF Cite this Publication

Text Only Version

A review on implementation of AI/ML in MNDT

Aksh Patel

Mechanical Engineering Department, ITNU

Nirma University, Ahemdabad, India

Abstract – Electrically insulating composites that are lighter, stronger, more rigid, and more durable than metals have replaced metals in numerous uses because to advancements in materials engineering. Due to the drawbacks or curb of the traditional non-destructive testing (NDT) methods, which include thermography, eddy current testing, ultrasonic, X-ray, and magnetic particles, these synthetic materials require alternative inspection procedures. Due to inadequate signal penetration, typical non-destructive inspection methods operating at low frequencies necessitate removing insulating material to permit examination. a number of high-frequency inspections Without removing the insulations, methods like the microwave approach have demonstrated successful examination in finding the problem under them. A potential method to find flaws in both metal and composite materials is the use of microwave NDT with open-ended rectangular waveguides (OERW). The microwave approach, however, confronts a number of difficulties, including inadequate spatial imaging, significant inaccuracies in defect size and depth caused by variations in stand-off distance, the best frequency point choice, and the existence of outliers in microwave measurement information. For determining corrosion beneath insulation, the microwave method in combination with machine learning strategies offers great promise and feasibility. For the purpose described here, microwave NDT using OERW in combination with strong artificial intelligence techniques has a great deal of potential and feasibility. Because the influence of artificial intelligence methods has been demonstrated in several conventional NDT techniques, combining methods of artificial intelligence with methods for signal processing is extremely likely to increase the effectiveness and resolution of the microwave NDT technique.

Keywords Artificial Intelligence, Machine Learning, Microwave Inspection


    Corrosion Under Insulation (CUI) is a serious flaw in many essential applications, particularly in the oil and gas, nuclear, marine, aerospace, and power developing sectors[1], [2]. Due to inherent coating flaws, moisture intrusion, and structural weakening in these industries, corrosion grows covertly in the metal backing behind insulation[3], [4]. This structure's vulnerability might result in a catastrophic collapse of asset integrity, which would have a number of negative effects on people's health and safety as well as production losses and

    maintenance costs[5], [6]. Therefore, to minimise the aforesaid effects and improve the system's all inclusive integrity, a precise and speedy examination for CUI identification is necessary. NDT is a auspicious idea in practical use to identify, track, and gather important data on the severity of CUI.

    In the realm of non-destructive testing (NDT), the advancement of imaging techniques for examining items that are physically inaccessible has been a study focus for many years. Nondestructive testing (NDT) is the process of assessing changes in a material's various properties, such as delamination, which corrosion, cracks, and fatigue, as well as any inside flaw or metallurgical condition, without compromising the material's integrity or suitability for use[1]. NDT is essential to many industrial applications, particularly in the fields of aviation, power generation, petro chemistry, railroads, and automobiles. To reduce maintenance costs and increase the overall system's safety and dependability, NDT examinations to evaluate structural or component damage are essential[7].

    Due to field penetration restrictions, the requirements of replacing or coating composite materials with metals in numerous uses need the use of an alternate inspection methodology to the usual NDT method. These components' capabilities are affected by a variety of issues as a result of wearing and the repeated process, such as corrosion, cracks occurring in the undercoating of metallic substrates, and breakdown between coverings and substrates made of metal. Since electromagnetic waves at microwave wavelengths have the potential to provide composite materials with improved inspection resolution, the microwave non-destructive testing (MNDT) approach is ideally suited for examining composite materials [4].

    Microwave signals are a very appealing applicant for composite inspection because, unlike ultrasonic sound and acoustic signals, they can interact with the inner structure of composite materials like dielectric insulations and can penetrate inside them. The conventional microwave-based techniques, Notwithstanding their encouraging outcomes, they struggle with an assortment of issues, such as data intricacy, poor geographic quality of images, hazy defect form as an outcome of distance between objects fluctuations, and optimal frequency site selection, all of which affect the geometric assessments of faults[5], [8]. Consequently, the development of smart systems, including methods based on neural networks, makes it feasible for soft computing techniques to emerge, which in turn makes it viable to address these issues. Machine learning techniques' great capacity to handle complicated data makes them useful to handle the complexity of microwave data and minor fluctuations as well as to increase the sensor's sensitivity and spatial resolution of images[9].

    Some studies recommend the use of artificial intelligence (AI) for finding flaws and identification, citing the development of a classification system based on machine learning that automates defect detection throughout fabrication or in service and raises the standard of NDT inspection [6]. The present efforts to combine MNDT and AI, notwithstanding the individual approaches' bright futures, have a great chance of overcoming the aforementioned difficulties.

  2. CONVENTIONAL NON-DESTRUCTIVE TESTING Conventional non-destructive testing (NDT) techniques are

    employed in many different factories, such as from the oil and gas, renewable energy, rail, engineering and manufacturing, shipbuilding, steel manufacturing, biochemistry, and amusement parks and fairs to name a few.

    Conventional NDT techniques include x-ray inspection, thermographic inspection, eddy inspection, magnetic inspection and ultrasonic inspection nevertheless, they all have one thing in common: the substances being tested are not affected in any way. These techniques are tried-and-true and often used.

    1. X-ray Inspection

      One of the radiometric inspection methods for spotting penetrating radiation modification in objects is the X-ray examination methodology [10]. The X-rays approach uses radiation from the electromagnetic spectrum with short wavelengths to assess the thickness of the specimen and capture photographs of the framework's contour [11]. A detector is used to calculate the quantity of radiation that is flowing through the test sample. In contrast to defect-free sections, Figure 1 shows how cavities and interruptions change the amount of rays obtained at the detector, revealing thickness variations in the item under investigation.

      Figure 1 :- Xray inspection

      In epoxy-coated mild steel, the Compton X-ray backscattering technique is used to locate and examine the corrosive undercoating [12]. The radiographic picture and the usual profile of grey levels in the radiological picture show the thickness variance in the test specimen. The existence of

      thicknes loss indicates that the specimen has corrosion damage. The greyscale picture generated by scanning the experimental sample in a linear motion serves as evidence of the test samples level of corrosion. But the quantitative analysis the corrosion level cannot be reliably estimated using information gleaned from the observed corrosion's grayscale. Additionally, wearing protective gear is necessary when doing an X-ray check since prolonged exposure to radiation from X-rays has negative effects on one's health and safety [13].

    2. Thermographic Inspection

      Figure 2 :- Thermographic inspection

      Safety and medical concerns brought on by radioactive sources can be eliminated using the thermography examination method. This method is another NDT methodology for detecting the specimen Under Test (SUT)s temperature distribution[14]. An infrared camera that can detect even the smallest thermal differences is used to record the temperature distribution. Figure 2 [15] depicts the classification of thermography inspection procedures into active as well as passive thermography.

    3. Eddy Current Inspection

      Quantitative information, such as the extent of CUI, may be provided via the Eddy Current Examination (ECT) approach. Heavy industries often analyse conductive materials using the ECT method to find surface and subsurface flaws[16], [17]. As seen in Figure 3 [15], the ECT probe employs a main transmitter coil to determine the magnetic field surrounding the examined metallic spot. As a result, ripples are produced on the metallic item. The eddy currents generate an additional magnetization that is polarisation in the opposite manner as the originating field [18]. Whenever an error exists, the eddy currents deteriorate and have an effect on the second-degree field of magnetism [19]. For the purpose of trying to differentiate amongst a site with a defect and one that is free of errors, a coil or magnetised sensor is employed to determine the disparity in magnetization.

      Figure 3 :- Eddy currents inspection

    4. Magnetic Inspection

      Figure 4 :- Magnetic separation method

      The irregularities in the sample or specimes causes magnetic flux to seep [20]. In Figure 5, a magnetised particulate scanner that is placed close to the surface being examined may absorb magnetic ions and provide the precise position, shape, size, and level of the gap. In the areas of heavy engineering, welding faults, and applications in aerospace, MT is frequently used to evaluate surface breaks, flaws, and discontinuities on surfaces that are either being produced or utilised.Because it relies on the materials capacity to be magnetised and can only identify flaws a few millimetres below the surface, the MT approach in NDT currently struggles with dependability and sensitive to identify surface fractures [21]. In [22], it is examined if MT can reliably and sensitively identify fractures in welded components. The study demonstrates that the method is unreliable for detecting fractures with a tiny size and is not able to identify any faults with an expanse of less than 1.5 mm.

    5. Ultrasonic Inspection

    Figure 3:- Ultrasonic testing

    The conventional ultrasonic testing (UT) system is shown in Figure 3. It uses a transducer to produce a beam of high- frequency sound waves that are directed towards the specimen being tested, pass through it, and are reflected off of its rear surface or other flaws [23]. The wave that is reflected is changed into an electrical signal that may be analysed to find flaws and their locations.

    Ultrasonic technology is utilised in NDT for a number of inspection purposes, including the identification of cracks, delaminations, and corrosion [24] studies the various amounts of corrosion between the gran of many stainless-steel pipes. The investigation measured a longitudinal ultrasound wave over the tube wall using both receiver and transmitter ultrasonic transducers. The study demonstrates that when corrosion levels rise, its ultrasonic non-linear coefficient increases dramatically. As an outcome, ultrasonic NDT can identify and gauge the degree of corrosion in steel pipes. Nevertheless, since the length of the examined pipe degrades along the longitudinal wave, it had a significant impact on the unpredictable coefficient.

  3. MICROWAVE NON-DESTRUCTIVE TESTING Microwave NDT techniques have had a lot of success in

    inspecting composite materials during the past 20 years. [25] [27]. Due to their reliability, strength, and lightweight qualities, composite materials have gained popularity and have begun to replace metals in numerous applications, including the aerospace, automotive, and aviation sectors. Since microwave signals may pass through composite materials and communicate with the materials exterior and internal structure, this necessitates the widespread use of microwave techniques [28], [29], present microwave nondestructive testing (NDT) employing a microwave line of communication (MTL) sensor. The examined specimen serves as an electrically conducting substance for a microwave circuit in the MTL process. Any flaw in the specimen causes a change in the material permittivity, which is measured in the signals responses.These alterations are utilised to spot anomalies in the specimen under inspection, such as variations in the material layers and defect size and position.

    The amount of pollen cleanliness is determined using MTL, a very efficient technology for assessing the permeability of resources, which is used to ensure manufacturing quality [30]. Additionally, MTL is employed in [29] to find hidden glass and stone bits that were added to the vegetable salad. To find foreign things in food, the suggested approach calculated the ratio of the transmitted signals real and imaginary components.Microwave NDT is a strong method for evaluating under-coating flaws like CUI in comparison to traditional NDT. The composite insulation

    may be deeply penetrated by electromagnetic waves with frequencies between 300 MHz and 300 GHz, and these waves are sensitive to variations in the thickness of the metal substrates surface [15], [31], [32]. The microwave NDT method has a number of benefits, including non-contact inspection and the lack of a couplant required for the transmission of signals into the material being tested [33][35]. Unlike ultrasonic waves, microwave impulses can pass through the insulation of composite constructions and interact with the underlying structural materials [36].

    Microwave waveguides known as "Open-Ended Cylindrical Waveguides" (OERW) are frequently used to investigate metallic surfaces coated with hybrid and slippery polymers. In order to check faults including delamination, which fractures, and CUI, OERW is frequently utilised. A high-frequency signal is sent in the course of the SUT by the OERW. Modifications in the radio frequency signal's reflection index are used for capturing the faults, resonance frequency, magnitude, and phase. The microwave NDT methods described in [37] use a sweeping frequency. In order to assess the plated environments corrosion on mild steel, the OERW detector runs in the K-band.


    Conventional NDT methods continue to have various drawbacks that lower the effectiveness of examination of CUI, such as inadequate spatial image processing, field penetration restrictions, and fuzzy defect shapes. On the advancements of NDT automation, recent study is very trustworthy. As a result, there is less reliance on the knowledge and experience of the operators. To increase the accuracy of the investigation of CUI, signal processing using classifiers constructed from machine learning is used in NDT procedures. Additionally, it can raise the CUI levels forecast accuracy. Three phases are often involved in NDT techniques based on Machine Learning (ML) approaches: pre-processing, feature extrication, and categorization [38]. Prior to data analysis, a seriesof activities called pre-processing are carried out with the goal of cleaning up and removing extraneous data in order to reduce analytical mistakes. The feature extraction step, on the other hand, seeks to extract a number of useful characteristics from a huge amount of data for improved data interpretation.

    Figure 6 :- The process of machine learning

    The collected characteristics are categorised using a classifier based on machine learning during the classification step in order to divide comparable data into different categories, such as defect versus fault-free. Simulation as well as information processing have definitely improved thanks to machine learning. ML-based models have proven to have high fitting analysis capabilities and good forecast precision. Configuration simplification, structural hypotheses, fault performance evaluation, experimental advancement in the last few years, and materials science all apply machine learning (ML) [39]. The ML-based models employed in CUI are used to identify, image, and assess the seriousness of CUI or compute the failure probabilities and growth rates of CUI defects. Numerous algorithms that are often employed in the three phases are explained in the following sub-sections.

    1. Pre-Processing

      Only acknowledged early processing techniques, such as the separate wavelet transforms transform (DWT), the transformation using Hilbert (HT), and variational mode dismemberment (VMD), are described as examples for the signal processing step in this section. For the purpose of identifying and eliminating electrical noise from indications, the DWT is a divided wavelet transformation (WT) [128]. WT divides a data stream into several levels that represent various frequencies [40], [41]. The electrical noise is based on the WTs location on each scale. The noise data is used to create a threshold. By removing a threshold number below the threshold, the amount of noise is effectively reduced. The DWT approach is used to remove background noise from an electrical signal acquired using the plated Perfect Renewable Conductor (PEC) in order to identify the miniatured delamination a dual-ridges OERW [42].To forecast the extent of the delamination, are the DWT is used when combined with the signal strength analysis. The use of DWT output combination of methods significantly

      improves the accuracy of the fault magnitude measurement. To ensure that the signals data will be retained, the quantity of scales that are used to deconstruct the signal must be carefully chosen.

    2. Feature Extraction

      The practise of feature extraction makes data sets easier to manage for analysis of data [43]. A selection technique used in feature extraction tries to retrieve the dominant information and eliminate redundant raw data factors. These repeated properties reduce the predicting model's reliability and acceleration, which has a significant impact on the model's diagnostic correctness and execution efficacy. The frequency sweeping process in microwave NDT creates a feature vector with a high degree of dimension. The large dimensionality of the characteristics makes the machine learning methods more computationally complicated and time-consuming to process. Due to their ability to conduct analysis of features and dimensionality reduction simultaneously, Partial Least Squares (PLS), Principal Com component analysis (PCA), and Nonnegative Mapping Factorization (NMF) are covered in this section.

    3. Classification Stage

    Throughout the classification process, a classifier approach is employed to classify a set of information into subgroup groups according to their the outdoors. The two forms of data segmentation involve supervised deep learning (SML) methodologies and unorganised machine learning approaches. Supervised training provides a function for translating inputs to expected outputs [44]. These techniques need input informationor status labelingfrom well-known domains. Throughout the classification process, a classifier approach is employed to classify a set of information into subgroup groups according to their the outdoors. The two forms of data segmentation involve supervised deep learning (SML) methodologies and unorganised machine learning approaches. Supervised training provides a function for translating inputs to expected outputs [44]. These techniques need input informationor status labelingfrom well-known domains. Figure 8 depicts the SML procedure. When learning without supervision, it picks up framework from the data that have not been labelled. When previous information is unavailable or the number of samples is tiny, unsupervised learning can be helpful [45].

    One of the SML algorithms is the artificial neural network (ANN). The artificial neural network (ANN) model is a crucial tool for handling complex problems [46]. Information is fed to the first level's brain neurons through the hypothetical network from the information source. The neurons transform the data into a sensation and send the impulse as an input to the synapses in the layer below. As the artificial brain learns, the accuracy with which the outcomes were generated increases. The neural networks extrapolate on earlier learning to arrive at a choice. A typical ANN's architecture is shown in Figure 9.


    AI based on machinelearning algorithms play a crucial role in terms of output processing for intelligent systems in NDT. Post-processing for evaluating and detecting flaw. AI can deal with the intricate nature of the data that has been gathered to increase the sensitivity of fault detection [47]. Machine learning algorithms offer reliable and real-time inspection in traditional NDT methods while eliminating complicated mathematical modelling. However, despite its success in other NDT approaches that attain high inspection accuracy, the application of AI in electromagnetic NDT (MNDT) is still limited. As already noted, conductive and non-conductive objects may be inspected more effectively using microwave-based . The OERW sensor responsiveness was improved in the first effort to deploy adaptive MNDT in [47]. To locate the minute fluctuations in the reflected coefficient that are hard to see, the approach used SVM and ANN classifiers. In order to improve fault detection, the great capacity of the artificial intelligence (AI) model to locate minute changes is applied. The method is used to divide coated steel items into fault and defect-free categories. The flawed and defect-free specimens are used to calculate the reflection coefficient, which is then divided into both testing and training samples, totaling 45 and 371 samples, respectfully. The samples from training are classified as having defects and not having defects. By 30 MHz steps, the spectrum is swept from 12 to 18 GHz using a network analyzer with vector networks (VNA). 201 frequency points are thus created and used as characteristics. To reduce the little stand-off fluctuations, every repetition point is normalised to its highest value. On the basis of the PCAs principle component analysis (PCA) characteristics, the rate of the point vector is categorised. Prior to being utilised as data sources for SVM and ANN, PCA is used to identify the important characteristics. PCA breaks down the starting point into a small number of uncorrelated values using a set of diagonal transformations.

    Figure 7 :- Neural networks of aritificial intelligence Though the approach is confined to categorising the

    examined specimen as either defective or without defects without doing a fault evaluation, intelligent inspection nonetheless manages to accomplish accurate flaw identification even when no complex noise preprocessing is taken into account. Additionally, when there are little stand-off fluctuations, the features normalisation greatly enhances the fault identification. Because it is not restricted to categorising the examined material into defects and fault-free groups, an AI model may contribute more t MNDT. Another goal in defect

    analysis that intelligent MNDT may accomplish is flaws imaging. The microwave waves' ability to interact with the interior layers of a complex structure effectively accounts for the high classification rate. Additionally, [48]s novel fusion of microwave and AI approaches examination of the fault in terms of its location and extent. Despite the minimal efforts made to include AI approaches into NDT, the microwave performed well. Therefore, further research is needed in the future to fully integrate AI and MNDT and also to reduce the amount of data collected during the first stage of microwave inspection.


    Noise in signal quality and signal intricacy can be resolved using comfort computing techniques like processing signals and artificial intelligence-based learning algorithms. Therefore, the constraints may be overcome and a successful outcome for the assessment of CUI can be obtained by combining soft computation methods with microwave NDT techniques. As previously noted, microwave NDT methods used for materials analysis might result in fallback-free divergence on the borders of the defect. Because machine learning can solve complicated and indiscriminate data, it may be used to detect fault and non- defect regions. Additionally, because statistical algorithms for machine learning have a great capacity to estimate unknown values based on training set instances, they may be used to anticipate the depth of a defect using examples of known flaws. However, the effectiveness of machine learning algorithms for both classification and estimate depends on the volume of the retrieved features to distinguish error position and depth. By choosing the proper and distinctive characteristics, one may increase classifier accuracy, shorten processing time, and make the system implementable in a real setting. In this case, a thorough investigation is required to determine the impact of transferring traditional feature extraction techniques, including HOG and LBP, from the image processing to the signal processing domains.


OERW is a standard MNDT technique that is heavily utilised in non-destructive assessment and has shown excellent results in terms of erros, CFRP, GFRP, TBC, and dielectric components. However, due to the absence of application of comfort computing approaches in MNDT applications, researchers have focused their study on sensor-based improvement, which presents a number of obstacles for OERW-based solutions, including stand-off variations, the right frequency point, and poor picture quality. Rectangular waveguides outperform all other microwave NDT methods in the examination of CUI. The rectangular broadband can work independently or in tandem with other elements. Rectangular waveguides still have a problem with being able to identify the entire region of corrosion. This is due to the researchers' emphasis on strengthening the standard of sensing without the use of soft computing approaches for microwave NDT. Soft computing techniques may be used to eliminate the outliers and increase the precision of CUI detection.Additionally, the degree of severity of the CUI may be detected and visualised using AI-based machine learning approaches. Additionally, machine learning

can anticipate the CUI and calculate how long the material is expected to last. Therefore, companies can make early preparations to fix or replace the items being tested. Utilising deep learning in electromagnetic NDT techniques may also provide an automated system for improving the efficacy and management of goods throughout production and use. Artificial intelligence (AI) provides a substantial influence on many sectors, notably detection of patterns, data extraction, systems digitization, and traditional NDT procedures. It is thus extremely likely that signal processing methods and AI approaches will be combined in MNDT. The efficacy of the system of checks would enhance as a result. The application of AI in MNDT goes beyond only overcoming the aforementioned difficulties; it can also be used to construct automated systems that may enhance quality assurance as well as tracking during manufacture or in service, which improves NDT applications qualitatively. Additionally, MNDT automation attempts to avoid depending on the knowledge and expertise of the operators, reducing the chance of human mistake.


[1] Q. Cao, M. Brameld, N. Birbilis, and S. Thomas, On the Mitigation of Corrosion Under Insulation (CUI) of Mild Steel Using Local Cathodic Protection, Corrosion, vol. 75, no. 12, pp. 15411551, Dec. 2019, doi: 10.5006/3197.

[2] M. F. Akbar, N. H. M. M. Shrifan, G. N. Jawad, and N. A. M. Isa, Assessment of Delamination Under Insulation Using Ridge Waveguide, IEEE Access, vol. 10, pp. 3617736187, 2022, doi: 10.1109/ACCESS.2022.3163308.

[3] Z. Yuan, Y. Tu, Y. Zhao, H. Jiang, and C. Wang, Degradation behavior and aging mechanism of decay-like fractured GRP rod in composite insulator, IEEE Transactions on Dielectrics and Electrical Insulation, vol. 26, no. 3, pp. 10271034, Jun. 2019, doi: 10.1109/TDEI.2019.007788.

[4] R. Yang, Y. He, H. Zhang, and S. Huang, Through coating imaging and nondestructive visualization evaluation of early marine corrosion using electromagnetic induction thermography, Ocean Engineering, vol. 147, pp. 277288, Jan. 2018, doi: 10.1016/J.OCEANENG.2017.09.023.

[5] M. J. Cullin, G. Birmingham, R. Srinivasan, and G. Hailu, Injectable Sodium Bentonite Inhibitors for Corrosion under Insulation, J Pipeline Syst Eng Pract, vol. 11, no. 4, p. 04020036, Jun. 2020, doi: 10.1061/(ASCE)PS.1949-1204.0000488.

[6] Design for the prevention of CUI, Corrosion-Under-Insulation (CUI) Guidelines, pp. 6773, Jan. 2016, doi: 10.1016/B978-0-08-100714- 3.00008-1.

[7] N. R. A. Burhani, M. Muhammad, and M. C. Ismail, Corrosion under insulation rate prediction model for piping by two stages of artificial neural network, AIP Conf Proc, vol. 2035, no. 1, Nov. 2018, doi: 10.1063/1.5075560/792804.

[8] E. O. Eltai, F. Musharavati, and E. S. Mahdi, Severity of corrosion under insulation (CUI) to structures and strategies to detect it, Corrosion Reviews, vol. 37, no. 6, pp. 553564, Dec. 2019, doi: 10.1515/CORRREV-2018-


[9] M. Yang and J. Liu, In Situ Monitoring of Corrosion under Insulation Using Electrochemical and Mass Loss Measurements, International Journal of Corrosion, vol. 2022, 2022, doi: 10.1155/2022/6681008.

[10] R. Wu, H. Zhang, R. Yang, W. Chen, and G. Chen, Nondestructive Testing for Corrosion Evaluation of Metal under Coating, J Sens, vol. 2021, 2021, doi: 10.1155/2021/6640406.

[11] M. Jolly et al., Review of Non-destructive Testing (NDT) Techniques and their applicability to thick walled composites, Procedia CIRP, vol. 38, pp. 129136, Oct. 2015, doi: 10.1016/J.PROCIR.2015.07.043.

[12] M. Margret, M. Menaka, V. Subramanian, R. Baskaran, and B. Venkatraman, Non-destructive inspection of hidden corrosion through Compton backscattering technique, Radiation Physics and Chemistry, vol. 152, pp. 158164, Nov. 2018, doi: 10.1016/J.RADPHYSCHEM.2018.07.015.

[13] S. Doshvarpassand, C. Wu, and X. Wang, An overview of corrosion defect characterization using active infrared thermography, Infrared

Phys Technol, vol. 96, pp. 366389, Jan. 2019, doi: 10.1016/J.INFRARED.2018.12.006.

[14] D. S. Herd, Microwave based monitoring system for corrosion under insulation, 2016, Accessed: Apr. 26, 2023. [Online]. Available: https://www.ros.hw.ac.uk/handle/10399/3240

[15] N. H. M. M. Shrifan, M. F. Akbar, and N. A. M. Isa, Prospect of using artificial intelligence for microwave nondestructive testing technique: A review, IEEE Access, vol. 7, pp. 110628110650, 2019, doi: 10.1109/ACCESS.2019.2934143.

[16] A. N. AbdAlla, M. A. Faraj, F. Samsuri, D. Rifai, K. Ali, and Y. Al- Douri, Challenges in improving the performance of eddy current testing: Review, Measureent and Control (United Kingdom), vol. 52, no. 12, pp. 4664, Jan. 2019, doi: 10.1177/0020294018801382/ASSET/IMAGES/LARGE/10.1177_0020


[17] L. Xie, B. Gao, G. Y. Tian, J. Tan, B. Feng, and Y. Yin, Coupling pulse eddy current sensor for deeper defects NDT, Sens Actuators A Phys, vol. 293, pp. 189199, Jul. 2019, doi: 10.1016/J.SNA.2019.03.029.

[18] R. Mardaninejad and M. S. Safizadeh, Gas Pipeline Corrosion Mapping Through Coating Using Pulsed Eddy Current Technique, Russian Journal of Nondestructive Testing, vol. 55, no. 11, pp. 858867, Nov. 2019, doi: 10.1134/S1061830919110068/METRICS.

[19] S. A. Denenberg et al., Advancements in Imaging Corrosion Under Insulation for Piping and Vessels, Mater Eval, 2015.

[20] Y. Deng and X. Liu, Electromagnetic imaging methods for nondestructive evaluation applications, Sensors (Basel), vol. 11, no. 12, pp. 1177411808, Dec. 2011, doi: 10.3390/S111211774.

[21] Z. Lu, Q. Zhang, and X. Liu, New magnetic particle cassette NDT intelligent detection device, Proceedings – 2013 4th International Conference on Intelligent Systems Design and Engineering Applications, ISDEA 2013, pp. 403406, 2013, doi: 10.1109/ISDEA.2013.496.

[22] A. Zolfaghari, A. Zolfaghari, and F. Kolahan, Reliability and sensitivity of magnetic particle nondestructive testing in detecting the surface cracks of welded components, Nondestructive Testing and Evaluation, vol. 33, no. 3, pp. 290300, Jul. 2018, doi: 10.1080/10589759.2018.1428322.

[23] A. Lopez, R. Bacelar, I. Pires, T. G. Santos, J. P. Sousa, and L. Quintino, Non-destructive testing application of radiography and ultrasound for wire and arc additive manufacturing, Addit Manuf, vol. 21, pp. 298 306, May 2018, doi: 10.1016/J.ADDMA.2018.03.020.

[24] F. Zhong, C. Zhang, W. Li, J. Jiao, and L. Zhong, Nonlinear ultrasonic characterization of intergranular corrosion damage in super 304H steel tube, Anti-Corrosion Methods and Materials, vol. 63, no. 2, pp. 145 152, Mar. 2016, doi: 10.1108/ACMM-05-2014-1390/FULL/XML.

[25] J. C. Bolomey, Recent European Developments in Active Microwave Imaging for Industrial, Scientific, and Medical Applications, IEEE Trans Microw Theory Tech, vol. 37, no. 12, pp. 21092117, 1989, doi: 10.1109/22.44129.

[26] P. Butz, C. Hofmann, and B. Tauscher, Recent Developments in Noninvasive Techniques for Fresh Fruit and Vegetable Internal Quality Analysis, J Food Sci, vol. 70, no. 9, 2006, doi: 10.1111/J.1365- 2621.2005.TB08328.X.

[27] A. Franchois and C. Pichot, Microwave imaging-complex permittivity reconstruction with a levenberg-marquardt method, IEEE Trans Antennas Propag, vol. 45, no. 2, pp. 203215, 1997, doi: 10.1109/8.560338.

[28] A. Todoroki, K. Yamada, Y. Mizutani, Y. Suzuki, and R. Matsuzaki, Impact damage detection of a carbon-fibre-reinforced-polymer plate employing self-sensing time-domain reflectometry, Compos Struct, vol. 130, pp. 174179, Oct. 2015, doi: 10.1016/J.COMPSTRUCT.2015.04.020.

[29] S. Kon, K. Watabe, and M. Horibe, Nondestructive method using transmission line for detection of foreign objects in food, 2018 IEEE Sensors Applications Symposium (SAS), vol. 2018-January, pp. 14, Apr. 2018, doi: 10.1109/SAS.2018.8336728.

[30] Z. Li, A. Haigh, C. Soutis, A. Gibson, and R. Sloan, Evaluation of water content in honey using microwave transmission line technique, J Food Eng, vol. 215, pp. 113125, Dec. 2017, doi: 10.1016/J.JFOODENG.2017.07.009.

[31] M. T. Ghasr, M. J. Horst, M. Lechuga, R. Rapoza, C. J. Renoud, and R. Zoughi, Accurate One-Sided Microwave Thickness Evaluation of Lined-Fiberglass Composites, IEEE Trans Instrum Meas, vol. 64, no. 10, pp. 28022812, Oct. 2015, doi: 10.1109/TIM.2015.2426352.

[32] G. N. Jawad and M. F. Akbar, IFFT-Based Microwave Non-Destructive Testing for Delamination Detection and Thickness Estimation, IEEE Access, vol. 9, pp. 9856198572, 2021, doi: 10.1109/ACCESS.2021.3095105.

[33] T. D. Carrigan et al., Nondestructive Testing of Nonmetallic Pipelines Using Microwave Reflectometry on an In-Line Inspection Robot, IEEE Trans Instrum Meas, vol. 68, no. 2, pp. 586594, Feb. 2019, doi: 10.1109/TIM.2018.2847780.

[34] M. Saif Ur Rahman, A. Yassin, and M. A. Abou-Khousa, Microwave imaging of thick composite structures using circular aperture probe, Meas Sci Technol, vol. 29, no. 9, p. 095403, Aug. 2018, doi: 10.1088/1361-6501/AAD2CF.

[35] Z. Li, A. Haigh, C. Soutis, A. Gibson, and P. Wang, A review of microwave testing of glass fibre-reinforced polymer composites, https://doi.org/10.1080/10589759.2019.1605603, vol. 34, no. 4, pp.

429458, Oct. 2019, doi: 10.1080/10589759.2019.1605603.

[36] M. F. Akbar, G. N. Jawad, L. R. Danoon, and R. Sloan, Delamination Detection in Glass-Fibre Reinforced Polymer (GFRP) Using Microwave Time Domain Reflectometry, 2018 15th European Radar Conference, EuRAD 2018, pp. 253256, Nov. 2018, doi: 10.23919/EURAD.2018.8546540.

[37] H. Zhang, Y. He, B. Gao, G. Y. Tian, L. Xu, and R. Wu, Evaluation of Atmospheric Corrosion on Coated Steel Using K-Band Sweep Frequency Microwave Imaging, IEEE Sens J, vol. 16, no. 9, pp. 3025 3033, May 2016, doi: 10.1109/JSEN.2016.2522983.

[38] B. Ahn, J. Kim, and B. Choi, Artificial intelligence-based machine learning considering flow and temperature of the pipeline for leak early detection using acoustic emission, Eng Fract Mech, vol. 210, pp. 381 392, Apr. 2019, doi: 10.1016/J.ENGFRACMECH.2018.03.010.

[39] Y. Diao, L. Yan, and K. Gao, Improvement of the machine learning- based corrosion rate prediction model through the optimization of input features, Mater Des, vol. 198, p. 109326, Jan. 2021, doi: 10.1016/J.MATDES.2020.109326.

[40] T. S. Yee, M. F. Akbar, N. A. Ghazali, and M. F. P. Mohamed, Defects Detection Using Complementary Split Ring Resonator with Microstrip Patch Antenna, Lecture Notes in Electrical Engineering, vol. 829 LNEE, pp. 625631, 2022, doi: 10.1007/978-981-16-8129- 5_95/COVER.

[41] N. H. M. M. Shrifan, M. F. Akbar, and N. A. Mat Isa, Maximal overlap discrete wavelet-packet transform aided microwave nondestructive testing, NDT & E International, vol. 119, p. 102414, Apr. 2021, doi: 10.1016/J.NDTEINT.2021.102414.

[42] T. S. Yee and M. F. Akbar, Under Insulation Microwave Non- destructive Testing Using Dual-Ridges Open-Ended Rectangular Waveguide, Lecture Notes in Electrical Engineering, vol. 829 LNEE, pp. 684689, 2022, doi: 10.1007/978-981-16-8129-5_104/COVER.

[43] R. R. Zebari, A. Mohsin Abdulazeez, D. Q. Zeebaree, D. A. Zebari, and

J. N. Saeed, A Comprehensive Review of Dimensionality Reduction Techniques for Feature Selection and Feature Extraction, Journal of Applied Science and Technology Trends, vol. 1, no. 2, pp. 5670, May 2020, doi: 10.38094/jastt1224.

[44] O. F.Y, A. J.E.T, A. O, H. J. O, O. O, and A. J, Supervised Machine Learning Algorithms: Classification and Comparison, International Journal of Computer Trends and Technology, vol. 48, no. 3, pp. 128 138, Jun. 2017, doi: 10.14445/22312803/IJCTT-V48P126.

[45] X. Yuan, B. Huang, Y. Wang, C. Yang, and W. Gui, Deep Learning- Based Feature Representation and Its Application for Soft Sensor Modeling with Variable-Wise Weighted SAE, IEEE Trans Industr Inform, vol. 14, no. 7, pp. 32353243, Jul. 2018, doi: 10.1109/TII.2018.2809730.

[46] J. S. Chou, N. T. Ngo, and W. K. Chong, The use of artificial intelligence combiners for modeling steel pitting risk and corrosion rate, Eng Appl Artif Intell, vol. 65, pp. 471483, Oct. 2017, doi: 10.1016/J.ENGAPPAI.2016.09.008.

[47] A. Ali, B. Hu, and O. Ramahi, Intelligent detection of cracks in metallic surfaces using a waveguide sensor loaded with metamaterial elements, Sensors (Basel), vol. 15, no. 5, pp. 1140211416, May 2015, doi: 10.3390/S150511402.

[48] A. Ali, A. Albasir, and O. M. Ramahi, Microwave sensor for imaging corrosion under coatings utilizing pattern recognition, 2016 IEEE Antennas and Propagation Sociey International Symposium, APSURSI 2016 – Proceedings, pp. 951952, Oct. 2016, doi: 10.1109/APS.2016.7696184.