1 Introduction

The proportion of carbon fibre-reinforced polymer (CFRP) composites being used in aerospace, automotive and marine structures is increasing year on year [1]. Fatigue tolerance of composite structures is found attractive in addition to lightweight and corrosion resistance. In the composites carbon fibres of high strength and Young’s modulus are embedded in a polymer resin as a matrix. Carbon fibres are usually produced from natural cellulose, synthetic polyacrylonitrile (PAN) and pitch, by carbonisation and/or graphitisation at high temperatures to eliminate other chemical elements and generate polycrystalline graphitic structures [2]. In general, fibres with a diameter of 5 to 10 μm are the primary load-carrying members, while the surrounding matrix acts as a load transfer medium and keeps the fibres in the desired location and orientation [3].

Defects like air voids and surface unevenness may be generated during the manufacturing process. And in the field carbon fibre composites remain vulnerable to various hazards, like impact caused by hailstones, runway debris, collision with ground equipment, tool drops and bird strikes. The types of the composite damage induced by impact include surface dents, delamination, matrix cracking and fibre breakage. And in many occasions, these happen internally and are hardly visible. Hence, it is of importance to routinely and accurately assess the conditions of composite structures with an effective inspection approach. In the aircraft industry visual inspection is commonly conducted before each flight, while the accuracy depends on the lighting conditions, access for detection, inspector’s experience and time relaxation (e.g., dent depth of the impact damage). In addition to that conventional method, various non-destructive testing (NDT) techniques have been applied, such as ultrasonic testing, acoustic emission, thermography, shearography, vibration testing, optical fibre sensing, Lamb waves and digital image correlation (DIC). At present, no single method exists that can detect all types of manufacturing defects and in-service damage [4]. Each method has its own specific advantages and limitations. For example, couplants (e.g., water or gel) are needed in ultrasonic testing, and the acoustic emission sensors should be placed near the damage source for accurate measurement [5]. For thermography, the possibility of unwanted thermal damage to the structures should be taken into account caused during the test. In the setups of shearography and vibration testing, it is required to apply mechanical loads to the structure. In optical fibre sensing systems, the weight penalty, possibility of failures in the wiring network, manufacturing and installation costs are significant. For Lamb wave-based techniques, piezoelectric transducers are permanently mounted on the surface of the test piece [6]. And before the DIC measurement, the surface of the sample under test should be speckled with black and white paints [7], which is not practical for large structures.

An alternative method for damage detection is the microwave technique. Microwaves are electromagnetic (EM) radiation with frequencies between 300 MHz (wavelength of 1 m) and 300 GHz (wavelength of 1 mm). Microwaves can propagate in air and dielectric materials with low attenuation. The amplitude and phase of the microwave signals can be affected by the variations of the thickness or electromagnetic properties (i.e., permittivity and permeability) of the material due to damage/defects. There are a number of attributes when applying microwave NDT, such as non-contact, no need for transducers bonded on the surface or couplants, no need for complicated signal post-processing, operator friendly, relatively inexpensive and one-sided scanning [8,9,10]. During in-service inspections, commonly only one side of the object under test can be accessed, so the microwave NDT methods are more advantageous. Safety precautions are usually not needed since the power of the signal used is relatively low (few milliwatts). It is noted that the microwave techniques should be differentiated from the lower frequency electromagnetic detection methods, like the eddy current testing (ECT) (100 Hz - 10 MHz) [11,12,13] and coupled spiral inductors (CSI) (1–100 MHz) [14,15,16], where the field that exists only around the coil cannot propagate. In addition, the ECT and CSI do not work well for dielectric materials.

In recognition of the growing interest in microwave testing, in 2011 the Expert committee for microwave and THz testing procedures of the German Society of Non-Destructive Testing (DGZfP) was founded. And in 2014 the Microwave Testing Committee of the American Society for Non-destructive Testing (ASNT) was established. Microwave testing was furthermore recognised as its own NDT method in the 2016 edition of the ASNT standards, and the certifications for Level I and Level II microwave testing inspectors are currently under development [9].

In this paper, the electromagnetic properties of carbon fibre composites closely related to the microwave-based detection are addressed first. Four geometric scales are adopted in the permittivity analysis to thoroughly study the differences in electromagnetic responses due to the fibre direction. Based on the detection principles, the microwave testing methods can be classified into five categories: self-sensing methods, near-field induction methods, near-field resonance methods, far-field methods and the methods with integration of another NDT. The mechanism and applications of each kind are discussed in detail. Finally, some viewpoints on the development of microwave detection research are presented.

2 Electromagnetic Properties of Carbon Fibre Composites

It is perquisite to understand the electromagnetic properties of each component first. Carbon fibres are both thermally and electrically conductive. The electrical conductivity of carbon fibres ranges from 3.84 × 104 S/m to 1 × 106 S/m depending on graphitisation treatment [17]. The high conductivity makes them behave like metals, which reflect most energy of the incident signal. However, epoxy resin is nonconductive, i.e., dielectric. Its dielectric constant is usually less than 6 at room temperature [18]. And the loss factor is negligible, which means that little energy is absorbed in the material and it is considered transparent to microwave radiation.

For the mixture the effective permittivity is a parameter of interest for evaluation of the overall electromagnetic performance. As CFRP is non-magnetic, its permeability μ is equal to that of free space μ0. The permittivity ε can be written as [19].

$$ \varepsilon ={\varepsilon}_0{\varepsilon}_r={\varepsilon}_0\left({\varepsilon}_r^{\prime }-j{\varepsilon}_r^{{\prime\prime}}\right) $$
(1)

where ε0 is the permittivity of free space (i.e., 8.8542 × 10−12 F·m−1), and εr is the relative permittivity. The real part \( {\varepsilon}_r^{\prime } \), or dielectric constant, is related to the ability of a material to store the electric field energy, while \( {\varepsilon}_r^{{\prime\prime} } \) accounts for the dissipation of energy within the material in the form of heat. \( {\varepsilon}_r^{{\prime\prime} } \) is positive due to energy conservation. The transmission line method, open-ended coaxial/waveguide probe method, resonant cavity method and free space measurement can be used for permittivity characterisation [20, 21].

The permittivity of the composite is anisotropic, as the fibres and fibre architecture affect the polarisation of the induced currents. In [22], it was reported that at 10 GHz, the relative permittivity of a T300 composite laminate was approximately 101-j73 when the fibres were parallel to the incident electric field, while it was 107-j31 when the fibres and the electric field were orthogonal. The relative permittivity of a neat resin sample cured under the same conditions as the composite was 2.90-j0.10. The differences in the permittivity are associated with the multiscale nature of the composite laminate. As illustrated in Fig. 1, there are four geometric scales: macro level, meso level, micro level and molecular level. On the macro/meso-scale, a number of thin layers are stacked with a desired sequence of fibre orientations. On the microscale, in a lamina, ideally the fibres do not touch and are completely isolated by the resin. However, the fibres, especially in 2D and 3D woven architectures, may contact at several points. At the molecular level, the electrons in carbon fibres (polycrystalline graphite) are delocalised, i.e., free to move along the planes of carbon atoms.

Fig. 1
figure 1

Four geometric scales involved in the electromagnetic study of a composite laminate

Full size image

On the microscale, due to the presence of the applied EM field, carbon fibres and dielectric epoxy make up a series of capacitors. The capacitance always exists when the incident field is applied at an arbitrary angle. Hence a similar energy storage capability is reflected in the real permittivity. And that permittivity is two orders of magnitude larger than that of the neat resin as would be expected. \( {\varepsilon}_{\mathrm{r}}^{{\prime\prime} } \) in the orthogonal case is approximately half of that in the parallel case. As the epoxy resin is a polar dielectric material [23], only the dipolar relaxation exists. The lower \( {\varepsilon}_{\mathrm{r}}^{{\prime\prime} } \) of the epoxy indicates that the epoxy itself does not contribute much to the high loss of the composite, so the primary loss is due to the carbon fibres. At the molecular level, when the electric field is parallel to the carbon fibres, the electrons move freely along the whole length of the fibres. In the orthogonal case, some electrons travel via the contact points between fibres, while the others are trapped at the interfaces between the fibres and resin (Maxwell-Wagner (interfacial) polarisation [24]), which leads to a relatively lower conductivity.

The signal penetration is a practical factor that should be paid attention to. Due to the lossy medium, the power of the EM signal decays exponentially through the thickness. A parameter used for the evaluation is the penetration depth dp, which is defined as the depth where the magnitude is reduced to 1/e (about 37%) in the medium. For a plane wave incident on a half space of a medium, dp can be given by:

$$ {d}_p=\frac{c}{\sqrt{2}\pi f{\left\{{\varepsilon}_r^{\prime}\left[\sqrt{1+{\left(\frac{\varepsilon_r^{{\prime\prime} }}{\varepsilon_r^{\prime }}\right)}^2}-1\right]\right\}}^{1/2}} $$
(2)

where f is the operating frequency, and c is the speed of light in free space. For the same laminate discussed above, at 10 GHz the depths dp in the parallel and orthogonal cases were 1.39 mm (approximately 11 layers assuming the thickness of a single layer is 0.125 mm) and 3.25 mm (approximately 26 layers), respectively.

It is seen that there is an inverse relationship between the signal frequency and the penetration depth. Hence, it is feasible that the distribution of the damage in the thickness direction can be revealed using multi-frequency inspection. For deeper penetration, a lower frequency and an electric field not parallel to the fibre direction are preferred.

3 Existing Microwave Testing Methods for CFRP Composites

3.1 Self-Sensing Methods

In this type of methods, CFRP composite acts as a conductive component of a microwave circuit. Any damage in the composite will lead to perturbation in the system. By detecting the changes in the signal response, self-sensing of the composite structure is enabled without any external setup.

  1. (a)

    Part of a transmission line

The composite can be ground plane of a transmission line. Todoroki et al. [25,26,27] constructed a microstrip line with the use of a copper tape on a glass fibre reinforced polymer (GFRP) plate as substrate, Fig. 2. When there is surface damage or near-surface damage, the characteristic impedance at that location is changed, and the incident signal is split into the reflection and transmission components. This discontinuity can be located from the reflected signal in the time domain. This measurement technique is known as Time Domain Reflectometry (TDR). It was reported that delamination, matrix cracks, fibre breakage and lightning strike damage can be successfully detected.

Fig. 2
figure 2

Diagram of the cross section of the self-sensing sensor proposed by Todoroki [21,22,23]

Full size image

In their setup shown in Fig. 2, a GFRP plate was cut with the same dimensions as those of the CFRP plate and glued together, while the copper tape and CFRP were soldered to a coaxial cable at the end of the transmission line. It is not suitable for practical applications, since the microstrip line is bonded to the sample under inspection. Li et al. [28] used an independent microstrip line on a Printed Circuit Board (PCB) for detection with the same TDR concept. As shown in Fig. 3, the strip conductor side of the board is placed close to the surface of the sample. Some fraction of the electromagnetic field is distributed between the strip conductor and the CFRP, which forms a new transmission line (similar to a covered microstrip line [29]). Here CFRP is separated from the measuring system, and the size of the sensor can be smaller than that of the CFRP sample. The whole area can be examined by conducting a line scan. In the test, a microstrip line was made on a FR4 substrate, and the distance between the board and an impacted sample (usually called standoff distance) was 100 μm. The microwave signal was generated with a vector network analyser (VNA). The frequency-domain data obtained were converted into the time domain by Inverse Fast Fourier Transform (IFFT). It was revealed that the location of the peak in the curve agreed with the real damage location.

  1. (b)

    Part of an antenna

Fig. 3
figure 3

Diagram of the detection system with a microstrip line sensor a experimental setup b cross section

Full size image

Matsuzaki et al. [30] treated CFRP as an element of a half-wavelength dipole antenna (e.g., CFRP wings in the electrically insulated case shown in Fig. 4) or a monopole antenna. The feasibility of this method was studied using unidirectional laminates and rotor blades of woven CFRP. It was demonstrated that the resonance frequency of the return loss was increased due to the damage in the specimen. Damage like fibre breakage or delamination caused the effective length of the dipole antenna shortened. However, the number of the damaged component (one or two blades) and damage location cannot be readily determined.

Fig. 4
figure 4

Insulated CFRP rotor blades modelled as a dipole antenna (adapted from [30])

Full size image

3.2 Near-Field Induction Methods

The near-field induction methods are primarily based on Faraday’s principle of electromagnetic induction. In the present case with CFRP composites it works the same as the eddy current technique, though microwaves can also be used for dielectric materials with the same setup [31]. The signal can be radiated from an antenna or an open-ended circular/rectangular waveguide performing like an antenna. To be accurate, the term “near field” refers to the non-radiative near field region, or reactive near field, which exists in the immediate vicinity of the antenna where mature propagating waves have not yet formed. The current distributions on the antenna are affected when an object is placed in the reactive region, the boundary of which is commonly given by [32]:

$$ R<0.62\sqrt{\frac{D^3}{\lambda }} $$
(3)

where D is the maximum linear dimension of the antenna, and λ is the wavelength. It is indicated that Eq. (2) is not suitable for evaluation of penetration in the near-field case, as the assumption of uniform plane waves is not valid. Here the penetration is associated with the standoff distance, operating frequency, dimensions of the object under test, electromagnetic properties of the materials and the boundary conditions. Rigorous theoretical analysis of a rectangular waveguide radiating into a laminated composite can be found in [33].

The schematic diagrams of the experimental setup and the equivalent lumped circuit model are illustrated in Fig. 5. When the CFRP sample is illuminated by the microwave radiation, currents are induced and predominated along the direction of high conductivity (i.e., carbon fibre direction), and the currents flow from one fibre to another at contact points. The secondary field produced can cause some energy reflected to the source, which is evaluated by the reflection coefficient S11 in the form of magnitude and phase. In the lumped circuit, the capacitance Cg is linked to the standoff distance, so surface unevenness can affect the value of this parameter. The resistance Rs and inductance Ls are associated with the material properties and sample dimensions (for the relatively high conductivity of the composite, the effect of its capacitance is not considered in the qualitative analysis). Change of the electric field direction with respect to the fibres will also lead to changes in the resistance and inductance and subsequently the signal received.

  1. (a)

    Detection of fibre direction

Fig. 5
figure 5

Illustration of the near-field microwave induction method a experimental setup b equivalent circuit model

Full size image

The basic application of this technique is determination of the fibre direction. When the fibres are not parallel to the electric field, less energy is reflected to the source, thus lower S11. Galehdar et al. [34] studied the effect of the orientation of the surface ply on the reflectivity. The carbon fibres that were parallel to the electric field behaved as good conductors, while the off-axis ply behaved as lossy dielectric layers with a finite conductivity. Kharkovsky et al. [35, 36] employed a waveguide to determine the fibre orientation in a CFRP patch with a thickness of 0.1 mm (one ply) and detect fibre breakage. It was implied that the electric field in the waveguide parallel with the fibres was optimal for fibre breakage detection. A small breakage with dimensions of 0.2 mm by 1 mm was found. The fibre orientation was determined by varying the angle between the fibre direction and signal polarisation. Wilson et al. [37] investigated the detailed relationship between the angular orientation of a carbon fibre tow and the S11 responses using a broadband horn antenna. The dependence on the tow orientation was strongly sinusoidal, and an average sensitivity of 97 kHz/degree resolution was obtained. A small angular misalignment of 1° was identified [38]. This technique could be used for automated non-contact inspection during the manufacturing process when the carbon fibre tow is being laid down by a fabrication robot.

  1. (b)

    Strain sensing

The engineering strain is the ratio of the total deformation to the initial dimension of the material along the force being applied, and it is an important indicator for stress concentration and crack growth. It is known that the resistance and inductance of a wire are linked to the length, and the deformation of the plate leads to changes in the capacitance associated with the air gap. Hence, by using the microwave near-field induction approach, as seen in Fig. 5b, any change of the fibre length due to the applied force could be observed in the received signal. Wilson et al. [39] first introduced the non-contact microwave method for strain measurement. In the test, the antenna was placed 12 cm from a quasi-isotropic CFRP laminate. From the retrieved signal, the reactance (i.e., imaginary part of the complex impedance) was calculated and proved most sensitive to the strain change. A close linear relationship between the reactance and the strain was established. However, the aperture of the antenna used was relatively large (approximately 24 cm by 14 cm), which suggested that the spatial resolution of the sensing was limited. Optimisation of the test setup, signal post-processing and characterisation of the temperature and humidity effects are aspects of further improvement.

  1. (c)

    Damage detection

The near-field induction technique has been widely used for damage detection of carbon fibre composites. The information of the delamination and surface damage can be provided. It is noted that there is a compromise between the signal penetration and the spatial resolution, as the resolution is highly dependent on the aperture size. For a rectangular waveguide, the higher the operating frequency range, the smaller the aperture size.

Akuthota et al. [40] applied an open-ended waveguide to detect the disbonds between a two-layer CFRP composite laminate and a concrete substrate. They pointed out that the frequency and the standoff distance should be chosen optimally (or near optimally) to maximise the sensitivity. It was found that the standoff distance variation, in the range of a few millimetres, had no adverse effect at 10 GHz, and a minimal effect at 24 GHz. The smallest detectable disbonded regions at 10 GHz and 24 GHz were 20 mm by 5 mm.

From theoretical simulations, Bin Sediq et al. [41] revealed that both rectangular and circular waveguides could be capable of detecting defects inside carbon-loaded composites. High attenuation was inevitable for the rectangular waveguide, due to the relatively linear polarisation of the fields radiated out of the waveguide and the linear nature of carbon fibres. It was suggested that the circular waveguide with the characteristic of circular polarisation was more attractive for inspection.

Yang et al. [42] detected the impact damage in CFRP specimens by using a horn antenna, which is shown in Fig. 6a. In the test, a frequency range of 65–67 GHz was used, and the E-field was applied parallel to the carbon fibres. The damage caused by impact energies of 13.21 J and 8.89 J was reported to be easily distinguished from the image produced. With the use of an edge detection image processing technique, a more natural image was produced and the shapes of the damage were clearly identified.

Fig. 6
figure 6

Microwave detection of impact damage in a CFRP specimen with a V band (50–75 GHz) horn antenna [42] a setup b imaging result of 8.89 J impact energy (unit: cm)

Full size image

3.3 Near-field resonance methods

  1. (a)

    Near-field scanning microwave microscopy (NSMM)

As shown in Fig. 7, a near-field microwave microscope is generally composed of a resonant cavity and a sharp tip. The sample is placed close to the probing tip, from which evanescent microwaves are emitted. The probe scans across the surface of the object at a fixed standoff distance. The unevenness or material discontinuity of the surface can result in a different capacitance of the air gap, subsequently changes in the resonance frequency and quality factor (or Q factor) of the whole resonant circuit. Thus, a surface contour plot of the resonance frequency/Q factor can be produced. The sharper the tip is, the higher the spatial resolution that would be obtained. The penetration offered by this method is poor for the low intensity of the evanescent field.

Fig. 7
figure 7

A general near-field microwave microscope for detection of surface damage a schematic diagram b lumped circuit model

Full size image

Li et al. [43] utilised a near-field microwave profiler to detect low velocity impact damage in a 4-mm thick composite laminate. The damage categorised as barely visible impact damage (BVID) was created by a drop-weight impact energy of 20 J. In the test, the resonance frequency at each measurement position was recorded from a Marconi 6200A scalar network analyser. The centre frequency was set to 3.05 GHz with a frequency span of 100 MHz. A 2D scan was performed over the sample with a standoff distance of 100 μm and a step size of 280 μm. As shown in Fig. 8, the area of the dent is clearly defined in the image generated, and the symmetric and circular damage shape shows better image quality than that by the microwave imaging with an open-ended waveguide.

  1. (b)

    Microwave planar resonator

Fig. 8
figure 8

Scanned image of impact damage in CFRP by near-field scanning microwave microscopy

Full size image

Instead of a resonant cavity, a microwave planar resonator was developed in [44]. As illustrated in Fig. 9, a complementary split-ring resonator (CSRR) was made on the lower side of a FR4 substrate, and a microstrip line with a width of 2.8 mm was made on the upper side for signal feed. Compared with the NSMM, this sensor exhibits some distinct advantages, such as low cost, easy operation and simple design. From the simulation analysis, it was shown that the most sensitive region of the sensor was immediately under the resonator. In the test, the sensor was connected to an HP 3720D VNA by two coaxial cables and placed above a CFRP sample with a standoff distance of 100 μm. First, the location without impact damage underneath was measured as a reference; then, the region with BVID was tested for comparison. The resonance frequency was shifted upwards from 2.23 GHz to 2.295 GHz with a frequency shift of 65 MHz.

Fig. 9
figure 9

Geometry of a microwave planar resonator for impact damage detection [44]

Full size image

3.4 Far-Field Sensing Methods

The far field is the region of operation for most antennae, in which the radiation does not change with distance. The far-field (Fraunhofer) distance must satisfy three conditions:

$$ R>\frac{2{D}^2}{\lambda } $$
(4a)
$$ R\gg D $$
(4b)
$$ R\gg \lambda $$
(4c)

It is mentioned that there is a radiative near field between the reactive near field and the far field. The energy in the radiative near field is all radiant energy, but its mixture of magnetic and electric components is still different from the far field. Depending on the values of R and the wavelength, this field may or may not exist. Hence the radiative near-field region is not in the scope of this paper.

Absorption of the radiation in the far-field region does not feed back to the transmitting antenna. Either bistatic configuration (a pair of transmitting and receiving antennae) or monostatic configuration (a single antenna with a circulator) can be used. Each kind has inherent advantages and disadvantages. The bistatic configuration has a high isolation between transmitting and receiving channels, but the dimensions are larger compared with the monostatic counterpart and accurate alignment between the transmitter and receiver is required. The monostatic configuration is compact, as the microwave circulator can separate the transmitting signal from the receiving one in a single component. However, the lower isolation of the circulator can lead to a significant leakage [45]. Additional measures must be undertaken to reduce or suppress undesirable leakage.

Gubinelli et al. [46] used two ultra-wideband (UWB) antennae to detect the presence of a hole in CFRP. The experimental setup is presented in Fig. 10. UWB pulses were generated with a frequency range of 6.0–8.5 GHz, and the radar-to-sample distance was set to 40 cm. A CFRP sheet with no defect, one with a 3 mm through-thickness hole in the centre and another one with a 3-mm hole patched in the back with the same CFRP were tested separately. The radar was in the central position with respect to the sheet. By scanning a healthy sheet (reference) and “damaged” sheets, the difference in the signal responses was recognisable.

Fig. 10
figure 10

Setup of an UWB radar system for far-field detection of a hole in a CFRP specimen [46]

Full size image

3.5 Integration with Other NDT

The microwave techniques can be combined with other NDT methods for better detection performance. Contact-free microwave-based thermography is commonly found in the literature, in which microwaves are used as a heat source (like microwave curing [47]). When a high-power microwave signal is applied, the intensity of the induced currents in composites will increase and the temperature will rise due to the Joule effect. Damage and defects could be revealed in the thermogram with an infrared (IR) camera. Owing to good performance by commercial infrared cameras, high sensitivity and high spatial resolution (much better than the microwave wavelength) can be offered. In addition, rapid inspection can be achieved, as a large area can be inspected within a short time. Compared with the available thermal excitation methods, such as thermal lamps, laser, ultrasonic and pulsed eddy current (PEC), microwaves can provide more uniform, volumetric and selective heating [48,49,50]. If the penetration depth of the microwave signal is much smaller than the sample thickness, only the material within the penetration depth is heated and the rest of the material is heated by thermal conduction. Therefore, the penetration ability of the microwave-based thermography is better than that of conventional microwave NDT techniques.

When using this specific technique, special attention should be paid to the electromagnetic shielding as high power is used. According to the guideline issued by the International Commission on Non-Ionizing Radiation Protection (ICNIRP) [51], the reference level for occupational exposure to electromagnetic fields is 50 W/m2 over 2–300 GHz, which is equivalent to an electric field strength of approximately 137 V/m in free space. This health and safety requirement should be carefully considered in the implementation.

A schematic diagram of the microwave thermography approach is illustrated in Fig. 11. A horn antenna is used to direct the waves into the region of interest. A microwave source can be a magnetron, travelling wave tube or signal generator [52]. An off-the-shelf IR camera is located on the same side of the excitation source (reflection configuration). A personal computer (PC) is used for control of the measuring instruments, data acquisition and signal post-processing.

Fig. 11
figure 11

Schematic diagram of the experimental setup for microwave-based thermography

Full size image

Keo et al. [53] used a commercial 2.45 GHz magnetron and an IR camera to detect a defect (absence of adhesive) in a CFRP plate. A 360 W microwave signal was applied to heat the sample for 150 s. It was found that the defect area was hotter than the non-defect area. Foudazi et al. [54] utilised a horn antenna to heat a cured mortar sample containing a 2-mm thick surface-bonded CFRP, as shown in Fig. 12. The surface was under microwave illumination of 50 W at 2.4 GHz for 5 s. It was shown that the simulated delamination between the mortar and the CFRP was visible in the thermal image.

Fig. 12
figure 12

Microwave-based thermography for delamination detection [54] a setup with a standoff distance of 6 cm b thermogram of the cured mortar sample

Full size image

In the work of Palumbo [55], sandwich specimens were put into a microwave chamber (power 750 W, frequency 2.45 GHz) for 2 s. The heating phase and subsequent cooling phase were both used for impact damage evaluation. Good agreement was achieved in comparison with X-ray imaging and lock-in thermography using two halogen lamps with 500 W power.

4 Concluding Remarks and Future Work

An overview of the progress made in the development of the microwave testing methods for carbon fibre-reinforced polymer composites have been presented here. The available microwave NDT methods have been summarised into five categories. Various applications have been demonstrated: strain sensing and detection of absence of adhesive, fibre direction, fibre breakage, delamination, impact damage and lightning strike damage. The advantages and limitations of the five categories of methods have been discussed for the readers’ benefit. It is noted that the potential of microwaves for wider detection applications is yet to be fully exploited. Near future work could focus on the following aspects:

  1. (1)

    Theoretical analysis: the existing work is mainly based on experiments. It is also of importance to understand the wave propagation in the materials and interaction between the wave field and the material system, which would help to optimise the setup for detection and quantify the damage size. A multi-physics approach is needed when investigating the microwave-based thermography, where electromagnetism and thermodynamics are involved.

  2. (2)

    Damage identification: various types of defects/damage can be observed during fabrication or in service, while there is a lack of research on differentiating the damage. Artificial intelligence (AI) can be introduced for automated identification of damage types. For example, an artificial neural network (ANN) can be used to build a classification system. The ANN can be modelled with input nodes that are matched to the input data format, output nodes in the form of a damage type probability and intermediate hidden layer nodes. The connection parameters between the nodes are initially defined randomly, but can be optimised by training, using a set of tests with and without damage.

  3. (3)

    Far-field imaging: in the far-field region, microwave imaging methods can be adopted to generate two-dimensional images of objects, such as synthetic aperture radar (SAR) imaging, phased array (digital beam forming) and microwave holography. The measurements can be performed with a single antenna raster scanning over the area, a 1D antenna array sweeping over the area or a 2D antenna array covering the area.

  4. (4)

    Use of the RFID technique: the active radio frequency identification (RFID) technology with the capability of damage inspection has started to attract some attention [56]. With a RFID tag attached on the surface of the object under test, the local damage information can be wirelessly interrogated with a RFID reader. Hence, the microwave RFID concept can be introduced, and further Internet of Things (IoT) system can be integrated for remote inspection and long-term structural health monitoring (SHM).

  5. (5)

    Wider applications of microwave-based thermography: the current applications of microwave-based thermography are mainly limited to detection of absence of adhesive and delamination. The potential of the technique for rapid detection of other damage types could be explored.

  6. (6)

    Automated inspection: there is a high demand for automated inspection by the industry, while many of the existing applications are still at the experimental stages. Future research could be focused on facilitating the industrial adoption.

  7. (7)

    Potential applications in additive manufacturing (AM): additive manufacturing is an emerging technique where successive layers of material are formed under computer control to create a three-dimensional object. It significantly revolutionises the design of products that feature complex geometries and concept of spare parts management. However, at present the complex parts are posing difficulties for inspection, and a standard verification procedure for quality control is not established [57]. For this reason, the 3D printing has not been widely used in industry. It is required that any defect should be identified and corrected before finishing parts. For example, when fabricating continuous carbon fibre reinforced polymer composites, porosity of each layer should be checked. Resin micro cracks [58] that could lead to delaminations [59] and fibre instability [60], especially under compressive loading, need also to be identified. Microwave testing can be applied for in-process monitoring of additive manufacturing. In addition, a customised microwave sensor can be made possible by AM to fit complex (e.g., curved) surfaces.