A novel dielectric sensor for process monitoring of carbon fibre composites manufacture

A dielectric sensor appropriate for process monitoring of composites manufacture with carbon reinforcement has been developed in this study. The sensor concept overcomes problems of electrical sorting and interference with the electric field occurring when electrical/dielectric sensors are used with carbon reinforcement. The sensor comprises two uniformly twisted insulated copper wires. Two sensor designs based on the same concept have been implemented; a lineal sensor for flow front position tracking and a woven sensor used to monitor the cure. Resin Transfer Moulding (RTM) processing has been employed to evaluate the lineal sensor performance against visual monitoring of the flow front. Vacuum Assisted RTM (VARTM) has been carried out to validate the results of the woven cure sensor against calorimetric cure kinetics models. Both the lineal flow and woven cure sensors provide accurate monitoring signals.


Introduction
The development in recent years of process monitoring techniques appropriate for continuous fibre composite manufacture has been motivated by the need to track critical manufacturing parameters such as flow front position and cure reaction progress. Lineal and spot sensors have been developed for monitoring the filing in liquid moulding processes. Lineal sensors based on dielectrics [1,2], fibre optics [3], time domain reflectometry [4] and DC [5,6] methods can perform continuous monitoring of the flow front position mainly in the presence of glass fibre reinforcement. Local sensors, such as thermocouples [7], fibre optic sensors [8,9] and pressure transducers [10][11][12], implemented in the filling stage can instantly provide information of resin arrival, whilst the technology applied is in a mature stage for integration in industrial applications.
Cure monitoring systems are based on tracking the evolution of a physical quantity connected either directly or indirectly to the cure state. Near-infrared [13][14][15] and infrared spectroscopy [16,17] provide direct chemical information on the cure evolution. The correlation between fluorescence and viscosity has been used as the basis for monitoring of the cure [18][19][20]. Similarly, the sensitivity of ultrasonic wave propagation to macroscopic polymer structure, viscosity and modulus has been used to capture the evolution of the curing reaction [21,22]. The refractive index is directly related to the density of the resin and can be used for cure monitoring through fibre optic sensors based on the densification occurring during cross linking [9,23]. Dielectric cure monitoring methods are based on the dependence of the electric and dielectric properties on structural properties of the resin and have been used successfully to follow the degree of cure, viscosity and vitrification [24][25][26][27][28][29][30][31]. Monitoring based on impedance/dielectric spectroscopy is considered advantageous due to the high sensitivity of sensor response, the robustness and low cost of the measurement setup and the capability for incorporation on tooling.
High specification parts, which are the primary target of monitoring methodologies, usually involve carbon fibres. The conductive nature of carbon fibres introduces measurement issues in cases where the sensing system operation is based on measuring the electromagnetic or optical response of the resin system. Lineal dielectric flow sensors and cure dielectric sensors are appropriate for use with non-conductive reinforcement [1] as presence of carbon disturbs the electric field, whilst using the conductive reinforcement as one of the electrodes of the sensing system involves significant practical complexity as it requires electrical insulation of the reinforcement from the tooling assembly [2]. The solution adopted for carbon composites in cure applications is to cover the sensor with a permeable nonconductive material such as glass cloth or a polymer weave [24,29,30]. This type of solution increases the intrusiveness of the sensing system and causes a difference between the material monitored and the material of the composite. Fibre optics techniques for flow monitoring rely on the evanescent field around the optical fibre [8,9], which is also disturbed by absorption by the carbon. Vacuum sensors [32], pressure sensors [12], ultrasonic [33] or thermal probes [7] are not affected by the presence of carbon fibre, and thus can be utilised for this purpose. The implementation of pressure transducers within the mould provides local information regarding the resin arrival [11], whilst 2-D pressure sensors allow the mapping of pressure distribution within the mould cavity area and thus the resin filling pattern. However, the accuracy of the measured flow front depends on the sensitivity of the monitoring system [34,35].
Furthermore, the implementation of 2-D pressure sensors requires significant adjustments and modifications of the mould cavity, increasing the monitoring system complexity. The utilisation of thermocouples placed in discrete positions within the part for monitoring of resin arrival does not depend on fabric type, but it is limited only to cases in which there are high temperature differences between resin and the mould [7,36]. Furthermore, the integration of thermocouples within the preform induces local disturbances in preform architecture. The local character of these sensors makes them not appropriate to monitor continuously the flow front, whilst the implementation of multi spot sensor arrays at sufficient resolution within the tooling assembly is cumbersome in practice and increases the intrusiveness of the monitoring system.
The present study focuses on the development of a dielectric sensor capable of operating with the presence of carbon reinforcement. The main concept is used for the design of two sensor types; a lineal sensor used to monitor the flow front and a woven arrangement able to track the cure. Sensor signals are analysed using the corresponding equivalent circuits.
The lineal sensor is validated in RTM processing against visual monitoring. The cure sensor is evaluated in isothermal neat resin cure experiments and in VARTM processing of carbon fibre composites.

Principle of operation
The design of the dielectric sensor is illustrated in Figure 1a. It comprises two twisted solid copper wires coated by an insulator. The insulating coating prevents contact of the copper wire with the conductive fibres. An electric field is formed between the wires upon application of voltage. The field goes through the insulating coating and penetrates the gaps between wires. The twisting enhances the robustness and sensitivity of the sensor. The signal strength is proportional to the actual length of wires and of the corresponding contact region (L ) which depends on the wire diameter (D) and twist per unit length (T) as follows [37]: where L denotes the length of the sensor. Increasing the twist per unit length results in greater actual length, which leads to higher sensitivity.
The twisting makes the pair more robust and eliminates the need of additional bonding needed to hold the wires in places. In comparison to a pair of parallel wires, the twisted configuration is also advantageous in terms of interference of carbon reinforcement with the field. The gaps formed between parallel wires are exposed to carbon reinforcement reducing the signal strength. In the case of a twisted pair the reinforcement interferes with the fringing field; however, the constantly changing orientation of the gap between the wires keeps limits the penetration of reinforcement in this region. The sensor principle can be adapted to address both flow and cure monitoring applications.
A lineal sensor made using the twisted pair can be used to monitor the resin flow front position. An example of the configuration of the lineal sensor in an RTM tool is presented in only causes a small disturbance to the fibre architecture, similar in size to the diameter of the wires. If the wire diameter is significantly smaller than that of the tow width the disturbance is negligible and similar to the resin rich channels that are present in the composite material.
In the case of cure monitoring, the local material state is estimated based on the sensitivity of electrical properties on cure progress. For the sensing concept based on the twisted pair proposed here, the strength of the signal is governed by the total length of the sensor. Utilising a woven configuration allows fitting a large length within a small area. This results in sufficient sensitivity combined with local cure sensing.

Analysis of lineal flow sensor signal
The lineal sensor response has been examined following the analysis in [1]. The electric circuit representing the sensor response is illustrated in Figure 2. The wetted and dry part of the sensor are connected in parallel, whilst within each part the element corresponding to the gaps and the element corresponding to the coating are connected in series. The admittance per unit length of the wetted part is: and of the dry part y = where y is the admittance per unit length of the coating, y the admittance per unit length of the filled gaps and y the admittance per unit length of the dry gaps.
The admittance measured by the sensor is: where the length of the sensor is L and the length covered by resin L . Eq (4) considering Eq (2) and (3) can be rearranged as follows: where Y = Ly is the admittance of the fully wetted sensor and Y = Ly the admittance of the dry sensor.
Eq (5) describes the linear response of sensor measurement to flow front position and allows the online estimation of the length of sensor covered by resin using the measured admittance, the admittance of the dry sensor and the admittance of the fully wetted sensor. It should be noted that the latter two need to be determined for the conditions of the measurement, matching the reinforcement architecture and pressure applied during impregnation.

Analysis of cure sensor signal
In order to uncover the quantitative characteristics of the resin reaction during cure process it is necessary to translate the dielectric/impedance signal to information related to resin reaction state. The behaviour of a thermoset under an AC excitation is governed by three phenomena: dipolar relaxation; charge migration and; electrode polarisation. These phenomena can be represented by the equivalent circuit illustrated in Figure 3a where A and n are coefficients of the constant phase element and ω the angular frequency.
For a lossy dielectric such as a curing epoxy resin with electrical behaviour dominated by migrating charges and a small contribution by dipolar relaxation, the equivalent circuit can be simplified by replacing the sub-circuit corresponding to dipolar relaxation and migration charges with a parallel R-C circuit [39]. Hence, Eq (6) is transformed as follows: The first term in the left hand side of Eq (7) denotes the material impedance (Z " ) and the second term the constant phase element. Figure 3b illustrates the imaginary impedance spectrum of the simplified equivalent circuit. The imaginary impedance decreases linearly in a log-log plot with a slope equal to the exponent of the electrode polarisation and coating term (n) in the low frequency zone of the spectrum where this term dominates the signal. The peak region is dominated by migrating charges, whilst the local imaginary impedance maximum (IIM) is related directly to the resistor corresponding to migrating charges (Z ≈ R 2 ⁄ ) [39]. The high frequency zone is governed by the capacitive elements of the circuit, which in the case of the simplified spectrum expressed by Eq (7), is manifested as a line with a slope of -1 in a log-log plot. In the case of the more complex behaviour represented by Eq (6), the high frequency response incorporates a small shoulder corresponding to dipolar relaxation with a linear behaviour with a slop of -1 at very high frequencies.
The imaginary impedance maximum is used in practice to monitor the cure based on the dependence of resistivity on migrating charges mobility, which in turn depends on local material viscosity [40]. This can implemented through a normalised form of the IIM [26,27,39] or through treating its time derivative as a signal equivalent to heat flow in differential scanning calorimetry [41].

Experimental set-up
The evaluation of the lineal sensor was performed during RTM processing of two carbon fibre/epoxy composite panels under different injection pressures. The RTM tool is a rectangular cavity with dimensions 900×340×3.3 mm. The reinforcement material was a 5H satin weave carbon fabric (Hexcel HexForce ® G0926) with an areal density of 375 g/m 2 [42].
The resin system was Hexcel HexFlow ® RTM6 epoxy [43]. The preform with dimensions 800×340×3.3 mm comprised nine fabric layers in a [(0F/90F)2/0F/(90F/0F)2] sequence resulting in a volume fraction of 57%. The filling was carried out at a constant temperature of 120 °C under a pressure of 2 and 3 bar for the first and second RTM run respectively, with the simultaneous application of vacuum at 10 mbar. The curing was performed after the end of filling at 160 °C for 2 h where the heating ramp was equal to 1.5 °C/min. The lineal sensor was made by twisting two 136-AWP solid copper wires with polyurethane enamel coating [44] using a hand drill. The diameter of each wire is 127 μm, whilst the pitch of the twist is 500 twists/m as illustrated in Figure 4. The lineal sensor with length 800 mm was placed in the centre of cavity of the RTM tool as illustrated in Figure 5 and was connected to a Solartron 1260 Impedance Analyser. Impedance data were acquired over seven frequencies swept logarithmically in the 100 Hz -100 KHz range. The analyser communicates with a computer via an IEEE interface. An in-house code developed in LabVIEW was utilised to drive the measurements and acquire the data. The code controls the data acquisition from the impedance analyser and thermocouples where necessary. It operates in a loop from the highest frequency to the lowest frequency, triggering a measurement and the corresponding acquisition at each frequency, followed by acquisition of temperature data where applicable.
The impedance analyser data are acquired in the form of a resistance and capacitance of the overall system, which are assumed to be connected in parallel. These are subsequently translated to real and imaginary values for further processing of results. A toughened glass top plate was used in the RTM tool to allow the visual measurement of the flow front for validation purposes using a digital camera acquiring images during the filling. Eq (5) was used to calculate the flow front position from the admittance data for the two RTM runs. The experimental set-up and the data acquisition system are illustrated in Figure 6. of the VARTM assembly. Impedance data were acquired over 25 frequencies swept logarithmically in the 1 Hz -1 MHz range using the Solartron 1260 Impedance Analyser. The results of temperature monitoring were used to compute the evolution of the degree of cure based on a non-parametric kinetics model of the cure of the resin system of this study [45].

Flow monitoring
The evolution of the flow front for the case of the 3 bar injection is shown in Figure 8.  According to Eq (5) the length estimated using the admittance measurements is the ratio of two complex numbers. In an ideal situation, in which the sensor field, material state and environmental conditions are identical across the length of the sensor, the numerator and denominator of the ratio are in phase resulting in a negligible imaginary part of the estimated length. Deviations from these ideal conditions, as well as measurement noise, cause higher values of imaginary length. Consequently, the sensor response provides a direct indication of measurement and corresponding analysis errors [1]. In the results presented in Figure 9 the imaginary impedance length in predominantly between -10 and 10 mm with a few extreme values reaching an absolute value of 25 mm. This is below 3% of the overall length, indicating a high quality estimation based on the real part of the length computed using Eq (5).
The results obtained using the flow sensor show clearly that the concept developed is applicable to liquid moulding of carbon composites under industrial conditions. The sensor withstands RTM level pressures, whilst its placement between the metal tool and the carbon fabric tested here is the worst case scenario in terms of potential interference by conductors as well as potential damage to the insulating coating. The air/resin pockets formed around the sensor by the fabric are sufficient to guarantee high sensitivity to filling state. The disturbance in the fabric architecture caused by the presence of the sensor is minimal, whilst the sensor is easily removed after curing if it is placed on the surface. The fabric conforms around the sensor forming a groove with a maximum dimension of around 250 μm as illustrated in Figure 10. The dimension of the disturbance is governed by the wire diameter, which can be minimised by selection of a thinner coated wire for the twisted pair.

Cure monitoring
The evolution of the imaginary impedance spectrum during the isothermal cure of neat resin at 150 °C and 160 °C is illustrated in Figures 11a and b respectively. A linear log-log reduction of imaginary impedance at low frequency is observed as expected. At intermediate frequencies, the plot shows a shoulder at the location where the local imaginary impedance peak is expected, whilst at high frequency the plot reverts to a linear log-log behaviour. The manifestation of migrating charges as a shoulder in the spectrum instead of a peak, which is the case of standard sensors [41], can be attributed to the insulating coating of the twisted wires, which behaves as an additional -mostly capacitive -element acting alongside electrode polarisation and results in a higher effective value of A in the equivalent circuit. In terms of physical phenomena the presence of the coating limits direct charge migration towards the electrodes, with the migration mostly occurring up to the boundary of the coating and the curing material and limited by the polarisation of the insulating layer as the field alternates.
As the curing progresses the spectrum shifts to lower frequencies and high impedances. This behaviour, which is the same as observed with conventional sensing elements, is attributed to the effect of increasing viscosity during the cure, which results in reduced mobility of charge carriers and increased timescale in their response.
The manifestation of migrating charges as a shoulder instead of a peak in imaginary impedance does not allow use of the conventional analysis based on the IIM to monitor progress of cure [26,27,39].  (7) is illustrated in Figure 12 alongside the original spectrum. The material impedance spectrum incorporates a pronounced peak which can be used instead of the IIM for estimating the state of cure. Figures 13a and b illustrate the evolution of the maximum of Z " for the two isothermal neat resin experiments and compare it with the results of the cure kinetics model based on calorimetric data [45]. In both temperatures the response of the cure sensor follows closely the resin reaction. The results highlight the cure sensor efficiency in terms of monitoring the degree of cure evolution during the whole process.
Imaginary impedance spectroscopy can also be used for identifying the vitrification point of the resin during the cure. This follows from the influence of vitrification on dipolar relaxation and is manifested as a secondary shoulder in the evolution of imaginary impedance at fixed frequency [46]. Figures 14a and b illustrate the imaginary impedance evolution at 1 KHz and 10 KHz alongside the specific heat capacity for the isothermal curing of neat resin at 150 °C and 160 °C. The imaginary impedance at fixed frequency shows a two-step behaviour; the first major step corresponding to the effect of curing on migrating charges and the secondary step to vitrification. The vitrification is manifested at 50 -70 min and 35 -55 min at 150°C and 160°C respectively. This period can be compared with the vitrification time identified as a step change in specific heat capacity during the cure. The specific heat capacity for the resin system of this study was calculated using the model reported in [47]. The step in specific heat capacity curves occurs at 65 min and 50 min at 150°C and 160°C respectively, which shows that the sensor indications of vitrification are in agreement with the calorimetric manifestation of the phenomenon. The comparison highlights the capability of the cure sensor to identify the vitrification of the resin during isothermal runs.
The results of cure monitoring during VARTM of a carbon reinforced composite are illustrated in Figure 15. The response is very similar to the case of the neat resin, showing that the sensor monitors resin material changes in the presence of carbon reinforcement and is able to follow the cure. The reaction progress as monitored by the cure sensor is in very good agreement with the corresponding estimation of the cure kinetics model, as observed in Figure   15a. Figure 15b illustrates the imaginary impedance evolution at fixed frequencies alongside the evolution of specific heat capacity. The vitrification is manifested as the second step of The capability demonstrated here is the same as that of state the art systems used for cure monitoring of neat resin or non-conductive reinforcement such as glass. This has significant industrial implications given that application of monitoring and control is far more relevant in high end applications dominated by the utilisation of carbon fibre reinforcements.
In addition to the capability to operate during the manufacturing of carbon reinforced composites the sensing concept has a low cost in terms of both raw materials and manufacturing method. Furthermore, it is applicable to non-conductive reinforcement and compatible with current industrial and research dielectric cure monitoring systems. The geometry and material of the sensor can be optimised further to allow full exploitation of the concept presented in this work in the composites manufacturing industry. The contribution of wire insulation can be controlled through selection of a material that minimises its influence, while making sure that its behaviour is highly stable at temperatures relevant to composites processing. In high-end applications the disturbance in fabric architecture, introduced by the sensor, can be a limitation. The ability to scale down the size of the sensor -currently at about 250 μm -by utilising thinner insulated wires can enhance the non-intrusive character of the concept. Furthermore, the form of the sensing element lends itself to direct incorporation onto fabrics, offering potential routes for producing smart materials with process monitoring capabilities.