Improved sensitivity of liquid sensing melt-spun polymer fibers filled with carbon nanoparticles by considering solvent-polymer solubility parameters

The Hansen Solubility Parameters (HSPs) and the Relative Energy Differences (REDs) were used to select suitable polymers to perform sensing experiments of electrically conductive nanocomposites against different solvents to employ such materials for sensor applications. From the solvent-polymer HSPs and their REDs, it was determined that polycarbonate (PC) is a polymer with potential towards high liquid sensitivity for different organic solvents. Furthermore, PC is spinnable and sensing fibers of different diameters can be easily produced. In order to get electrically conductive materials, PC was melt-mixed with carbon nanoparticles such as Multi-walled Carbon Nanotubes (MW) and Carbon Black (CB). The materials were then spun into fibers via melt-spinning and evaluated for liquid sensing. It was found that combining MW and CB (50/50 wt%) improved the fiber spinnability and their sensing range in comparison to fibers made of PC and only MW. Liquid sensing evaluations showed that knowing the REDs in advance allows predictions on the ability of nanocomposite fibers to be highly sensitive to specific solvents. From the sensing evaluations it was found that fibers made with hybrid fillers of CB+MW, drawing down ratios of 4 to 12 and filler contents between 3 and 6 wt% showed the best liquid sensing abilities. Depending on the composition and the conditions of fiber production, the maximum values of relative resistance change for fibers with CB+MW were always higher than for fibers containing MW only. Testing for instance butyl acetate as solvent, the fibers with MW+CB achieved values between 600%–3200% for long immersion times, while values between 390 and 1200% were obtained for the fibers with only MW. Testing a selected fiber under a simulated leakage scenario and as sheath/core bi-component fiber showed the effectiveness of these fibers working as liquid detector as well as potential for applications beyond single component fiber sensors.


Introduction
The growing development of new technological advances exposes the demand of diverse kinds of sensors, such as biological sensors [1], mechanical sensors [2], and chemical sensors [3]. Nanotechnology is one of the most promising areas for achieving high efficiency, easy-to-use, up-scalable and low-cost sensor fabrication. In specific, nanotechnology based on conductive polymer composites (CPC) filled with carbon nanoparticles offers the advantage of wide versatility and high processability in diverse fabrication techniques over other conventional sensor fabrication techniques. For this reason, nanocomposites (NCs) used as sensing materials can be found in a broad variety of shapes and dimensions, such as films [4], disks [5], balloons [6], 3D-printed structures [7], filaments [8], fibers [9], and many other structures [10]. In addition, various types of conductive carbon fillers such as carbon nanotubes (CNTs), carbon black (CB), graphene and graphene-like structures, nanofibers, onion-like structures, and nanohorns can be used, further expanding the range of possibilities [11][12][13][14][15][16][17].

Nanocomposite fibers as sensitive materials
In contrast to any other kind of custom-shaped nanocomposite, polymer-based NC fibers fabricated through melt-spinning offer the chance of making up to kilometers of sensing material (in this case, in the shape of a fiber) with only few grams of nanocomposite (NC) material. This is something which almost cannot be taken to practice with any other material rather than with polymer-based NCs. In addition, conductive polymer-based NCs have the advantage over other new technologies that the processing techniques required to produce them as sensors already exist and are widely used [18]. This further expands the possibilities of weaving the conductive NC fibers into textiles that can later be employed as sensing fabrics (e.g. wearables). In addition, contrary to traditional sensors, no additional components such as integrated circuits are needed to enable electrically conductive polymers to convert physical stimuli into electrical signals. Nevertheless, most common meltspinnable polymers are inherently electrically insolating, showing the need to add conductive elements to the polymer for providing them sensing abilities. Therefore, in the production of sensor fibers, attention is paid to the particles used as conductive elements as well as to the processing technique employed for fabricating the CPC sensors. Due to their high aspect ratio (length/diameter ratio), electrical conductivity and proven ability to endow polymers with sensing abilities, carbon particles such as multi-walled carbon nanotubes (MWCNTs) are among the most commonly used particles for the production of CPC-based sensing materials [19][20][21]. On the other hand, melt spinning has proven to be an effective method for fabricating sensing fibers made from polymers filled with carbon particles [22,23].
Although there are many reports of the use carbon particle filled NC fibers as sensing elements, only few reports exist on their use for detecting liquids [24][25][26]. On liquid sensing fibers made of poly(lactid acid) (PLA), Pötschke et al found that when using MWCNTs, lower filler amounts lead to higher sensitivity against ethanol [27]. Similarly, in their work they found that an increase in temperature led to higher changes in electrical resistance, which is an effect of modified diffusion kinetics. Rentenberger et al reported the use of melt-spun multi-filaments based on a blend of two biodegradable polymers, poly(ε-caprolactone) (PCL) blended with PLA containing 4 wt% of MWCNTs, for detecting solvents such as acetone and ethyl acetate [28]. In their work they exposed PCL/PLA/MWCNT fibers to several immersion/drying cycles showing that these fibers have a very fast response and recovery (drying after exposition to the solvent was removed). However, an increase in the maximum change in resistance was seen after each cycle and the resistance does not return to its original value after the first cycle. They attributed this behavior to a rearrangement of the polymer chains after the first immersion of the fiber into the solvent.
Besides MWCNTs, other conductive carbon fillers such as CB have been reported to be added for preparing polymer composites with liquid sensing properties, yet no report has been made on the use of CB filled polymer fibers for its use specifically as liquid sensors. This is most probably due to the fact that, in contrast to polymer/ MWCNT composites, high amounts of CB have to be added to achieve appropriate liquid sensing properties. Due to the spherical shape of its primary particles, which can form aggregates of different shapes, CB has a much higher percolation threshold to achieve conductive composites, which are a prerequisite for the sensor capability. CB amounts in the range of 4-20 wt% have been reported as the minimum needed to have stable liquid sensing responses in non-fiber shaped composites [29][30][31][32]. In terms of spinnability, such high amounts of filler may significantly difficult the melt-spinning process and, in particular, could cause blockages at the filter and at the cavities inside the die. On the other hand, polymer fibers with hybrid filler systems of MWCNTs and CB have not yet been evaluated as liquid sensing materials. Hybrid filler systems could potentially present an interesting synergic/combined effect in the liquid sensing behavior that is worth investigating. In particular, CB and MWCNTs are interesting as they have different aspect ratios and surface areas. Thus, melt-spun polymer composites with hybrid carbon fillers are a promising way to fabricate liquid sensing materials with a wide range of possibilities for implementation. Furthermore, through their versatility, hybrid filler NC fibers open a door to many potential applications such as such as strain sensing, structural health monitoring, detection of vapors/ liquids and flexible sensors and enables the fabrication of smart multifunctional textiles.

Material selection aspects for sensing
In order to get a fast and high response in nanocomposites that can detect the presence of predetermined liquids, a proper polymer matrix has to be selected. This is based on the fact, that the liquid-polymer interactions determine the sensing capability of the sensing materials as well as the selectivity towards a variety of liquids. In the specific case of preventing leakage of hazardous liquids such as solvents, the sensing polymer-based material should have a certain behavior when exposed to solvents in order to function as a sensor for liquids (e.g. solvent). In this case, the polymer matrix should be able to swell without dissolution during the solvent exposure so that the electrically conductive network inside the matrix alters its arrangement and by that, the electrical resistance changes, which is the signal to be detected. There are certain shortcomings in some of the polymer-based liquid sensor materials described in the literature. For instance, some of them are based on biodegradable polymers which are not durable over extended periods of time, especially when exposed to typical environmental conditions [28,32,33]. Others need a high amount of a conductive fillers to achieve a fair solvent sensitivity or react only slowly and relatively poorly to the presence of liquids [29][30][31].
In the early 2000 s, the first attempts were made towards liquid sensing materials using CPCs containing carbon nanoparticles (CNPs) such as CB [29][30][31][32]. Tsubokawa et al achieved a change of 5%-10% in electric resistance when exposing paste coatings of poly(ether glycol), poly(ether amine) and PCL grafted with CB to chloroform, methanol and trichloroethane [32]. By using thermoplastic polyurethane (TPU) rods added with 4 phr CB, Segal et al successfully detected the presence of methanol, ethanol and 1-propanol [29]. However, the response upon immersion was very slow requiring even minutes to show a significant change in the electrical resistance and cyclic tests were unsuccessful. In another study, Srivastava et al found in a circular extrudate of a blend composite consisting of polypropylene(PP)/polyamide 6/CB (75/25/3) that cyclic liquid sensing tests are possible for benzene and n-heptane [30]. Nonetheless, the relative resistance is not fully recovered after the first immersion. Frackowiak et al immersed PP/CB and polystyrene (PS)/CB composite filaments with different CB amounts in benzene, toluene, xylene and ethylbenzene and found that the sensitivity is higher for CB loadings near the electrical percolation threshold [31].
These early works proved that polymer/CB composites can be used as liquid sensor materials, but proved to be poorly efficient. One main reason for that low efficiency is that most authors did not consider the solventpolymer interactions. However, for liquid sensing applications, the degree of interaction between a desired solvent to be sensed and a polymer has to be taken into consideration. A more efficient route to get suitable sensor materials is by selecting polymers and solvent combinations according to their solubility parameters. Nevertheless, among other works in this direction, Narkis et al reported for a high impact PS/ethylene-vinyl acetate co-polymer blend filled with CB upon immersion in various solvents (benzene, methanol, methyl methacrylate and n-heptane), that the standard (Hildebrand) solubility parameters of the polymers and solvents do not have a very significant impact on the magnitude of the composite's sensitivity to the tested solvents [34]. This is supported by the findings in a work by Pötschke et al where the electrical response of PLA/MWCNT composite fibers tested for ethanol, methanol and n-hexane could not be easily related to the differences between the Hildebrand solubility parameter of the different solvents and the PLA [27].
On the other hand, the Hansen solubility parameter theory introduced by Dr Charles M. Hansen in 1967 could be a reliable tool to select solvent-polymer combinations with higher interactions that leads to liquids sensors with improved sensitivity. Contrary to the most commonly used Hildebrand's solubility parameter, the Hansen solubility parameters take into account the energy from dispersion (δ D ), intermolecular (δ P ) forces, and hydrogen bonds (δ H ) together with the interaction radius (Ri) of the polymer [35,36]. The Hansen solubility parameters (HSP) were conceived to be able to predict the solubility of polymers in solvents. Currently, they are still widely used successfully in the paint/coatings industry and more recently even in printable energetics [37]. The principle is based on the idea that 'like dissolves like', meaning that if a solvent has HSPs very similar to that of a polymer, a high interaction between them is expected. According to HSP theory and diffusion kinetics, it is possible to predict if a selected polymer will remain unaffected, if it will swell, or if it will dissolve upon exposition to a solvent as a result of solvent diffusion into the polymer chains. When the solvent diffuses into the carbon particle filled polymer matrix, it then induces an increased distancing between the conductive particles [23,35,38]. On the other hand, it is also possible to detect which polymer is the best suitable to sense a certain solvent or a range of solvents. In the case of CPCs with carbon fillers the swelling effect of the polymer matrix is of most interest, as the swelling ultimately translates as rearrangement of the percolated particle network inside the polymer composite. The extension of the conductive network leads to partial disruption of particle contacts and increases the distances between neighbored CNTs up to values above the electron tunneling or hopping distance. This phenomenon can be applied to design suitable CPC liquid sensor materials as the swelling behavior of the selected polymer can be determined from the knowledge of the change in electrical response of the conductive nanocomposite during swelling when in direct contact with different methodically selected solvents. Yet, a proper selection of solvent-polymer combinations is necessary to achieve high solvent sensitivity that comes from a significant swelling effect, but without dissolution of the polymer matrix. Figure 1 depicts this working principle of a CPC based sensor material, in which the increased distances between the conductive particles promoted by polymer matrix swelling causes an electrical resistance change in the CPC, which can be monitored in situ during solvent exposure. Following this solvent-polymer interaction-based swelling principle, some researchers have proven that the HSP can be a promising tool to manufacture sensitive MWCNT filled polymer composites capable of detecting common use hazardous liquids such as organic solvents [23,27,28,33]. Furthermore, this principle it is not restricted to solvents as liquids, it also has been reported to apply to their vapors since this process is related to interaction of the polymer chains and the solvent molecules regardless of its physical state [11,39,40]. This shows the great potential of CPCs as organic solvent sensors, as vapor and as liquid.
In the specific case of liquid sensing using small-diameter filaments and fibers, some efforts have been reported. Qi et al fabricated a liquid sensing fiber based on a cellulose/CNT composite and tested its sensitivity against water [41]. In their work, the cellulose fibers were dip coated in dispersions with different amounts of MWCNTs achieving a fast response to water and also to diverse aqueous electrolyte solutions. Furthermore, from analyzing the solvent vapor sensing abilities of PLA/MWCNT biodegradable composite filaments, Silva et al concluded that if aiming at the production of a wearable functional textile or textiles for leakage detection, smaller fiber diameters are preferred [11]. This additionally suggests that the use of fibers as solvent sensing elements is a promising approach towards up-scalable nanocomposite sensors. Studying the influence of particle concentration in the detection of solvents, Pötschke et al found that the capabilities of MWCNT/PP/PCL filament blend composites to sense n-hexane, ethanol and methanol decrease with increasing amount of CNTs. This can be attributed to a reduction in the changes in the particle-particle distances upon swelling due to the initially denser particle networks at higher filler contents [33]. Furthermore, Fan et al found that the intensity of the electrical response of multifilaments of TPU coated with CNTs is correlated with the solvent-polymer interactions [42]. This supports the hypothesis that higher sensitivity can be achieved when taking into consideration solvent-polymer solubility interactions. Moreover, following this principle, Villmow et al successfully selected polycarbonate (PC) as matrix to detect the presence of tetrahydrofuran, acetone and ethyl acetate using a MWCNT/PC composite and achieved a very fast response within the first seconds of exposure [38]. Furthermore, Villmow et al reported that polymer/MWCNT fibers can be woven into textiles showing the great potential and versatility for developing smart structures based on polymer/CNT composites [23].
Similarly, Rentenberger et al tested the sensing abilities of a textile made of biodegradable PLA/PCL + 4 wt% MWCNT = 50/50 multifilaments against ethyl acetate and acetone and measured changes in resistance up to 3 times of its initial resistance [28]. Nevertheless, they highlighted the complexity added by the fiber fabrication which becomes more difficult as the melt viscosity of the composites increase with increasing CNT volume fraction. In this direction and added to the early findings using polymer/CB fibers, a hybrid filler system of MWCNTs and CB seems to be a possible solution. Such systems could combine the lower electrical percolation threshold of CNTs with the better flowability and lower price of CB containing composites [43]. For example, previous studies on PC melt-mixed with combinations of CB and MWCNTs found an electrical percolation threshold of 1.4 wt% for 50/50 wt% mixtures compared to about 5 wt% for CB and 0.7 wt% for MWCNTs. On the other hand, the transient melt viscosity (at 230°C) of a composite with 3 wt% hybrid filler is only 63% of the viscosity when CNTs alone are used as fillers [44]. Lower needed filler content and lower melt viscosity could help to enhance the spinnability of electrically conductive fibers.
Even though, liquid sensing capabilities of carbon particle filled polymers are promising, there are still critical issues and open questions that have to be addressed in order to significantly improve the sensitivity of these sensors in the shape of fibers. For instance, orientation processes in the polymer chains and the fillers occur during the fiber spinning process, especially when using high-aspect ratio fillers which not only changes the electrical resistance of the fibers but also their rearrangement processes during solvent exposure [22,27,45]. However, the approach to design melt-spun fibers made from non-biodegradable CPCs with a hybrid filler system and a suitable polymer selected following the HSP theory is very interesting towards the development of fast responsive liquid sensing fibers that are durable, highly sensitive to specific organic solvents and up-scalable for large production with standard fiber fabrication techniques. This kind of melt-spun fibers has tremendous potential towards low-cost multi-solvent sensing fibers and thus more investigation is needed for transferring the knowledge of this kind of materials into tangible industrial applications, textiles and wearables.
In this work we fabricated melt-spun liquid sensing fibers from melt-mixed polymers enabled to have conductive character by adding a hybrid filler system of MWCNTs and CB (50/50 wt%) using different spinning parameters. The hybrid filler system of MWCNTs and CB was selected in order to enhance the spinnability and to enhance the stability of the conductive network structure upon the spinning process. In addition, a better sensing ability can be expected from combined sensing contributions of both kind of particles. The polymer finally used was selected employing the tool of the HSP from various commercially available spinning grade polymers in order to detect a wide range of common solvents. The fiber fabrication parameters, the carbon particle content, and the fibers sensing behavior were studied aiming to distinguish which conditions provide the best performing liquid sensing fibers. At the end of the work the best performing fiber was selected and put under test on a simulated contamination scenario for evaluation its capabilities for real-time leakage detection. In addition, a new approach towards bicomponent carbon particle filled NC fibers for liquid sensing is presented.

Materials and methods
2.1. Polymer matrix selection using HSP As described in section 1.2, the polymer chosen for fiber production must be swellable in the presence of the liquid to be detected so that conductive polymer fibers filled with carbon particles give sensing signals. A reliable approach to estimate if a polymer will swell or not in the presence of a particular fluid is by knowing their affinity using the Hansen Solubility Parameters (HSP) of both, the polymer, in this case the fiber matrix material and the liquid that is to be sensed; a solvent in our case. When the HSPs of both elements are known, a way to quantify the affinity between a polymer and a solvent is by calculating their Relative Energy Difference (RED). The RED correlates the radius of interaction of the polymer and the solvent-polymer distance in the HSP tridimensional space (Hansen Space) between the polymer and the solvent, as described by equation (1), where R i (in MPa 0.5 ) is the interaction radius of the polymer molecules and R a (in MPa 0.5 ) is the distance between the polymer and the solvent in the Hansen Space. Ri represents the degree of interaction and strength between the polymer molecules and R a indicates the distance between the HSP of the polymer and the solvent in the Hansen Space. Calculation of R a is possible by employing equation (2), where δ D , δ P , and δ H are the energies from the dispersion forces (D), dipolar intermolecular forces (P) and hydrogen bonds (H) between molecules; subscripts 1 and 2 indicates polymer and solvent, respectively. For the estimation of the RED, the values of δ D , δ P , δ H and Ri were taken from literature and datasheets [36,46]. Accordingly, a RED magnitude well above 1 indicates that the solvent will neither swell nor dissolve the polymer, a RED magnitude around or equal to 1 indicates that the solvent could only partially dissolve the polymer or that probably the polymer would just slightly swell and a RED magnitude below 1 indicates that exposure of the polymer into the solvent will swell and/or up to certain degree dissolve the polymer. This in turns means that as closer the RED magnitude is to 0, the solvent-polymer interaction is more likely to result in polymer dissolution if the polymer exposed to the solvent over a longer period. Thus, it is possible to know beforehand which polymers are potential candidates to work as liquid sensor based on the solvent-polymer RED magnitudes. However, knowing the RED magnitude leads only to a qualification of the degree of polymersolvent interaction and its accuracy on the swelling magnitude has to be validated prior polymer selection. Moreover, it has been established that the ability of electrically conductive polymer nanocomposites to sense liquids (to change their electrical resistance) is directly related to their swelling magnitude in the presence of solvents. This can be verified by quantifying to which extent a polymer swells in the presence of diverse solvents with known HSPs and calculating the solvent-polymer REDs which can be later correlated with the sensitivity achieved by the sensor material.

Swelling measurements
A way to validate estimated RED numbers, is by determining the magnitude of polymer swelling in the presence of high and low affinity solvents and correlate it to their REDs. This in turn allows to afterwards choose a proper polymer with high affinity towards sensing several solvents. The RED numbers between a total of 24 solvents plus water and 8 highly spinnable polymers were estimated using equations (1) and (2) and the HSPs parameters taken from [36,[46][47][48]. The parameters employed for solvents and polymers are listed in tables 1 and 2 respectively; units of HSPs are MPa 0.5 , equivalent to J cm −3 .
Based on their contrasting REDs estimated, their high spinnability, and in order to know the validity of the calculated RED values, PA 6 (Novamid ® 2.7; DSM, The Netherlands), PA 66 (Ultramid ® A27; BASF, Germany), PET (RT20; Invista, USA) and polycarbonate (PC) (Makrolon ® 2205; Bayer MaterialScience AG, Germany) were the polymers selected to carry out swelling measurements, while the most commonly employed solvents from table 1 were used.
The polymer samples as pellets were fully immersed into a Petri dish filled with 20 ml of the solvent under evaluation and the Petri dish was covered with glass to prevent solvent evaporation during testing at room temperature (RT). The previously weighted sample was immersed into the solvent for a defined time (5, 15, 30 and 60 min); during this time the samples were under observation for any possible change. After the desired time, the sample was carefully taken out, any solvent on the surface was removed and then weighted again; weights were measured in a high-resolution balance.
The volume gained by the polymer after immersion into the solvent was calculated by employing the following equation, where ρ p and ρ s are the densities of the polymer and the solvent, respectively, and m i and m f are the weights before and after immersion, respectively. The calculated volume increase based on the densities provided by the material suppliers may differ slightly from the actual increase, yet this difference should be negligible for the scale of this test. Densities of solvents and polymers were taken as reported by suppliers (ρ PA6 = 1.14, ρ PA6,6 = 1.13, ρ PET = 1.45 and ρ PC = 1.19; units are in g cm −3 ) and public data bases [49].

Materials and fiber fabrication methodology
The polymer fibers were manufactured following a melt-spinning process where pelletized nanocomposite material made of previously dispersed carbon nanoparticles (CNPs) into the polymer is melted and melt-spun into fibers. Based on the results of the swelling experiments, the selected polymer, PC, was employed to produce nanocomposites using a microcompounder MC15 (Xplore, Sittard, The Netherlands) with a capacity of 15 cm 3 under the following conditions: melt temperature: 280°C, mixing time: 5 min, screw speed: 250 rpm, screw type: twin conical co-rotating screws (see figure 2(a)). The polymer and composites were dried in a vacuum oven at 100°C for 8 h before each processing step. The manually pelletized nanocomposite extrudate was then placed inside a piston-type spinning device where the pellets were allowed to melt for 5 min inside the barrel at the temperature of 280°C before spinning. Then the molten composite was pushed through a die (capillary diameter D 0 of 0.60 mm and capillary length to diameter ratio of 2) and the exiting thread was pulled by a mechanical winder to draw down the molten thread and form fibers with diverse diameters (⌀). Spinning conditions such as take-up speed and throughput were varied in order to obtain as many draw down ratios (DDRs) as possible at which electrically conductive fibers can be produced. Equations for determining the DDR are detailed in [22] from which it was obtained that DDR = D 0 2 /⌀ 2 . The DDR provides a quantifiable value of the pulling and thinning processes that are used to produce the fibers from the composite material. For more details on the fabrication process of the polymer/CNP composites and the fibers used in this work, the reader is referred to Bautista-Quijano et al [22]. The DDRs obtained after spinning ranged between 4.04 and 481. The content of nanoparticles was adjusted depending on the spinnability of the nanocomposites and their electrical properties and ranged between 1 and 6 wt%. The MWCNTs used were NC7000 (Nanocyl ® S.A., Belgium) and the CB employed was Printex XE 2-B (Orion Engineered Carbons, Germany).
In order to analyze changes in the spinnability of the melt from the added fillers, the melt pressure was measured from pressure sensors in the piston at the processing temperature. In addition, melt viscosity measurements were done on a parallel-plate rheometer (Haake TM Mars TM rheometer) in oscillation mode at the spinning process temperature. The electrical resistance of the fibers was measured employing the alternating polarity method in a Keithley Electrometer 6517 A with a test fixture 8002 A from which the resistivity was obtained. The maximum electrical resistance measurable is ∼200 TΩ; which for our samples' geometry is equivalent to an electrical resistivity of ∼10 12 Ω cm.
Finally, to observe the arrangement of the particles inside the polymer fibers, Scanning Electron Microscope (SEM) images were taken in Charge Contrast Imaging (CCI) mode on NC fibers. The fibers were frozen in nitrogen and then samples were taken from thin transversal cut along the fiber direction.

Liquid sensing characterization
Prior sensing characterizations, the electrical resistivity of the fibers (R 0 ) was measured in order to determine whether the fibers are conductive enough and suitable for sensing characterizations. Thereby, the resistivity should be lower than 50 MΩ-cm. The resistivity measurements were carried out following the methodology described in [22]. From this initial evaluation, the broad options of processing conditions to produce fibers were then narrowed down and the most suitable fibers for liquid sensing characterizations were selected based on: low electrical resistivity, high spinnability, homogenous morphology, and low brittleness.
The fibers were immersed in the solvents for a certain immersion time t i followed by a drying time (t d ) in air at RT, where changes in the electrical resistance were measured during the immersion-drying process; t i and t d were defined according to the sensing behavior of the fibers. A dependency between the duration of the first cycle and the behavior in subsequent cycles was noticed. Thus, t i for the single immersion cycles was chosen to be long enough to reach a plateau (equilibrium). The solvents were selected according to table 1, the calculated RED values, and the swelling behavior found after swelling characterization. By properly selecting t i and t d as well as the solvents based on their HSP, the fibers were neither damaged nor changed in shape or appearance after repeated exposure. The electrical resistance (R measured ) was measured in situ using a Keithley Sourcemeter 2400 with computer communication. The relative resistance change (R rel = ΔR/R 0 with ΔR = R measured -R 0 ) was determined from the continuously over time recorded resistance values taken at every 2 s. A vertically adjustable in-house-constructed carriage was used to regulate and control the test-fixture height during immersion-drying cycles and to reduce any external stimuli, as depicted in figure 3. The electrical response of the fibers was stable presenting negligible noise ratio and thus no signal correction was needed. The total fiber length employed was 3 cm and the immersed fiber length was 1 cm. Silver paint was applied at a distance of 0.5 cm from both clamping zones to reduce the contact resistance between clamps and fiber. The temperature of the liquid was controlled by using an oil bath and a thermostat. Immersion/drying cycles were performed in order to evaluate the reproducibility of the liquid sensing capability of the fibers. Depending on the swelling of the fibers and the physical changes of the fibers during immersion, the immersion-drying times were adjusted. For comparison purposes, the magnitude of R rel can be considered as a quantitative measure of the sensitivity of the fiber sensor.

Polymer matrix selection
To evaluate the accuracy of the HSPs and then select an adequate polymer for liquid sensing against many different solvents, the RED values of several polymer-solvent combinations were calculated using the HSPs listed in tables 1 and 2 as inputs for equations (1) and (2). Only spinnable polymer types were considered, as they are also expected to enable the spinnability of their nanocomposites. Similarly, commercially available organic solvents of common use in different kinds of industries were taken into account. Polymers with RED magnitudes too high (well above 1) or too low (very close to zero) were initially discarded as they are not desirable for liquid sensing applications Solvents with too high or too low RED values show no effect or dissolve the polymer. In order to verify whether the RED estimations are reliable, swelling tests were carried out on specific spinnable polymers with high and low theoretical solvent interactions. Such polymers are those which show few or many RED values to solvents close or below 1. As polymer with poor solvent interaction PA 66 was selected, while PC was selected as the best candidate polymer due to its high interaction with many solvents. PET and PA 6 were also tested for validation of the theoretical findings as they account for intermediate solvent interactions limited to specific solvents.
The selected polymers were tested in 24 solvents and the swelling gained (Vg) was determined with equation (3); the solvents with the most significant changes in Vg are shown in figure 4 along with their respective {REDs}.
In the case of PET, PA 6 and PA 66, all solvents that caused swelling are shown, while in the case of PC that had several high swelling solvents, only 8 of them are presented in figure 4. Since depending on the density of the solvent, a variation in weight of ∼0.1 mg could give an erroneous change of ∼1.25% in Vg, a dotted line was fixed at Vg = 1.25% in figure 4. All values of Vg below the dotted line were not considered as swelling. Additionally, given that PA 6, PA 66 and PET exhibited low swelling after 60 min, the evaluation time was quadruplicated for these three polymers. However, only PET showed a further increase in Vg after 240 min, while PA 6 and PA 66 showed no further swelling after 240 min in direct contact with the solvents tested. Figure 4 shows that increasing the time of exposure also increases the amount of swelling, which is expected from extended solvent absorption. Moreover, a fast increase on the swelling in the first minutes is seen, followed by a slower increase rate in the subsequent time. The swelling behavior exhibited by these polymers is similar to a power law function, which follows the diffusion kinetics, described e.g. for PC by Villmow et al [38].
Also visible from figure 4, is that both PA6 and PA66 show small amounts of swelling. In particular, for PA66 most solvents produced swelling below 5% showing its poor interaction with the solvents tested, which is in accordance with the RED estimations. PA6 also shows only small amounts of swelling, however it was swollen by higher amounts of solvents than PA66.Similarly, although some swelling occurred with PET, the maximum swelling was 15% for chloroform (RED = 0.70) and 8% for DCM (RED = 0.10). Interestingly, upon extended exposition (240 min), the PET samples tested in DCM suffered significant dissolution, which matches the prediction by the estimated RED.
In the case of PC, the extent of swelling was largely in line with the REDs' expectations. As figure 4 shows, with the different solvents tested, much higher swelling was achieved in PC after only 5 min of exposition than in the other polymers. For instance, for PC it was possible to achieve up to nearly 30% of swelling after 60 min for various solvents such as acetone (RED = 0.  RED = 0.49) resulted in complete polymer dissolution in less than 1 h. It is not in the scope of this work to investigate whether there is a correlation between the low RED values and the degree of dissolution of the polymer. Nevertheless, with a few exceptions, it can be stated that the behaviors found in these tests in terms of swelling and dissolution of the tested polymers correspond well with the RED value. REDs close to zero led to polymer dissolution, REDs below (yet close to) 1 had a swelling effect and REDs too high had practically no effect on the polymers.
After the swelling evaluation it can also be stated that PET is highly selective since it presented high amount of swelling by only very few solvents. In contrast, PA 6 and PA 66 are poorly swellable, while contrary to the rest, PC shows a high amount of swelling. These results show that with the use of the HSPs it was possible to predict to certain extent the polymers swellability in accordance with their estimated RED values, which is essential for designing/tailoring a liquid sensing polymer composite. Therefore, these results suggest that it is indeed viable to select a polymer for a potential use as liquid sensor by knowing its RED magnitude against different liquids. Additionally, PC exhibited a very fast swelling response to high affinity solvent which is desirable for liquid sensor applications. Moreover, PC showed great potential for its use as a detector for a variety of solvents. Subsequently, solvents with 0.70 RED 0.90 should be primarily considered when evaluating the liquid sensing ability of PC polymer fibers.  3.2. Spinnability and resistivity of the melt-spun fibers As described in section 2.2, different amounts of MWCNTs and CB were melt-mixed in PC and subsequently the nanocomposites were melt-spun by varying the spinning parameters so that fibers at different DDRs were obtained. For reference, the molecular weight (M W ) of the specific PC used, (Makrolon 2205) was measured as 20,100 g(mol [50]. After preliminary spinning evaluations with the hybrid filler system, it was decided to use equal (1:1) weight concentrations of MWCNT (abbreviated MW) and CB. Thus, the nomenclature for composites with mixed fillers is from now on referred as PC/CB+MW 'X' wt%, where 'X' is the sum of the amount of both fillers at equal weight concentrations ratio. Table 3 summarizes the fibers fabricated showing their DDRs, diameter, electrical resistivity and the melt pressure measured during spinning. Among all the fabricated fibers only the ones of particular interest for discussion are shown in table 3.
As it can be seen from table 3, a wide composition range of nanocomposite materials were spinnable. Thereby, the highest achievable DDR was 481 (PC+1 wt% MWCNT) and the lowest 4.04 (PC+3-5 wt% CB +MWCNT). However, the spinnability of fibers was significantly reduced after the incorporation of the carbon particles. For instance, the maximum DDR achievable was 491 for neat PC and 481 for PC/MWCNT 1 wt%, while the highest possible DDR for PC/MWCNT 6 wt% was 4.83 (the lowest DDR also for all PC/MWCNT fibers). All attempts to produce MWCNT containing fibers above these DDR ranges caused breakage of the fibers. Some reports consider the reduced spinnability upon increased carbon particle content as a result of the presence of filler agglomerates remaining from the melt-mixing process [22,27]. The presence of nanocarbon particle agglomerates in the fibers generates defects that cause the fiber breakage upon stretching due to stress concentration around the defects. Hence, the presence of remaining agglomerates may increase the brittleness of the molten polymer composite upon drawing and thereby reducing its spinnability.
As in this work, Mazinani et al achieved a DDR maximum between 450 and 500 for melt-spun PET fibers with a MWCNT content of 3 wt% [51]. However, PET composites with MWCNT contents above 3 wt% were not spinnable, whereas in this work fibers with double the weight percentage were successfully melt-spun. Hooshmand et al achieved PP/PA/MWCNT fibers with 5 wt% MWCNT when using very low DDRs of 2 and 3 [52]. In literature, the maximum amount of MWCNTs to be added to melt-spun polymer/MWCNT fiber was reported by Soroudi and Skrifvars also using a polymer blend [53]. In difference to the results of this work, they were able to add up to 7.5 wt% of MWCNT to PP at a DDR of 4. They achieved these high amounts by blending PANI and PP to fabricate PANI/PP/MWCNT melt-spun fibers. However, the drawability of their fibers was consistently reduced as the take-up speed increases (increasing the DDR). Similar behavior was reported by Pötschke et al where take-up speeds higher than 20 m min −1 led to fiber breakage in PLA/MWCNT fibers [27]. Nevertheless, at MWCNT contents above 3 wt% Pötschke et al did not achieve PLA/MWCNT melt-spun composites due to fiber breakage upon drawing. Furthermore, the clear trend recognized as a decrease in drawability as a function of particle weight concentration was experimentally supported by the observed increase in the melt pressure inside the heating barrel. Such pressure increase could be mainly due to an increase in the melt viscosity upon particle addition which reduces the flow rate of the molten polymer composite making it more difficult to be pushed out through the die.
This was also visible as increased die-swell (droplet formation) at the die exit. To better understand this phenomenon, figure 6 shows the complex viscosity and the storage modulus of the nanocomposites that were later melt-spun into fibers.
As can be seen from figure 6, the complex viscosity and the storage modulus of the MWCNT and CB+MW filled melt increase significantly already after the addition of 1 wt% of carbon particles when compared to neat PC. Furthermore, the melt viscosity of PC/MWCNT 1 wt% is significantly higher compared to PC/CB+MW 1 wt%. The same is the case for 4 wt% filler; the viscosity of the hybrid filler system is lower than of the composite with only MWCNTs. Likewise, after the addition of CB+MW 1 wt% the storage modulus of the composite increases significantly and further increases at higher CB+MW contents. This finding points out a significant enhancement of the rigidity of the molten polymer nanocomposite with increasing wt%s. Furthermore, a lower increase in storage modulus is observed when CB+MW are added than when MWCNTs are added. This can be seen at 1 wt% filler, but even at higher filler content. The storage modulus of the PC/MWCNT 4 wt% composite is higher compared to PC/CB+MW 4 wt% and is very similar to the storage modulus of PC/CB+MW 5 wt%.
In correlation with the experimental spinning tests, these rheology results indicate that the storage modulus of the composites is so high at >3 wt% that the spinnability of the fibers is significantly hindered. These results prove that the increases in the melt viscosity and storage modulus of the composites containing carbon nanoparticles can greatly influence the capability of such composites to be melt-spun. Overall, experimentally the PC/CB+MW fibers showed better spinnability at high (>3 wt%) filler concentrations than those based on PC/MWCNT. This behavior can be attributed to the lower melt viscosity of the PC/CB+MW composites, which is expected to be due to the different density of the network structures of CB and MWCNTs inside the PC matrix.
Finally, since these fibers are intended to be used as liquid sensing elements, the ρ r of the fibers has to be within a suitable range for sensing. In this work, the suitable range was defined as the resistivity corresponding to resistance magnitudes measurable by most standards electrometers in the market. Interestingly, independently of the DDR not much change in ρ r was visible for most fibers below 3 wt% of carbon particles where it mostly remains around or above 10 10 Ω cm. Moreover, increased DDR led to an increase of the resistivity, which has been reported to be related to distancing of oriented carbon particles in the direction of the fiber axis [22,45]. This is clearly visible for instance for PC/MWCNT 2 wt% and 3wt% where an increase of up to 4 orders of magnitude in resistivity is found when increasing the DDR from 4.83 to 12.03. Such significant change at increased DDR was not visible for fibers with higher filler content. This indicates that drawing has a more substantial effect on the electrical resistivity at low particle contents and is more evident for higher aspect ratio MWCNTs. This is mainly due to the less interconnected nanotubes inside the fibers at lower MWCNT content, in contrast to the denser packed network in fibers with high MWCNT content. In correlation to the melt viscosity measurements, fibers whose melt viscosity were higher were less spinnable and more brittle, while they showed the lowest electrical resistivity, both of which can be attributed to the more densely packed particle network. Thus, there is a compromise between spinnability and electrical properties suitable for sensing. Interestingly, similar ρ r (and ⌀) was found for PC/MWCNT and PC/CB+MW for low DDRs and wt% above 4, but at lower melt pressure for PC/CB+MW. It is worth mentioning that it was also possible to produce fibers with different CB contents, whereby the resistivity of PC/CB 6 wt.% was still in the order of 10 6 Ω cm. Larger amounts of CB resulted in inhomogeneous and brittle fibers or even die blockage.
Summing up and based on the spinnability analysis of the composites, a more stable spinning is achieved for composites with complex viscosity and storage modulus commonly below 10 6 Pa·s and 10 5 Pa (at zero angular frequency), respectively, which corresponded to fibers with DDRs above 4 and showing during the spinning process melt pressures below 45 bar. The measured ρ r indicate that fibers with particle contents 3 wt% fabricated at DDRs below 12.03 would be most suitable for sensing applications. Thus, only fibers spun with DDRs between 4 and 12 with particle contents of 3-6 wt% were selected for liquid sensing evaluations.

The RED and liquid sensitivity
Using the methodology described in 2.3, the R rel was calculated after immersing the previously selected fibers as described in 3.2 and measuring their R measured in situ. Since it is desirable that the ΔR of the fibers can be followed by most of the standard electrometers, but the resistance increases significantly when the PC fiber is immersed in a high affinity solvent selected using HSP, a maximum resistivity of ∼48 kΩ cm was set. Taking into account the dimensions of the fibers, this value of ρ r corresponds to an electrical resistance of 10-100 MΩ* The maximum R rel values observed after the defined immersion time are used to compare the sensitivity maximum (SM) of the fibers upon solvent exposure. In the further discussions, SM will be considered as a quantity liquid sensitivity of the fibers. In addition, since the fibers were fully immersed into the solvents, the fibers' surface roughness does not play a significant role in the observed behavior, thus not affecting the values of R rel .and SM. Figure 7 shows the relative resistance change (R rel ) versus time of PC/MWCNT 6 wt% fibers immersed into different high affinity solvents, selected in accordance to their RED to PC and to their swelling behavior as discussed in section 3.1. The RED of the selected liquids varies from 3.74 to 0.71. The RED as well as the SM are included for each solvent in figure 7.
As can be seen from figure 7, there is a good correlation between the RED values and the liquid sensitivity of the PC/MWCNT 6 wt% fiber with higher REDs leading to higher SMs. For instance, PC with MEK has a RED of 0.71 leading to a very high SM above 1000%, while diethyl ether with a RED to PC of 1.16 has a much lower value of SM of ∼120%. In addition, solvents with low RED to PC caused a practically immediate high increase in Rrel, which increases rapidly until the plateau zone is reached. Furthermore, all the solvents with low RED to PC resulted in R rel plateau after few minutes of immersion. However, the time for reaching the plateau varies from solvent to solvent and the shape of the curves and the SMs for solvents with similar REDs to PC seems to be not related on the RED. The shape of the sensitivity curve of ethyl acetate and acetone, as well as toluene and xylene, are considerably different. The reason for that behavior is the different diffusion kinetics of each solvent into PC. The REDs of polymers and solvents indeed depend on the similarity of the HSPs of both parts (solvent and polymer). However, the volume expansion of the polymer comes from the diffusion kinetics which involves the occupation of empty spaces of molecules and atoms by thermodynamical means. Therefore, given that each solvent has different molecular size, but the polymer remains the same, each solvent will diffuse into the polymer differently and independently on the RED. This suggests that while HSP can predict whether the polymer will swell (slightly or strongly), it is not the only factor that plays a role in the swelling behavior of the polymer composite. In a model developed by Villmow et al it was reported that the solvent molecular size plays an important role in the different diffusion kinetics of the solvents into polymers [54]. They compared the HSPs and the solvent molecular size relation with the response of u-shaped PC/MWCNT composite films immersed in different solvents with high and moderate affinity with PC. They found that the solvent's molecular size influences the diffusion kinetics, where increasing molecule size led to decelerated diffusion kinetics. Therefore, the response speed and the shape of the R rel versus time curve are not directly correlated with the RED magnitude, but with the molecular size of the solvent, which in turn influences the diffusion kinetics. Additionally, as reported by Villmow et al, the chemical structure of the solvents plays a role in the diffusion process. Non-planar cyclic structures reduce the ability of solvents to diffuse into the polymer matrix of the composite [54]. This finding was expanded by Fan et al where the R rel versus time behavior was related to the polar component of the solubility parameter of the solvent, the solvent dielectric constant, the size of solvent molecules, and the Flory-Huggins interaction parameter between solvent and polymer [42]. This explains the differences seen in figure 7 for solvents with different shapes of the R rel curves and largely different SMs but similar REDs, such as for xylene (RED = 0.88, SM = 375%) that has non-planar cyclic structure and ethyl acetate (RED = 0.83, SM = 985%) that has a more linear planar structure.
Nevertheless, as seen in figure 7, low RED values clearly resulted in high R rel and fast reaction while low RED resulted in low R rel and slow reaction. Additionally, water had a very low SM value (<10%) with a RED = 3.74 which can be neglected compared to other values, such as that for ethyl acetate (SM ∼ 980%). Furthermore, acetone that has a relatively low RED with PC of 0.83 resulted in a SM of ∼910%, while toluene and xylene presented a SM of ∼590% and ∼370%, respectively. On the other hand, diethyl ether which has a RED with PC of 1.16, shows an SM of ∼120%, which is very low when compared to the SM obtained for solvents with a lower RED. Overall, it can be concluded that the values and differences in the SM and the shape of the R rel versus time curves for each solvent shows the great potential of PC nanocomposite fibers of being capable to distinguish between different solvents.

Liquid sensing properties of the fibers
To evaluate the effect of the content of carbon particles on the liquid sensing capabilities of the NC fibers, fibers with different amounts of MWCNT and CB+MW in PC were immersed in the same solvent, in this case butyl acetate, as shown in figure 8. Interestingly, figure 8 shows that when immersed in butyl acetate at longer immersion times (up to 1600 s) larger R rel and SM were found for the fibers with higher filler amounts. This was independent of whether MWCNT or CB+MW were added as fillers and of the DDRs used to melt-spin the fibers. This is contrary to the findings by Pötschke et al where melt-spun PLA/MWCNT fibers immersed in ethanol showed the opposite behavior from the filler content [27]. The behavior found in this work might be due to changes in the diffusivity of the solvent molecules as they pass through the filler network structure of the nanocomposites. Specifically, the increase in the liquid sensitivity upon increased filler amount could be due to the differences in the orientation of the carbon particles and the polymer chains. Figure 9 shows SEM images taken in CCI mode on samples as described in section 2.2, the arrow illustrates the fiber direction. The SEM images show a clear alignment of the carbon particles in the fiber direction, which supports the hypothesis that the observed sensing behavior is influenced by changes in the diffusivity due to filler alignment upon fiber fabrication. Furthermore, the addition of more particles results in larger remaining agglomerates as shown for other similar NC fibers, e.g. using light microscopy [22]. It is also expected that the filler alignment will be reduced with agglomerated particles compared to better dispersed particles, as individual particles are easier to align than large agglomerates.
Consequently, at higher wt% the alignment of the fillers inside the fibers is lower than at lower wt% and therefore the ability of the polymer chains to align in the drawing direction is obstructed by the presence of higher numbers or larger remaining filler agglomerates.
Thus, at higher filler contents, it is easier for the solvent molecules to reach and change the percolated network inside the fiber, resulting in higher resistance changes for fibers with higher wt%. A deeper analysis on the effect of polymer chain alignment on the R rel of the fibers will be later assessed by evaluating the liquid sensing behavior upon different DDRs ( figure 10). Interestingly, as shown in figure 8, at the same loading, PC/ CB+MW fibers have larger sensitivity and SM than the PC/MWCNT fibers. For instance, PC/MWCNT 4 wt% fibers at DDR = 4.83 have a SM of ∼ 910% while for PC/CB + MW 4 wt% the SM is ∼1370%. Similarly, for PC/ MWCNT 5 wt% fiber at DDR = 4.83 the SM is ∼990% while at the same conditions and wt% the PC/CB + MW fiber has SM of ∼ 1700%. Furthermore, among all the fibers evaluated at all the tested conditions the most sensitive fiber was PC/CB+MW 6 wt% at DDR = 4.04, reaching a very high SM of ∼3170%. The reason for PC/ CB + MW fibers to have higher liquid sensitivity than PC/MWCNT fibers might be due to the different initial resistances ρ r of both systems. PC/CB + MW fibers have higher initial resistivity values (table 3) which is due to the geometrical and conductivity differences between MWCNTs and CB. MWCNTs have a large aspect ratio (∼10 3 ) while the aspect ratio of CB is close to 1. This results in the formation of a conductive network with lower network density when using MWCNTs and CB compared to using MWCNTs alone, which also higher electrical conductivity than CB. The way in which the morphology differences influence the formation of the percolated network is formed and affect the solvent infiltration is discussed later.
Moreover, due to the already high initial electrical resistance (only slightly above the named sensing suitability limit) of the PC/CB + MW 3 wt% sample fibers at DDR = 4.04, the test could not be terminated. During the immersion, the resistance of the sample went above the measuring limit of the electrometer used. This issue occurred even if the SM is relatively low compared to the other PC/CB + MW fibers at DDR = 4.04. In order to avoid this specific issue, the sensing suitability limit was introduced as described earlier in this section. Nevertheless, it is worth noting that the resistance of this sample, before exceeding the measurement limit, had already reached an SM value of over 600%. In order to investigate the effect of drawing the nanocomposite into fibers on the liquid sensing answer, NC fibers with DDRs from 1 to 12.03 within the suitable sensing range were selected for sensing evaluations. As detailed in section 3.2, not all the fiber compositions were spinnable or had resistance values within the sensing range at all spinning conditions. Figure 10 shows the R rel versus time of PC/MWCNT 4 wt% fibers for DDRs from 4.83 to 12.03 immersed in MEK, PC/MWCNT 4 wt% for DDRs from 1 to 12.02 immersed in butyl acetate and PC/CB+MW 5 wt% for DDRs from 1 to 12.03 immersed in butyl acetate. Here, DDR = 1 stands for undrawn rods coming out from the spin die. Interestingly, figure 10 shows two distinct phenomena for both PC/MWCNT and PC/CB+MW fibers. The first one is a trend occurring when passing from the undrawn rods (DDR = 1) to the draw down fibers, where a significant decrease in SM was seen for all the fiber compositions evaluated. The second one is a decrease in SM with an increase in DDR. For instance, the SM of the PC/MWCNT 4 wt% when immersed in butyl acetate is ∼250% for the undrawn rod (DDR = 1), while the SM at DDR = 4 is ∼900%. Similarly, there is a significant increase in SM for PC/CB+MW 5 wt% immersed in butyl acetate from the undrawn rod to the DDR of 4.83 from ∼510% to ∼1700. Moreover, in all cases the liquid sensitivity is largely reduced when increasing the DDR. While the SMs for a DDR of 12.02 are ∼40%, ∼450%, and ∼720% for PC/MWCNT 4wt% (in MEK), PC/CB +MW 4 and 5 wt% (in butyl acetate), the SM for the same composites at DDR = 4.83 are ∼170%, ∼900%, and ∼1710%, respectively. Similar behavior was seen by Pötschke et al for PLA/MWCNT 3 wt% fibers where after 600 s of immersion in ethanol the fiber with larger drawing (increased take-up speed up to 50 m min −1 ) had a maximum R rel of ∼8% while lower amount of drawing (take-up = 20 m min −1 ) showed a R rel ∼26% [27].
From these results, it appears that the liquid sensitivity of the composites is significantly affected during the transition from the unstretched state to any stretching. In addition, as shown in figure 10, the undrawn filaments have a faster response to solvent exposure than the drawn fibers. The explanation for that behavior might come from the polymer chain alignment. It is known that polymer chains align themselves along the fiber axis during the fiber drawing process [55]. Additionally, as the DDR increases, the polymer chains alignment along the fiber axis also increases resulting in a highly anisotropic fiber structure. Consequently, a reduction in the diffusivity of the solvent molecules (in this case butyl acetate and MEK) into the highly anisotropic fiber structure occurs. This in turns decreases the swelling ability of the polymer, which is the underlying principle for the change in the percolation network, leading to a reduction of the liquid sensitivity in the drawn fibers. Nevertheless, is worth noting that certain polymer chain alignment in the undrawn filaments is expected due to shear stresses and high pressure before and at the capillary hole of the die. However, this alignment is significantly lower than in the draw down fibers.
Moreover, given the large difference observed in SM and R rel between the undrawn filaments and the meltspun fibers, and noting that the alignment of the MWCNTs was already observed at a DDR value of 4.83 (figure 9), it can be concluded that the alignment of the polymer chains and the MWCNTs, as well as the distancing of the carbon particles, significantly affect the liquid sensing capabilities of the fibers even at low drawing. Additionally, the solvent diffusion process is hindered by the fibers' anisotropic structure (that increases as the DDR increases), which in turns reduces the capability of the solvent molecules to modify the percolated network. Hence, drawing has a significant negative impact on the fibers' liquid sensitivity indicating an optimal DDR at which the sensitivity is maximized. Still, the liquid sensitivities found in this work are higher when compared with those reported in literature for other nanocomposite fibers using diverse approaches. For instance, Pötschke et al reported for PLA/MWCNT 2wt% a SM = 85% for undrawn extruded rods when immersed for 600 s in ethanol in ethanol [27].
On undrawn filaments of high-impact-PS/ethylene-vinyl-acetate/CB (85/15/4 phr) Narkis et al found a Δρ/ρ 0 of 180 when exposed 15 min to benzene [34]. The only paper that found R rel or SM similar to this study was by Qi et al [41]. They found a Rrel up to ∼8,000% in cellulose/MWCNT fibers immersed in water for 12 s. Nevertheless, these fibers were not fabricated by melt-mixing carbon particles with a polymer. Instead, they dip coated natural cellulose fibers with MWCNTs dispersed in a surfactant aqueous solution to evaluate the ability of the MWCNT-coated fiber as water sensor. As well, in their case water is a well-known excellent solvent for cellulose causing the fiber to highly swell and eventually dissolve. However, as they mentioned in their manuscript, it is hard to know the exact amount of CNTs on the surface of the cellulose fiber given that they repeat the dip-coating process without strict control of the MWCNT amount added at each dip-coated layer.
For an easier direct comparison of the liquid sensing behavior of PC/MWCNT and PC/CB+MW fibers, figure 11 shows the R rel over time of PC/MWCNT and PC/CB+MW fibers the same DDR and immersed in the same solvents, for contents of 4 and 5 wt% and after allowed them to dry on air at room temperature.
As visible from figure 11, the PC/CB+MW fibers clearly show larger sensitivity and faster response time than the PC/MWCNT fibers. The SM achieved for fibers with 4 wt% was ∼1350% and ∼910% for PC/CB +MW and PC/MWCNT, respectively. Furthermore, in the case of 5 wt% filler, the SM of the fibers with MWCNTs was ∼1570%, but the PC/MWCNT fiber broke during the drying process. This effect might also be correlated with the smaller diameter (due to higher drawing), as a faster R rel increase implies faster solvent diffusion, which could lead to a faster change in morphology and thus fiber failure in smaller diameter fibers. On the other hand, for fibers with CB+MW 5 wt% the maximum R rel measurable was ∼900%, before the resistance values went above the measurement range of the electrometer. The latter is the result of the combination of two factors, one is the relatively high initial resistance of the CB+MW sample as a result of elevated drawing, the other is the high affinity of PC with ethyl acetate (RED = 0.83). The combination of these two factors induces a fast increase in R rel which in turns becomes too high to be measured with any standard electrometer. This further support that for carbon particle filled PC fibers the use of low DDRs should be preferred for the implementation of such fibers in liquid sensing applications. Taking into account the results observed in figures 8, 10 and 11 it becomes clear that overall, the PC/CB +MW fibers are more sensitive to liquids than the PC/MWCNT fibers at the same wt% of filler content. This is due to a combination of factors related to how the particle network formed inside the polymer is disrupted in hybrid filler system when affected by external stimuli. As seen for strain sensing PVDF/hybrid filler based NCs, Ke et al found that MWCNTs are the main contributors for building the backbone of conductive pathways inside the polymer, while the addition of CB affects the network structure by generating more slipping or loss of contacts/connections between particles during stretching [12]. Furthermore, Li et al reported for rectangular shaped vapor sensors made of PC/hybrid fillers that adding CB to an imperfectly connected MWCNT network resulted in high sensitivity when exposed to vapors of toluene, acetone and cyclohexane [39]. To visualize this, figure 12 shows a schema illustrating the phenomena occurring during swelling in these two types of composites. Figure 12 depicts that the MWCNTs with their high aspect ratio are largely entangled before immersion, and during the solvent diffusion process occurring within the swelling polymer the nanotube network is extended, where the CB particles promote the disruption of effective pathways yet the tubes still have connecting point. Thus, many conducting pathways still remain during solvent molecules diffusion. On the other hand, due to their low aspect ratio, the CB particles combined with the MWCNTs have less probability to form effective conductive paths already in the initial dry state but also within the fiber's expanded volume once the solvent molecules diffused inside the polymer.
Moreover, during the solvent swelling process the internal dimensions of the fiber increase not only in the fiber direction but in all directions and thus favoring the displacement of the more spherical shaped CB particles. This phenomenon has a positive effect on the liquid sensing abilities of the PC/CB+MW fibers since low volume changes will cause larger CB+MW pathway interruptions resulting in higher R rel and thus, higher sensitivity compared to composites with MWCNT alone. Following this principle, it can also be confirmed that low aspect ratio particles or combinations of high-aspect ratio with low-aspect ratio fillers are particularly advantageous to be used in the fabrication of highly sensitive liquid sensing polymer composites.
Finally, according to the liquid measurements made and the swelling tests, it can be stated that for cyclic immersion-drying tests, the exposure times to the solvent should be about 5 min or less. This enables more accurate measurements and prevents polymer dissolution and extended irreversible physical changes in the fibers. When thinking in terms of sensors, it is desirable that the sensor material allows repeatable immersions with different immersion times. Thus, different immersion/drying times were selected as a way to simulate 'quick use', 'middle use' and 'prolonged use'. In addition, the fibers that showed the highest sensing abilities were immersed in solvent with a medium affinity for PC, namely butyl acetate (RED = 0.90), which ensures that the PC fibers are not dissolved and yet swell enough to cause a significant increase in R rel . Figure 13 shows the relative resistance change of PC/MWCNT and PC/, both with 6 wt% filler, at the same DDR of 4.83 upon cyclic immersions in butyl acetate at increasing immersion/drying times, starting with 5 cycles of 20 s/40 s, then 10 cycles of 180 s/180 s, and 5 cycles of 900 s/300 s. The dotted line represents R rel = 0 which serve as refence for seeing positive and negative changes in R rel . For the shortest immersion/drying cycles, a contrasting behavior for PC/MWCNT and PC/CB+MW is visible. The R rel of the PC/CB+MW fiber increases significantly after each cycle compared to the R rel of the PC/ MWCNT. The latter remains in a similar R rel range (∼30%) in the subsequent cycles showing a stable shape for the 5 immersions, however with reaching slightly negative values in the drying parts. In contrast, the R rel of the PC/CB+MW 6 wt% fiber becomes larger after each immersion (from ∼30% to ∼230%). It gets also to negative R rel values with higher intensity as compared to PC/MWCNT starting at the 3rd cycle (from −5% to −50%) after each drying. These differences between PC/MWCNT and PC/CB+MW suggest that the reaccommodation effect during the drying has a more significant impact on the percolated network structure of PC/CB+MW fibers than in the PC/MWCNT fibers. It is worth mentioning that depending on the application, such as the leakage detection of a hazardous liquid, it is not so relevant whether the relative resistance change is positive or negative. It matters more whether the sensor responds quickly showing large enough changes in resistance and that this behavior is repeatable.
Interestingly, for 20 s/40 s cycles as well as for 180 s/180 s the R rel of the PC/CB+MW fibers increases after each cycle while the maximum R rel of PC/MWCNT reduces after each cycle and they stabilize starting at the 5th cycle (in the case of 180 s/180 s) for both fibers. The reason for that behavior is the difference in the conductive network structure in both fibers. (see figure 12). In the less dense hybrid CB+MW network there is more freedom of rearrangement across the volume of the fiber resulting in a cumulative increase in R rel after each cycle possible until it reaches an equilibrium state. A similar trend was reported by Rentenberger et al on PCL+4 wt%/MWCNT PLA = (50/50 wt%) multifilament fibers, where an increase in R rel is seen after every immersion, that levels off after a few cycles [28]. They attributed this behavior to a rearrangement of the polymer chains after the re-immersion in the solvent. Contrarily, for the longest immersion/drying times, the R rel observed in figure 13 does not become predominantly negative after the first cycles, as was the case for shorter cycles. This is especially observed for the PC/CB+MW fibers. This suggests that depending on the R rel achieved at the first cycle, the subsequent cycles can cause R values lower than the initial. Moreover, for all cases with longer cycle times the R rel of the PC/CB+MW fibers remains mostly positive. In the 900 s/300 s cycles, the R rel of PC/MWCNT fibers becomes increasingly negative in each cycle.
This behavior can be due to the formation of a more compacted MWCNT network as a result of the rearrangement of clustered MWCNT remaining agglomerates during the shrinkage process upon drying of the swollen polymer. On the other hand, the not so densely formed CB+MW network is still affected by the solvent diffusion/evaporation process allowing significant changes in the R rel of the PC/CB+MW fibers after reimmersion. In summary, the combination of the significant improvement in spinnability when using CB+MW Figure 13. Relative resistance change (R rel ) of NC fibers with 6 wt% of filler upon cyclic immersions in butyl acetate at different immersion/drying times. compared to MW alone due to the lower aspect ratio and surface area of CB (easier dispersion) with the less pronounced influence of orientation of the CB particles during melt-spinning and the resulting higher liquid sensitivity of these fibers, leads to the conclusion that hybrid filler systems such as CB+MW are clearly advantageous for the fabrication of liquid sensing fibers.

Potential applications and future outlook
From the spinnability analysis and the liquid sensing performance obtained for PC/CB+MW and PC/ MWCNT, it is evident that the fibers containing the hybrid filler system show the most promising potential for their application as liquid sensor materials. Moreover, the DDR should be selected to be low and the particle content should preferably be just slightly above 3 wt%. These conditions should be fulfilled to ensure high sensitivity while enabling functionality for longer periods of solvent exposure and allowing better scalability. In line with these features and taking advantage of melt-spun NCs with mixed fillers, some concepts for the application of sensing fibers are presented.

Nanocomposite fibers as fast leakage detector
As a proof of concept for the application case of leakage detection, a contamination scenario is examined. Based on its performance, the PC/CB+MW 4 wt% fiber at DDR = 4.83 was considered to have the most favorable properties for such an evaluation.
The 'contamination agent' (5 ml of acetone; RED with PC of 0.83) was poured inside a vessel containing water (RED = 3.74) using two different amounts of water, namely 100 ml and 25 ml; both fluids are miscible. The acetone drops were applied 1 min after the start of the experiment. Given the liquid sensing behavior seen for the fibers and the swelling behavior of PC in water, it is expected that the sensor electrical signal remains unchanged in water, whereas it responds quickly when interacting with acetone. Figure 14 shows the relative resistance change of the fiber before the acetone enters until after the fiber sensor has been removed from the container and dried in air.
As expected from previous evaluations, figure 14 shows that the R rel remains stationary while the fiber is immersed in water (up to 75 s) and is subsequently able to successfully detect the 'contamination' shortly after the 'leakage'. The electrical signal of the fiber within the container with the larger water volume is much lower than the signal in the smaller container and the responding time is longer. The lower response for the smaller 'leakage' is due to higher dilution of the low solvent volume within the water volume causing that smaller amount of solvent reaches the fibers in comparison to the larger 'leakage'. On the other hand, the longer time it takes for the smaller 'leak' to be detected is due to the fact that the smaller amount of solvent takes longer to run down the water and reach the fiber. The latter is a phenomenon unrelated to the speed response of the fiber, but to the small volume of the added droplets. Nevertheless, the R rel signal of the 5 ml of acetone in 100 ml water is as high as 100%. Thus, the liquid sensing fiber could successfully fulfil the tasks of leakage detection and sensing differences in the amount of the contamination. Moreover, the electrical resistance values measured for these PC/CB+MW fibers is in the range of kΩ and can be easily measured with any standard portable electrometer. It should be noted that for the implemenation of such sensors in uncontrolled escenarios other external stimili such as strain, pressure and temperature must be taken into account, as these could cause changes in R. Nevertheless, if the influence of these stimuli is known, its contribution can be eliminated and/or corrected by electronic or computing means (e.g. temperature compensation circuits or numerical calibration). For instance, fast calibration can be done on-site by using a graphical user interface at which an already trained microcontroller with reference data is responsible for detecting and/or correcting the proper electrical signal/ behavior of known solvents.

Implementing new fiber technologies towards liquid sensing
In the following, ways to improve the liquid sensing behavior of nanocomposite fibers using new fiber manufacturing technologies are presented. In recent years, the fabrication of sheath/core bicomponent fibers has been reported, which are used as functional fibers [56,57]. In this kind of fibers, the sheath (or the core), possess a desired function, while the other component of the fiber has a different property (e.g. a second desired function or simply an aid in the spinning process). To obtain functional Bi-Component (BiCo) fibers, a double piston spinning device with communicating double chambers was used, which was developed and built at the IPF facilities. Figure 15 shows a schematic of the BiCo fiber spinning device and images of a cross-section and transversal-section of the obtained fibers. Based on their melt properties, spinnability, and the sensing behavior of its subsequent fibers, PC/CB+MW 4 wt% was used as sheath component, while neat PC was used as core.
The conditions for BiCo fiber fabrication were obtained after several trials aimed on obtaining fibers with an electrical resistance suitable for liquid sensing, based on previous results with single component fibers. According to such pre-investigations, the best sheath component was PC/CB+MW 4 wt%, while the core was neat PC. The throughputs of the sheath (V S ) and core (V C ) were both equal to 1.50 cm 3 min −1 , while the take-up speed was adjusted in order to obtain the comparable DDR of 4.83, as for the previously fabricated single component fibers.
Employing the same methodology described in sections 2.3 and 3.4, the liquid sensing abilities of BiCo fibers was tested. Figure 16 shows the R rel of BiCo fibers with a CB+MW filler content of 4 wt% in the sheath compared to those of PC/MWCNT and PC/CB+MW, immersed in butyl acetate (RED with PC of 0.90). As visible from figure 16, from the three kinds of fibers, the BiCo fiber has a larger sensitivity to the solvent tested. In addition, in contrast to the single component fibers the BiCo fiber has a much faster response and is also the fastest to achieve its SM. This outstanding performance of the BiCo fiber can be due to the fact that unlike single component fibers, all the carbon particles are spread as well as restricted in the sheath which is only few μm thick and is the external layer of the BiCo fiber. Therefore, the part of the NC fiber responsible for the sensing abilities, this is the PC/CB+MW sheath, is closer to the fiber's outer surface which is where the initial solvent diffusion starts and also occurs faster than innermost the fiber.
Moreover, the shoulders visible in the BiCo response are possible due to damages occurring in different sheath sections during solvent diffusion passing through the thin sheath. This could be solved by increasing the diameter of the outer die hole and the thickness of the sheath layer. In addition, as it was reported by Villmow et al, there are two main phenomena occurring in CNP filled polymer composites. One is solvent diffusion into the bulk material and the other is the wet-skin/dry-core morphology formation [38].
In our case, the electrically conductive skin (in this case the CNP filled sheath) gets in the wet phase rapidly while the dry core remains unaffected. This in turns causes that changes in electrical resistance occur more abruptly. This gives the BiCo fibers an advantage on the liquid sensing response over the single component fibers, which is visible by its superior sensitivity. It is worth mentioning that apart from the findings of this work, there is no report yet on the use of a hybrid CB+MW filled polymers as bi-component fibers for their use as liquid sensors.

Conclusions
In this work, a hybrid filler system consisting of Multiwall Carbon Nanotubes (MWCNT) and carbon Black (CB) was incorporated in Polycarbonate (PC) by melt mixing and nanocomposite fibers with liquid sensing capabilities were successfully fabricated, characterized and implemented as sensors. Based on swelling measurements and liquid sensing evaluations, it has been demonstrated that knowing the Relative Energy Difference (RED) from the Hansen Solubility Parameters (HSPs) between polymer and solvents is a helpful tool for selecting a polymer that can be used as sensitive liquid sensor while ensuring high sensitivity. After a thorough spinnability analysis varying a number of processing conditions, it was found that improved spinnability is achieved when the filler content was below 6 wt% and the Draw Dawn Ratio (DDR) was above 4. However, increasing the DDR led to a significant reduction in electrical resistivity, so that the fibers could no longer be used as sensors. Furthermore, a compromise must be met between the spinnability and the desired electrical properties required for the use of the fibers as liquid sensing materials. In addition, among the well spinnable fibers, the fibers that showed electric resistivity values suitable for liquid sensing were those produced with a DDR below 12.
Although according to the sensing characterizations there are more factors than just the HSPs/REDs are involved in the behavior of the sensor response, it has been demonstrated that it is possible to predict whether the nanocomposite fiber will be highly sensitive to selected solvents based on HSPs and their REDs with a particular polymer. Yet also, the drawing of the fibers had a very significant effect in the sensing abilities, with fibers spun with higher DDRs having lower sensitivity. This is possibly related to the high anisotropy in the orientation of the polymer chains and fillers, which is more pronounced in fibers with only MWCNTs. Fibers produced at 4 < DDR < 12 with filler contents of 3 < wt% < 6 that contained a mix of CB and MWCNTs showed the most promising liquid sensing abilities. The nanocomposite sensing fibers with the best performance were tested as liquid sensor elements in a simulated water contamination scenario where the sensor immediately detected the presence of the pollutant. In addition, a novel approach to manufacture functional fibers by sheath/core Bi-Component (BiCo) fibers using the PC/CB+MW nanocomposite as sheath and neat PC as core was implemented. Surprisingly, the BiCo fibers showed superior liquid sensing response among the different kinds of fibers tested. This behavior is due to the carbon particles being spread along the fiber axis while at the same time restricted within the cross section of the thin sheath depth. Finally, this work has shown that this kind of fiber sensors are a promising prospect for alternative liquid sensors, while also offering the possibility for liquid-sensitive textiles and wearables.