Improved printability and performance of 3D printed hybrid thermoplastic polyurethane nanocomposite strain sensors

Abstract

Fused Deposition Modeling (FDM) has promised to revolutionize the fabrication of strain sensors, offering design flexibility and cost-effectiveness. However, printing flexible thermoplastic elastomers with conductive nanomaterials presents challenges, such as nozzle blockages and inconsistent feeding due to high melt viscosity and filler agglomeration. In this study, hybrid thermoplastic polyurethane (TPU) nanocomposites filaments containing carbon nanotubes (CNT), graphene nanoplatelets (GNP), and boron nitride (BN), were developed. The hybrid fillers reduced the filament extrusion force and improved the printability compared to CNT-only composites, while maintaining high electrical conductivity and strain-sensing capabilities. A novel extrusion force testing method was implemented to measure the printability of the nanocomposite filaments. The printed sensors, prepared via FDM, demonstrated an enhanced strain-sensing range and mechanical durability, with TPU-CNT-BN composites achieving sensing up to 250% strain. These findings highlight the potential of hybrid nanocomposites for reliable and scalable production of flexible strain sensors using FDM, offering applications in wearable electronics and structural health monitoring.

Keywords

FDM 3D printing strain sensing carbon nanotubes boron nitride

Introduction

Over the last decade, Fused Deposition Modeling (FDM) has emerged as the most used additive manufacturing technology, transforming the prototyping and production landscape. Main advantages include its low cost, relative high speed and reproducibility.^1,2 The layer-wise deposition allows for a high degree of design freedom and the creation of complex 3D structures, also containing different materials with different physical properties. This makes FDM an attractive technique for fabricating multifunctional components, including strain sensors.^3,4 The fabrication and printing of conductive filaments provides the manufacturers with the freedom to create sensors tailored to a specific application. However, FDM faces a number of significant challenges that prevents reliable printing of conductive materials.⁵ The layered deposition can introduce undesired anisotropies in the physical properties of the printed object,⁶ potentially influencing the sensor's response. The filaments melt properties directly influence the ease of extrusion along with the accumulation of conductive nanomaterials in the nozzle which can cause blockages, completely disrupting the printing progress. In addition, the FDM process may impart structural defects,⁷ which can lead to variations in the sensor's performance. Finally, achieving consistent conductivity and ensuring seamless integration of the sensing element within the printed matrix can be challenging, requiring careful optimization of both printing materials and the printing parameters.

Especially when printing thermoplastic elastomers, such as in the present work, extra care must be taken to ensure consistent feeding of polymer into the nozzle. While rigid polymers can be easily griped and fed via a feeding gear, flexible materials face challenges such as bending, slipping and kinking, which ultimately result in inconsistent feeding. To overcome this challenge, filament manufacturers suggest a reduced print speed, as much as 6 times lower than their rigid counterparts (e.g., poly(lactic acid)).

The FDM printing of conductive nanocomposites further extends the opportunities arising from 3D printing, compared to printing conventional flexible filaments, due to issues such as blockages, inconsistencies in the way agglomerated fillers can cause buildups of pressure resulting in sudden releases of resulting pressure causing inconsistent printing layers.

Rheological testing is normally the most suitable characterization method to determine the viscosity/printability of a polymer. The introduction of a nanofiller into a polymer-based filament can significantly modify the rheological properties and add complexity to the printing process. The interplay of nozzle geometry and potential agglomerations at the nozzle add further complications, making a straightforward comparison of viscosity and flow challenging. This limitation on rheological testing was noticed particularly when evaluating the printability of flexible nanocomposite filaments. Unlike conventional rigid filaments, flexible nanocomposite filaments face the limitation they cannot be easily gripped while extruded. In addition, high filler agglomeration/clogging takes place at the end of the nozzle. Filler agglomeration leading to clogging originates from fillers bonding to the inner walls of the nozzle during retractions and building up into larger agglomerated platelets over multiple retract and extrusion cycles (Fig. 1(c)). Although extruder design significantly influences the gripping force exerted on the filament, dual-gear direct drive extruders have successfully addressed many limitations, with increased gripping over conventional single drive systems on flexible materials (Fig. 1(b)). As a result, current challenges are predominantly attributed to material limitations.

Figure 1.

Illustration of: a. Direct drive FDM filament feeding setup with a short low-friction filament path, Bowden drive setup with a long filament path. | b. Gear contact with filament for different gear sizes where a larger drive gear diameter leads to a grater contact patch. | c. Filler agglomeration within nozzle for highly loaded thermoplastic composite filaments.

To address this challenge, we present testing apparatus for measuring the force required to extrude material from a nozzle at a range of flow rates, by coupling it with a loadcell commonly used in universal testing machines. The testing jig utilises commercially available FDM printer components along with mounts fabricated in-house to ensure the filament extrusion force is recorded directly from an in-line connected loadcell. The whole system is then controlled by a custom-Klipper interface.

The polymer-based filaments were functionalised with a combination of three nanofillers: carbon nanotubes (CNT), graphene nano platelets (GNP) and boron nitride (BN). Hybrid fillers combine two or more fillers with unique properties to aid towards the endowment of multifunctional properties,^8–13 while circumventing some of the negative impacts of using a single filler at higher loadings, both the processability and the sensing characteristics of a composite. A combination of conductive fillers can be used to create several conductive subnetworks within a matrix leading to a large sensing range. Interestingly, previous literature has revealed that combining conductive and non-conductive particles in a ternary composite increases conductivity at the same ultimate loading of conductive particles.^14,15 This conductive/non-conductive percolation mechanism is often explained by the volume exclusion theory, where volume occupied by a non-conductive filler (such as BN or clays) acts to densify the conductive network.^16,17 In an ideal filler distribution, the conductive and non-conductive nanoparticles would form their own interlinked subnetworks within the matrix leading to a highly sensitive conductive network at low loadings. BN was specifically selected as a non-conductive co-filler because its low-aspect-ratio particles are expected to increase melt viscosity far less than fibrous (CNT) or platelet (GNP) fillers at equivalent loading¹⁸; a critical consideration for FDM feedstock, where excessive melt viscosity causes nozzle back-pressure, clogging, and feeding instability. Additionally, the high intrinsic thermal conductivity of BN (∼250–300 W/mK)¹⁹ was anticipated to improve heat transfer within the melt zone of the FDM nozzle, reducing the risk of incomplete melting at higher feed rates. GNP was selected as a second hybrid candidate to examine whether a 2D platelet filler, which unlike BN, contributes to electrical conductivity via enhanced network formation, could offer a different balance of processability and sensing performance.^20,21 By fixing total filler loading at 5 wt% and 10 wt% across all hybrid compositions, direct comparison of processability and sensing response was made possible while remaining within the practically printable range identified in prior FDM nanocomposite studies.

Nanomaterials introduced into a flexible thermoplastic offer many avenues to produce high performance strain sensors.^22–24 TPU-CNT strain sensors have been found to display exceptional electrical and mechanical properties, emerging as one of the most promising materials for high strain (<100%) sensing applications.^25,26 Strain sensors can measure the changes exerted onto their physical dimensions under various mechanical stresses and convert this into an electrical sensor that can be picked up using simple electronics such as a voltage divider circuit. They are pivotal in a range of applications, including: wearable health monitoring, user input devices and structural health monitoring in civil engineering or transport.^24,27–29 The inherent sensitivity of TPU-CNT composites to deformation translates into real-time, accurate strain detection, making these composites an attractive choice for strain sensor fabrication. The filler loadings investigated in this study were selected based on the reported percolation threshold of CNT in TPU matrices, which typically occurs in the range of 1–3 wt% under melt-mixing conditions.³⁰ Loadings of 5 wt% and 10 wt% CNT were therefore chosen to ensure conductivity well above the percolation threshold, targeting the electrical conductivity range of 0.1–1 S/m required for voltage-divider-based strain sensing circuits. The 5 wt% total loading for hybrid composites (e.g., 2.5 wt% CNT + 2.5 wt% BN or GNP) was selected to maintain CNT content above percolation while simultaneously introducing a secondary filler at equal proportion, consistent with prior studies demonstrating enhanced excluded-volume effects at symmetric binary filler ratios.³¹

While FDM printing presents an intriguing path for crafting customized strain sensors, material constraints hinder the creation of consistently reproducible sensors. Although TPU-CNT sensors exhibit potential with sensing ranges extending up to 250%,²³ the current limitation lies in material processability, restricting these sensors to small, singular components, typically printed within controlled laboratory settings.³² This work aims to demonstrate how hybrid nanofillers can be utilized to fabricate reliable FDM strain sensors by exploiting hybrid fillers to reduce filament extrusion force, which in turn increases print reliability. Our novel statistically driven method of measuring the printability of flexible thermoplastic elastomers offers multiple advantages over rheological methods (in-situ or ex-situ). The FDM filaments produced herein further advance the understanding of how hybrid filler structures can help increase printability while maintaining electrical conductivity/sensitivity. The ultimate printed sensors were characterized using mechanical testing, alongside in situ electrical measurements in various strain sensing setups. The target performance criteria guiding formulation selection were: (i) electrical conductivity ≥0.1 S/m to enable measurement via a simple voltage-divider circuit, (ii) strain at break >150% to accommodate wearable and structural health monitoring applications, and (iii) an extrusion force compatible with the dual-drive gear system of the Ender-3 S1 printer at a feed rate of ≥3 mm/s. These criteria collectively constrain the accessible composition space and justify the selection of total filler loadings between 5 and 10 wt%, with CNT content maintained above the percolation threshold in all formulations.

Materials and methods

Materials

Thermoplastic polyurethane (TPU), Lubrizol MAXI 12T85E, a polyester based TPU with shore A hardness of 86 and density of 1.21 g/cm³. Multiwalled carbon nanotubes (CNT), NC7000™, produced via catalytic chemical vapor deposition (CCVD) with an average length of 1.5 µm and average diameter of 9.5 nm were procured from Nanocyl. Graphene nanoplatelets (GNP) AVA240 were produced by Avanzare.³³ Boron nitride (BN) powder, with average diameter of ∼1 μm was purchased from Sigma-Aldrich.³⁴

Methods

Polymer compounding

Melt compounding provides an eco-friendly and solvent-free method for mixing nanocomposite materials within a thermoplastic matrix. Initial masterbatches of CNT and GNP in TPU were prepared at a loading of 20 wt% using a twin-screw compounder (Colin ZK25, Germany), set to 150 rpm, a barrel temperature from 165°C (feeding zone) to 180°C (extrusion zone), gravitationally feeding pre-dried materials (overnight at 60°C). Subsequently, these masterbatches were diluted with unmodified TPU to achieve the desired weight percentages suitable for FDM printing using a micro compounder (Xplore 15 HT, Netherlands), set to 180°C with a 100 RPM screw speed.

Filament production

The blended TPU compounds were subsequently processed with a micro compounder (Xplore 15 HT) set up in filament making mode to produce filaments of 1.75 mm for FDM printing. The polymer melt was continuously extruded from a 2 mm nozzle, before cooled by a set of blower fans. An in-house proportional integral derivative (PID) controlled filament winding station was used to ensure filament uniformity; filaments were 1.75 ± 0.06 mm in diameter.

FDM 3D printing

The Ender 3 S1 FDM 3D printer was loaded with pre-generated geometry code (G-code) to produce test samples, according to the parameters listed in Table 1. The G-code for each sample type was exported once and executed from the SD card during each print run, ensuring that the printed output remained unaffected by any inadvertent slicer changes. The Ender 3 S1 printer was chosen as a low-cost FDM printer, with dual drive gears that have the ability to tightly grip flexible filaments.³⁵

A 1.4 mm diameter nozzle was used in place of standard, lower diameter nozzles to reduce the risk of clogging caused by filler agglomeration at the nozzle orifice, and to lower the melt back-pressure during extrusion of the high-viscosity nanocomposite filaments. This enlarged nozzle diameter is consistent with the approach used in the extrusion force testing described in Section 2.2.4, ensuring that printability data from both setups are directly comparable.

FDM samples production

Strain sensors were printed with deposition lines in line with the direction of strain, while the concentric layers and print direction work to evenly distribute strain throughout the sensor. The printed samples were designed (Fig. 2(a)) to reduce blockages and increase printing reliability. 2 sensors were printed at once and then cut (red lines) to 100 mm before fine copper mesh is fused in at 60 mm spacing (indicated in blue). Mechanical testing utilized ISO37 Type dumbbell specimens for the mechanical evaluation of thermoplastic rubbers; the specimens were printed on a FDM printer (Creality Ender 3 S1, China).

Figure 2.

a. Illustration of sensor designed to be printed in continuous extrusion ‘vase’ mode. | b. Schematic of the Extrusion Force Testing setup, showing the filament path (in red) along with custom mounts and commercially available printer components. | c. Image of extrusion force tester being externally controlled by Raspberry Pi running Klipper.

Extrusion force testing

For the determination of the printability of our nanocomposites we manufactured a data driven assembly equipment that was able to measure extrusion force, while controlling feed rate (material flow) and temperature. This was achieved by placing the extruder (cold end) and nozzle/heat block (hot end) on independent mountings, affixing them inline on a universal testing machine. The extruder used was a dual gear all metal extruder (Vz-HextrudORT),³⁶ with a large amount of range in its tensioning clamp (2.1 to 1.3 mm) to provide maximum gripping strength. A main board, coupled with a Raspberry Pi operating Klipper, was employed for controlling and recording data. Since the setup utilised readily available 3D printer components, there is flexibility for future upgrades to evaluate advancements, such as integrating Bondtech CHT nozzles. Between the hot and cold end assemblies presented in Fig. 2 an PTFE coupling can be seen, which gives the filament a low friction path while providing full force transfer to the loadcell. During testing, a 1.4 mm nozzle was used at a range of flow rates while the load cell was stationary in the universal testing machine (Instron 68TM-10).

Results and discussion

Printability of hybrid filaments

The force needed for filament extrusion is a critical consideration when choosing an FDM material for larger prints. This is because the consistency of the extruded bead significantly impacts the overall quality of the printed part, and defects in layer deposition can create stress concentrations, potentially causing premature failure due to crack propagation. In Fig. 3(a) filaments are extruded at 1 mm/s and 240°C. Extrusion starts at 1.5 s where all samples experience a sudden increase in extrusion force as fresh un-melted filament is forced into the print head. As the material begins to melt, the initial force/ pressure builds up and then the material is forced though the nozzle, where a steady build up in extrusion force is observed. A pronounced increase in print force is observed (at 4–6 s) before reaching a plateau value (6 s onwards). This plateau is attributed to steady state extrusion, where extrusion force and back pressure reach an equilibrium creating a consistent polymer bead to be extruded. Figure 3(a) indicates that TPU-CNT5 has the highest initial back pressure of the 3 materials at a total filler loading of 5 wt%. The large initial back pressure observed, indicates that the stiff TPU-CNT5 filament can exert a large degree of force on the printhead before melting, whereas the TPU-CNT2.5-BN2.5 filament displays a smaller initial peak that quickly subsides as the filament is melted. The hybrid TPU-CNT-BN filaments produced herein, required the lowest extrusion force compared to TPU-CNT and TPU-CNT-GNP filaments at the same ultimate loadings, both at a large range of temperatures (Fig. 3(b)) and at a range of extrusion speeds. It was found that at print temperatures between 200 and 250°C, the TPU-CNT-BN hybrids required ∼10% less extrusion force than TPU-CNT-GNP filaments and ∼20% less force than CNT only filaments, while maintaining similar conductivity levels as the TPU-CNT composites (Fig. 3(a)). Adding 5 wt% BN to a sample with 5 wt% CNT (Fig. 3(c)) showed that the presence of BN increased the print force by 34% at standardised print parameters (240°C) while increasing conductivity of the matrix by 9 times. The small increase in TPU-CNT5-BN5 extrusion force over TPU-CNT5 extrusion force is due to the increased melt viscosity of the additional 5 wt% of BN, hoverer the BN's contribution to localising conductive pathways and significantly increasing in conductivity due to excluded volume theory makes the TPU-CNT-BN composites an excellent choice where processability and conductivity are desirable material characteristics such as FDM printed strain sensors.

Figure 3.

Extrusion force data for TPU composite filaments extruded with 1.4 mm nozzle. a) Representative time vs extrusion force for composite filaments TPU-CNT5, TPU-CNT2.5-GNP2.5 and TPU-CNT2.5-GNP2.5 extruded at 240°C and feed rate of 1 mms⁻¹ sample. b) Extrusion force vs temperature data for samples at a total filler 5 wt% and feed rate of 1 mms⁻¹. c) Force temperature comparison for TPU-CNT5 vs TPU-CNT5-BN5 at feed rate of 1 mms⁻¹. d) Extrusion force vs feed rate data for composite filament extruded at 240°C.

Printability tests (Fig. 3(d)) showed a clear pattern of increasing extrusion force with increased feed rate, until a maximum feed rate was reached. At this point, the back-pressure from the hot end caused the extruder to fail to consistently feed the filament, as the filament in contact with the melt zone (indicated in Fig. 2(b)) had less time for heat to effectively transfer, resulting in an inconsistent melt and extruded bead. Tests showed that the highly loaded TPU-CNT5 and TPU-CNT2.5-GNP2.5 composites could not be consistently extruded at feed rates above 3 mm/s, due to filler agglomeration in the hot end, which caused high back pressure. In contrast, TPU-CNT2.5-BN2.5 was successfully extruded at the same total loading up to 4 mm/s. This improved performance is likely due to the low aspect ratio of BN particles, which do not reinforce the matrix as much and thus do not significantly increase viscosity. Additionally, the BN particles may aid in temperature transfer throughout the filament due to their high thermal conductivity. The aspect ratio shape appears to play a key role as feed rates increase, with TPU-CNT-BN filaments consistently requiring the lowest extrusion forces among the three composite types. TPU-CNT5 exhibited a significant rise in extrusion force, likely due to the high melt viscosity caused by the entangled CNTs. In comparison, the combination of 2D GNPs and 1D CNTs in TPU-CNT-GNP led to higher extrusion forces than in TPU-CNT-BN, likely because the GNP-CNT mix forms a rigid supporting filler network within the matrix.

Mechanical properties of FDM printed sensors

The modulus of FDM-printed TPU nanocomposites varies depending on the type of filler and its loading (Fig. 4(a)). TPU-CNT-BN hybrids show minimal change in modulus between 10 wt% and 5 wt% total filler loading, with a difference of less than 2.3%. In contrast, the addition of 2D GNP significantly increases stiffness, with the TPU-CNT-GNP composites showing a 24.6% rise in modulus from TPU-CNT2.5-GNP2.5 to TPU-CNT5-GNP5. Increasing the CNT content in TPU-CNT composites from 5 wt% to 10 wt% leads to TPU-CNT10 failing to reach 100% strain (Fig. 4(b)) before breaking, resulting in no measurable modulus at 100%. This is likely due to the high degree of reinforcement provided by CNTs when melt-mixed into the thermoplastic elastomer, producing a very stiff, non-elastic composite. The strain at break for both hybrid and mono-filler composites is considerably lower than that of unreinforced TPU, which reaches 493%. The highest-performing conductive nanocomposite is TPU-CNT2.5-GNP2.5, with a strain at break of 360%, representing a 97.8% improvement over TPU-CNT5, which breaks at 182%. At 5 wt% total filler, Fig. 4(b) shows a clear trend of increasing strain at break, progressing from TPU-CNT to TPU-CNT-BN, and finally to TPU-CNT-GNP, with TPU-CNT-GNP exhibiting the highest strain at break. At 10 wt% total filler, both BN and GNP hybrids fail within 3% of each other but outperform TPU-CNT10, which breaks at 64% strain. This is likely due to the high entanglement of CNTs, leading to filler agglomeration and stress concentration. It remains unclear whether the low loading of GNP helps act as a lubricant or disperses CNT aggregates during compounding, thereby reducing stress concentration in TPU-CNT-GNP hybrids. The 2D geometry and the low aspect ratio of BN particles likely localizes CNT aggregates but contributes minimally to the mechanical properties, as reflected by the lower strain at break in CNT-BN hybrids compared to CNT-GNP hybrids at 5 wt%.

Figure 4.

Mechanical properties for different FDM printed specimen samples: a) Youngs modulus, b) Strain at break, c) Ultimate tensile strength. Results from virgin printed TPU indicated by red line with shaded region as measured uncertainty. d) Electrical conductivity measured by 2-Point Probe.

When evaluating a sensor's performance, the required properties depend on the specific application. For structural health monitoring using this FDM sensor, the target is a high strain at break (>200%) paired with an electrical conductivity of around 0.1 S/m, as this lower bound level of conductivity supports measurements through simple voltage divider circuits. The ultimate tensile strength (UTS) of the tested thermoplastic nanocomposites is lower than that of raw FDM-printed TPU, which measures at 28.1 MPa (Fig. 4(c)). At 5 wt%, TPU-CNT-GNP exhibits the highest UTS due to superior load transfer and reinforcement provided by GNPs, while TPU-CNT-BN shows intermediate UTS, likely due to BN's role in reducing CNT agglomeration. TPU-CNT demonstrates the lowest UTS, possibly because of CNT agglomeration. At 10 wt%, all composites show a notable decline in UTS, with values within 23% of each other, likely due to increased filler agglomeration from high CNT loading, which introduces stress concentrators that weaken tensile strength.

Filler type and loading have a significant impact on the electrical conductivity of TPU FDM-printed nanocomposites (Fig. 4(d)). At 5 wt% total filler loading, TPU-CNT5 has the highest conductivity, followed by TPU-CNT2.5-BN2.5, while TPU-CNT2.5-GNP2.5 shows significantly lower conductivity. This suggests that CNTs provide better conductive pathways, whereas the presence of BN and GNP likely interferes with CNT network formation. CNT fillers appear to disrupt the filler network more than BN, as seen in the UTS results (Fig. 4(c)), where BN helps evenly disperse CNTs. At 10 wt%, TPU-CNT10 again shows the highest conductivity, closely followed by TPU-CNT5-GNP5, with TPU-CNT5-BN5 exhibiting conductivity similar to TPU-CNT5, indicating similar network formation in both cases. The increased conductivity of TPU-CNT-GNP at higher loadings may be attributed to enhanced network formation as GNP content rises, while the lower conductivity of TPU-CNT-BN composites is due to BN's non-conductive nature. Overall, CNT dominance in conductivity is evident, with hybrid fillers producing varying effects depending on their composition and loading. For structural health monitoring, the goal for this FDM sensor is a strain at break greater than 150% and an electrical conductivity above 0.1 S/m to enable measurements with voltage divider circuits. Based on these parameters, CNT-only samples fall short of the required mechanical properties.

Sensor performance

Results showed that the resistance of the printed nanocomposite sensors increased exponentially with strain (Fig. 5), making an exponential trend the most suitable for calibration, as highlighted by Tomes et al.²⁶ and demonstrated by O'Mara et al. for high strains.^26,37 This system fits well with exponential response to strain over most of the conducting strain range for FDM printed strain sensors. However, while the two previous studies have applied exponential models to study the electromechanical response of similar silicon-based polymer nanocomposite systems, we are applying this modelling approach to a thermoplastic FDM printed system. The ‘sensing limit’ reported herein is defined operationally as the strain at which the sensor's resistance–strain response departs irrecoverably from the exponential calibration model (Equation 1). In practice, two distinct failure modes were observed depending on the composite formulation. For CNT-only samples (TPU-CNT5), the sensing limit was governed by irreversible plastic deformation of the printed gauge: upon unloading from high strains, the baseline resistance did not return to R₀, indicating permanent rearrangement of the conductive network. For TPU-CNT5-GNP5 and other high-resistance composites, the sensing limit was additionally constrained by the measurement range of the 2-wire resistance meter (∼1 GΩ), beyond which resistance readings became unreliable due to the instrument switching between internal reference resistors. In both cases, the sensing limit is identified from the R–ε data as the strain beyond which the measured response either saturates, deviates by more than 10% from the exponential fit, or cannot be recovered upon unloading. This criterion was applied consistently across all samples presented in Fig. 5.

Figure 5.

Strain sensing data for TPU nanocomposite sensors including gauge factors calculated from the exponential model (Equation 1) and the strain-induced sensing limit.

The exponential gauge factor in flexible nanocomposite strain sensors can be expressed as:

R = R_{0} e^{G_{\exp} ε}

(1)

where G_exp is the gauge factor given by a regression fitting, R is resistance, R₀ is initial resistance and ε is strain.

The TPU-CNT5 printed sensors exhibited a notable sensing range from 0 to 318% strain, with a resistance change of 10⁴ and a linearity of R = 0.97. However, due to plastic deformation during sensing, they were only suitable for single-use at such high strain levels. Introducing BN into the matrix alongside CNT in the TPU-CNT5-BN5 sensor extended the sensing range to 250%, with a resistance change of 10⁵ and strong adherence to the calibration curve (R = 0.99). The broad sensing range and significant resistance change of the CNT-BN hybrids enabled the use of a simple voltage divider circuit and microcontroller for both calibration and strain monitoring. These properties make them ideal for wearable applications, where high strain range and sensor flexibility are essential. In contrast, TPU-CNT5-GNP5 and other samples displayed slight deviations from the line of best fit, likely due to systematic errors caused by the 2-wire resistance meter switching between internal reference resistors as the resistance increased from approximately 10 kOhms to 1 GOhm.

Static material evaluation

A radar plot (Fig. 6) is used to compare the measured properties discussed in Sections 3.1, 3.2 and 3.3. Compliance is scaled as the inverse of the modulus (Section 3.2), such that a lower stiffness corresponds to a more compliant sensor. Printability is scaled as the inverse of the measured extrusion force reported in Section 3.1. Sensitivity is scaled from the gauge factor (Gexp). Initial conductivity and sensing range are scaled from the conductivity (Section 3.2) and sensing limit (Section 3.3), respectively. Subjective analysis indicates that the three overall best-performing materials are those with a total loading of 5 wt%. Among these, TPU-CNT2.5 BN2.5 (solid red) emerges as the most suitable material for further study, demonstrating both excellent printability and strong sensing characteristics, while maintaining acceptable levels of initial conductivity and sensitivity.

Figure 6.

Radar plot of summary properties for FDM printed strain sensing materials. The aggregated data has been scaled between 1 and 5 for illustration/ comparison purposes as discussed in section 3.4.

Cyclic sensing performance

Cyclic testing was performed on a range of specimens at strains between 20% and 50%, simulating wearable sensors for healthcare applications, where strain levels are typically low. Tests were conducted on a universal testing machine (Instron 68TM-10) in displacement-control mode. A sinusoidal displacement waveform was applied at 0.008 Hz (i.e., a full cycle period of 125 s), cycling between 20% and 50% strain for 10 consecutive cycles. For the un-printed 1.75 mm filament specimens, a 20% pre-strain was applied prior to cyclic testing to remove the initial slack in the filament and ensure contact with the grips; no pre-strain was applied to the FDM-printed sensor specimens, as these were self-supporting. No hold time was imposed at the strain limits. Resistance was logged continuously throughout testing using an Keysight 970A DAQ in measuring using a 2-wire resistance configuration. FDM-printed sensors under cyclic loading exhibited a phenomenon known as double peaking, where the change in resistance was out of phase (Fig. 7(a)) and showed an unconventional resistance-strain response (Fig. 7(b)). After several cycles, the TPU-CNT2.5-BN2.5 FDM-printed sensor (Fig. 7(a) and 7(b)) displayed an unexpected sensing curve, where the resistance change was not proportional to the applied strain. To investigate further, a single extruded 1.75 mm filament of TPU-CNT2.5-BN2.5 (Fig. 7(c) and (d)) was tested, showing a more consistent sensing curve with uniform resistance changes between 20% and 50% strain, with a 20% pre-strain. The double peaking observed in the FDM sample is likely due to the intrinsic geometric characteristics of FDM-printed parts, such as voids, layer lines, and print beads.³⁸ Additional tests conducted on the 1.75 mm filament, as shown in Fig. 7(c), revealed that the same material, when not processed via FDM printing, did not exhibit signs of double peaking. This further supports the conclusion that the double peaking effect originates from the fibrillar structure inherent to FDM-printed parts. Figure 7(d) illustrates that the 1.75 mm filament, tested between 20% and 50% strain over 10 cycles, exhibited a higher R/R₀ ratio than the FDM-printed sensor. This was accompanied by a greater change in resistance, which is likely due to the filament's lower initial resistance.

Figure 7.

TPU-CNT2.5-BN2.5 sensor performance during cyclic testing at rate of 0.008 Hz. a) Time, Strain and Change in Resistance (R/R₀) graph for 10 cycles of the FDM printed sensor. b) Change in Resistance (R/R₀), strain graph showing sensor evolution of FDM printed sensor. c) Time, Strain and Change in Resistance (R/R₀) graph for 10 cycles for 1.75 mm filament. d) Change in Resistance (R/R₀), strain graph showing sensor evolution for 1.75 mm filament.

The double peaking phenomenon shown in Fig. 7(b) can be attributed to the formation of 1) irregular voids intrinsic to the inner structure of FDM prints, 2) the conductive and non-conductive fillers having different contributions when strained and 3) the effect of filler orientation within the printed structure.

The R/R₀–strain plots in Figs. 7(b) and 7(d) also reveal pronounced electromechanical hysteresis in both the FDM-printed sensor and the bare filament: the resistance measured during loading consistently differs from that measured at the same strain during unloading, producing the enclosed loop visible in both plots. This hysteresis arises because the rearrangement of the CNT conductive network under strain is not fully reversible within a single cycle: CNT junctions that are broken during extension do not immediately reform upon unloading, particularly at the relatively fast loading rate used here. Hysteresis of this type is a known limitation of piezoresistive nanocomposite sensors^39,40 and introduces a systematic error if a single-valued calibration curve (such as Equation 1, derived from monotonic loading) is applied to dynamic measurements. The area enclosed by the hysteresis loop in Fig. 7(d) (bare filament) is visually smaller than that in Fig. 7(b) (FDM sensor), suggesting that the additional structural complexity of the FDM geometry, specifically the triangular voids and layer interfaces identified in Figs. 8(c) and 8(d), amplifies hysteretic behaviour relative to the bulk filament. Future work should quantify the hysteresis index as a function of strain rate and cycle number to enable systematic optimisation. With respect to repeatability, examination of the time traces in Figs. 7(a) and 7(c) shows that the peak R/R₀ value in each cycle remains broadly consistent across the 10 cycles tested for both the FDM sensor and the bare filament, indicating that neither specimen undergoes catastrophic or progressive failure of the conductive network within this test duration. However, a gradual upward drift in the baseline resistance (R/R₀ at the minimum strain of each cycle) is observable in Fig. 7(a) for the FDM-printed sensor, which is not observed to the same degree in the bare filament (Fig. 7(c)). This baseline drift is attributed to the progressive irreversible collapse and misalignment of the triangular voids within the FDM print (Figs. 7(c) and 7(d)), which alters the effective gauge geometry with each cycle. It should be noted that the 10-cycle protocol employed here is intentionally limited in scope: the primary objective of this cyclic characterisation is to demonstrate proof-of-concept functionality of the FDM-printed nanocomposite strain sensors and to establish the qualitative nature of their electromechanical response under repeated loading. A rigorous assessment of long-term durability, cycle-to-cycle drift, and fatigue behaviour over hundreds or thousands of cycles falls outside the scope of the present study and is identified as a key priority for future work.

Figure 8.

a) Schematic illustration of strain application on FDM printed beads with progressive elongation shown through the reduction in lateral dimensions and increase in length. b) Electrical resistance (R/R₀) response during cyclic stretching up to 50% strain, highlighting different deformation stages (No Strain, Stage 1, Stage 2, Recovery). c) SEM micrographs of the triangular voids. d) Detailed view of the void with illustration.

The behaviour of the hybrid filler network within a single filament fibre during cyclic loading is presented in Fig. 8(a). The data and stages of the unusual double-peaking phenomenon can be seen in Fig. 8(b). Initially, in the no-strain condition, only a few CNTs make contact, providing a baseline resistance reading at 0% strain. The CNTs have uniform lateral spacing with some bridging, forming a conductive filler network. At this stage, the BN particles are assumed to be well dispersed. As strain is applied (Stage 1) there is a significant increase in the number of CNTs in contact, particularly in the through-plane direction. In Stage 2, the void-filled structure of an FDM print leads to a reduction in the number of CNTs in contact, with a more separated arrangement in the longitudinal direction compared to Stage 1. This indicates a relaxation and realignment toward a less dense configuration. The recovery stage, labelled in Fig. 8(b) and visible in the time traces of Figs. 7(a) and 7(c) as the return of R/R₀ toward its baseline value following unloading, proceeds over a few seconds for both the FDM sensor and the bare filament at the 0.008 Hz cycling rate used here. Importantly, the resistance does not fully recover to the initial R₀ within the time allowed by the 0.008 Hz cycle period, as evidenced by the non-closure of the R/R₀–strain loops in Figs. 7(b) and 7(d). This incomplete electrical recovery within each cycle is distinct from the cycle-to-cycle baseline drift discussed above, and reflects the viscoelastic nature of the TPU matrix, which also does not fully recover mechanically within the cycle period at this strain rate.

Figure 8(c) shows the void structures when printing a thermoplastic elastomer (TPU-GNP5-CNT5). The TPU based samples exhibit triangular voids (Fig. 8(d)), with no visible indication of layer lines, indicating a strong interface. The triangular voids arise due to the high printing temperatures required for the thermoplastic elastomer, resulting in lower viscosity and less precise shape retention. The collapse of these voids during cyclic loading is hypothesised to contribute to the observed double-peaking behaviour. The detailed examination of these structural characteristics provides insights into the material's response to cyclic loading and the factors influencing its mechanical performance.

Conclusions

This study demonstrates the potential of hybrid TPU nanocomposites reinforced with CNT, GNP, and BN as effective materials for FDM-printed strain sensors. Incorporating hybrid fillers mitigated challenges like high extrusion forces and nozzle blockages typically faced during FDM of flexible, conductive materials. Notably, CNT-BN composites required lower extrusion forces than CNT-only counterparts while maintaining or enhancing electrical conductivity and mechanical properties.

The hybrid CNT-BN composites exhibited excellent printability, conductivity, and strain-sensing performance, with strain ranges over 250%. These materials outperformed CNT-GNP and CNT-only composites in terms of reduced extrusion forces, improved conductivity, and greater strain at break, making them suitable for scalable, reliable FDM printing. The novel extrusion force measurement technique used offers a valuable tool for optimising FDM processes for nanocomposite filaments.

In summary, TPU-CNT-BN nanocomposite filaments provide a promising material for fabricating high-performance, flexible strain sensors via FDM, with potential applications in wearable electronics and structural health monitoring. Future work could further optimise filler distribution and investigating the long-term durability of these sensors under real-world conditions.

Table 1.

FDM parameters used for the 3D printing of TPU nanocomposites.

Parameter	Value	Parameter	Value
Print nozzle diameter (mm)	1.4	Layer height (mm)	0.3
Nozzle temperature (°C)	240	Print infill (%)	100
Bed temperature (°C)	60	Print speed (mm/s)	20

Footnotes

Acknowledgements

The authors acknowledge the support from “Graphene Core 3” GA: 881603 which is implemented under the EU-Horizon 2020 Research & Innovation Actions (RIA) and is financially supported by EC-financed parts of the Graphene Flagship. The authors would also like to acknowledge support from the Engineering and Physical Sciences Research Council (EP/V037234/1, EP/V037234/2, ESTEEM).

ORCID iD

Dimitrios G. Papageorgiou

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Horizon 2020 Framework Programme, Engineering and Physical Sciences Research Council, (grant number 881603 , EP/V037234/1 and EP/V037234/2).

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

References

Ngo

Kashani

Imbalzano

, et al. Additive manufacturing (3D printing): a review of materials, methods, applications and challenges. Compos B Eng 2018; 143: 172–196.

Ligon

Liska

Stampfl

, et al. Polymers for 3D printing and customized additive manufacturing. Chem Rev 2017; 117: 10212–10290.

Kim

Park

Suh

J-h

, et al. 3D Printing of multiaxial force sensors using carbon nanotube (CNT)/thermoplastic polyurethane (TPU) filaments. Sens Actuators, A 2017; 263: 493–500.

Zhang

, et al. Strain sensing behavior of FDM 3D printed carbon black filled TPU with periodic configurations and flexible substrates. J Manuf Process 2022; 74: 283–295.

Kwok

Goh

KHH

Tan

, et al. Electrically conductive filament for 3D-printed circuits and sensors. Applied Materials Today 2017; 9: 167–175.

Ahn

Montero

Odell

, et al. Anisotropic material properties of fused deposition modeling ABS. Rapid Prototyp J 2002; 8: 248–257.

Ursini

Collini

. Materials (Basel), 2021; 14: 5645.

P-C

Liu

M-Y

Zhang

, et al. Enhanced electrical conductivity of nanocomposites containing hybrid fillers of carbon nanotubes and carbon black. ACS Appl Mater Interfaces 2009; 1: 1090–1096.

Cataldi

Steiner

Raine

, et al. Multifunctional biocomposites based on polyhydroxyalkanoate and graphene/carbon nanofiber hybrids for electrical and thermal applications. ACS Applied Polymer Materials 2020; 2: 3525–3534.

10.

Papageorgiou

Terzopoulou

Fina

, et al. Enhanced thermal and fire retardancy properties of polypropylene reinforced with a hybrid graphene/glass-fibre filler. Compos Sci Technol 2018; 156: 95–102.

11.

Dong

Sun

Dunstan

, et al. Mechanical reinforcement from two-dimensional nanofillers: model, bulk and hybrid polymer nanocomposites. Nanoscale 2024; 16: 13247–13299.

12.

Yang

Chen

, et al. Interfacial engineering of hybrid MXene-Ni-CF tri-core–shell composites for electromagnetic interference shielding and E-heating applications. Compos Part A Appl Sci Manuf 2024; 178: 107990.

13.

Dong

Tomes

Soul

, et al. Hybrid Ti₃C₂T_xMXene and carbon nanotube reinforced epoxy nanocomposites for self-sensing and structural health monitoring. ACS Applied Nano Materials, 2024; 7: 3314–3325. 3C₂T _x MXene And carbon nanotube reinforced epoxy nanocomposites for self-sensing and structural health monitoring.>

14.

Szeluga

Kumanek

Trzebicka

. Synergy in hybrid polymer/nanocarbon composites. A review. Compos Part A Appl Sci Manuf 2015; 73: 204–231.

15.

Bao

H-D

Guo

Z-X

.Effect of electrically inert particulate filler on electrical resistivity of polymer/multi-walled carbon nanotube composites. Polymer 2008; 49: 3826–3831.

16.

Bilotti

Zhang

Deng

, et al. Controlling the dynamic percolation of carbon nanotube based conductive polymer composites by addition of secondary nanofillers: the effect on electrical conductivity and tuneable sensing behaviour. Compos Sci Technol 2013; 74: 85–90.

17.

Wang

Das

Thorn

TDS

, et al. Effect of boron nitride on the pyroresistive properties of smart conductive polymer composites: eliminating the negative temperature coefficient effect. Compos Sci Technol 2025; 267: 111203.

18.

Papageorgiou

Liu

, et al. Hybrid poly(ether ether ketone) composites reinforced with a combination of carbon fibres and graphene nanoplatelets. Compos Sci Technol 2019; 175: 60–68.

19.

Zhou

Zhu

Liu

, et al. High thermal conductivity of suspended few-layer hexagonal boron nitride sheets. Nano Res 2014; 7: 1232–1240.

20.

Papageorgiou

Kinloch

Young

. Hybrid multifunctional graphene/glass-fibre polypropylene composites. Compos Sci Technol 2016; 137: 44–51.

21.

Papageorgiou

Young

. 1.6 – Graphene Composites. In: Verpoest I, et al. (eds) The history of composites. Amsterdam: Elsevier, 2026, pp. 289–319.

22.

Wang

Liu

Schubert

. Highly sensitive ultrathin flexible thermoplastic polyurethane/carbon black fibrous film strain sensor with adjustable scaffold networks. Nano-micro Letters 2021; 13: 1–19.

23.

Xiang

Zhang

, et al. Enhanced performance of 3D printed highly elastic strain sensors of carbon nanotube/thermoplastic polyurethane nanocomposites via non-covalent interactions. Compos B Eng 2019; 176: 107250.

24.

Chen

Xie

Wang

, et al. Recent advances of flexible strain sensors based on conductive fillers and thermoplastic polyurethane matrixes. ACS Applied Polymer Materials 2021; 3: 5317–5338.

25.

Liu

Zhang

, et al. Electrically conductive polymer composites for smart flexible strain sensors: a critical review. J Mater Chem C 2018; 6: 12121–12141.

26.

Tomes

Soul

Zhang

, et al. A novel electrical depercolation model for stretchable nanocomposite strain sensors. Nanoscale 2024; 16: 6163–6175.

27.

Cochrane

Koncar

Lewandowski

, et al. Design and development of a flexible strain sensor for textile structures based on a conductive polymer composite. Sensors 2007; 7: 473–492.

28.

Dong

Soul

, et al. Transient starch-based nanocomposites for sustainable electronics and multifunctional sensing. Adv Funct Mater 2025; 35: 2412138.

29.

Dong

Cataldi

Zhang

, et al. Edible and recyclable gelatin‐based electronics for high‐precision health and environmental monitoring. Advanced Science 2025; 13 (15): e07950.

30.

Marsden

Papageorgiou

Vallés

, et al. Electrical percolation in graphene–polymer composites. 2D Mater 2018; 5: 032003.

31.

Koronio

Shachar-Michaely

Regev

, et al.A Study on Thermal Conductivity Enhancement in Composites Utilizing Excluded Volume Effects. Langmuir 2025; 41: 24484–24495.

32.

Georgopoulou

Vanderborght

Clemens

. Fabrication of a soft robotic gripper with integrated strain sensing elements using multi-material additive manufacturing. Frontiers in Robotics and AI 2021; 8: 615991.

33.

Avanzare. ‘Graphene’, 2023. viewed; Available from: https://www.avanzarematerials.com/graphene/.

34.

Sigma-Aldrich. ‘Boron nitride powder, 1um, 98 10043-11-5’, 2023. viewed; Available from: http://www.sigmaaldrich.com/.

35.

Creality. ‘Ender-3 S1 3D Printer’, 2023. viewed; Available from: https://www.creality.com/products/creality-ender-3-s1-3d-printer.

36.

VzBoT. ‘Vz-HextrudORT’, 2023 [viewed 2024 04-01-2024]; Available from: https://github.com/VzBoT3D/Vz-HextrudORT.

37.

O’Mara

Ogilvie

Large

, et al.Ultrasensitive strain gauges enabled by graphene‐stabilized silicone emulsions. Adv Funct Mater 2020; 30: 2002433.

38.

Christ

Aliheidari

Ameli

, et al. 3D Printed highly elastic strain sensors of multiwalled carbon nanotube/thermoplastic polyurethane nanocomposites. Mater Des 2017; 131: 394–401.

39.

Cho

M-Y

Lee

Kim

S-H

, et al. An extremely inexpensive, simple, and flexible carbon fiber electrode for tunable elastomeric piezo-resistive sensors and devices realized by LSTM RNN. ACS Appl Mater Interfaces 2019; 11: 11910–11919.

40.

del Bosque

Sánchez-Romate

Sánchez

, et al.Toward flexible piezoresistive strain sensors based on polymer nanocomposites: a review on fundamentals, performance, and applications. Nanotechnology 2024; 35: 292003.