Introduction

State of the art

Glass fiber reinforced plastics (GFRP) are materials commonly used in light weight design.1,2,3,4 As with many other composite materials, there is currently no satisfying process for separating GFRP in a value-preserving manner at the end of its service life.5,6,7,8 In the state of the art, GFRP often are deposited or used as a filler in granular form, mostly in concrete. This is associated with strong impairment due to the loss of fibrous structure.9,10 To realize a genuine circular economy for GFRP, it is essential to separate the glass fibers from the polymer matrix at end-of-life. There are some experimental approaches for this separation process.

First, there is the idea of pyrolysis which is commonly used in recycling of carbon fiber reinforced plastics. Here, the polymeric compound is thermally decomposed at high temperatures and usually in an oxidizing atmosphere. This process cannot be easily transferred to GFRP because of the different chemistry of the fibers. With pyrolysis, the tensile strength decreases extremely (− 50%) compared to virgin glass fibers.11 Another drawback is the loss of the polymer.5

Second, GFRP can be separated by solvolysis. Here, supercritical solvents are used to dissolve the polymer. An additional catalyst is used to decompose the polymer. In consequence, there are three fractions: the inorganic compounds (mostly glass fibers), the polymeric compounds and the monomeric compounds. Different from, e.g., pyrolysis, the polymer and resulting monomers can be reused as well. The greatest drawback of this method is the use of harmful solvents and the energy-consuming nature of the process.12

Third, separating the fibers from the polymer can be achieved by subcritical or supercritical hydrolysis. Here, the GFRP, applicable for polyester-based matrices, is treated with water instead of organic solvents. The polyester is hydrolyzed by acid catalysis. For this reaction, the critical point of water must be reached, in this example the reaction is carried out at 275°C and 60 bar, so it is highly energy-consuming. The main advantage is the absence of harmful solvents. Furthermore, the polymer is decomposed in its monomeric compounds which can be reused as well. On the other hand, the reaction is very slow, due to the necessary diffusion of hydroxonium ions.8,13 Another drawback is acid corrosion of the glass fibers which describes the reaction of the glass surface with acids. The E-glass which is commonly used in most of the glass fiber reinforced plastics is very sensitive to acids. This acid corrosion causes a significant weight loss of the glass and potential breaking.14,15

Debonding on demand

A possible solution for the outlined issues is the by-design integration of delamination features into the composites. Such approaches are commonly referred to by the term “debonding-on-demand” (DoD), which is already well established in adhesion technology. In this technology, different mechanisms are used to realize a reversible or irreversible debonding. For chemical debonding, reversible covalent, or noncovalent bonds are introduced into a polymeric network. Therefore, Diels–Alder chemistry commonly is used for covalent bonds,16,17 while metal coordination18 or π–π-stacking19 are examples for reversible and noncovalent bonding/debonding systems. The trigger which induces the debonding process can vary: commonly used are light-induced20 or thermally induced debonding.21

Furthermore, there are irreversible debonding mechanisms. For example, photolabile crosslinkers can be introduced in a polymeric network. If this network is exposed to UV light, the network degrades.22 It is also possible to introduce a physical debonding, for example, with gas formation in the interface phase where the delamination is expected to occur.23

DoD mechanisms are, however, not yet considered for GFRP recycling. In this work, we therefore aim to introduce the DoD concept as a promising rationale toward circular economy in GFRP. In comparison with the adhesion technology, GFRP have a much more complicated 3D structure. According to this, a trigger is necessary, which can also penetrate larger compounds of several centimeters in length and thickness with a complex structure and trigger the debonding. Therefore, common stimuli like UV light or hydroxonium/hydroxy ions cannot be used. One trigger, which is easy to apply, is heat. Depending on the size of the GFRP, heat conduction takes time. Therefore, a magnetic or electric field can be useful to prevent the time-consuming heat conduction. Another factor, which must be assumed, is the price of a recyclable GFRP. Most of the mentioned recycling methods are not economical because newly spun glass fibers are very cheap. Additionally, profound changes within the spinning process are mostly rejected by industrial producers. They must thus be avoided.

In the presented work, we aim to develop a DoD process that keeps the above-mentioned limitations in mind. As mentioned before, it is only possible to use triggers like heat or electric/magnetic fields which can easily penetrate the compounds. To keep changes to the spinning process as low as possible, our aim is to synthesize an additive which releases gases above approximately 300°C to initiate debonding. In the desired temperature range, decarboxylation of many substances like carbonates, carboxylic acids or amino acids takes place. Therefore, these substances are the initiation point to develop a suitable debonding additive.

The first requirement for the additive is a fitting decomposing temperature. As mentioned before, the ideal recycling temperature is around 300°C. This temperature is significantly higher than the usual service temperatures, yet low enough to preserve the glass’s tensile strength. Second, the additive must not reduce the adhesion between glass fibers and matrix polymer; thus pure carbonic acids cannot be used. In consequence, small molecules like carbonates, carboxylic acids or amino acids must be encapsulated or immobilized on, e.g., particles. Third, the additives must be easily implementable into existing sizing coatings. Commercial sizings are aqueous solutions or dispersions containing about 10 wt.% polymer and about 1 wt.% silane as adhesion promoter (e.g., 3-aminopropyl silane). The pH of sizings is determined by the chemistry of the polymer and varies between 3 and 10. In addition to the desired compatibility to such commercial sizings, the additive must also not negatively influence the spinning process. Usually, the quality of the spinning process is measured by fiber breaks per hour. Thus, this value must not be increased by the additives.

Experimental

Materials

All used chemicals were purchased from the usual suppliers and were used without further purification. The epoxy resin R1 and epoxy hardener H1 were purchased from DD Composite GmbH (Bad Liebenwerda, Germany). The polymer dispersion Baybond XL800 used for the sizings was purchased from Covestro AG (Leverkusen, Germany) and had a nonvolatile part of 44.5%. Embedding the fibers for the pull-out tests, a 2 K epoxy system was used (EPIKOTE Resin TRAC 06150 and EPIKURE Curing Agent TRAC 06150) which was purchased from Hexion Inc. (Columbus, Ohio, USA).

Debonding experiments on flat glass

For a proof of concept, 6 wt.% solutions of β-alanine, citric acid, tartaric acid or maleic acid and water were prepared. To begin, 30 µL of this solution was placed on a glass substrate. After drying, the pretreated substrate was coated by an epoxy resin and cured at 60°C for 24 h in an oven. Following this, these samples were treated at 200°C for 30 min simulating the recycling process.

Preparation of capsules

Synthesizing capsules, hollow spheres were prepared in the first step. Therefore, magnetite nanoparticles (MNPs) were used as a template. MNPs were synthesized via a one-step precipitation. First, 2.45 g of water-free FeCl3 (15.1 mmol) and 2.10 g of FeSO4 heptahydrate (7.55 mmol) were solved in 100 mL deionized water and heated to 40°C while stirring (500 rpm). Afterward, 0.8 mL (2.53 mmol) of oleic acid was added as a spacer for stabilization. With addition of sodium hydroxide solution (5.87 g in 10 mL DI water), magnetite nanoparticles were precipitated and ripped for 30 min. These MNPs were separated with the help of a magnet and washed with water until the dispersion was pH-level neutral.

In the second step, the magnetite was coated with silica in a Stöber-like process. Therefore, 5.00 g of the MNP dispersion (6,67 wt.% particles in water) was mixed with 40 mL ethanol and 2 mL ammonia solution (25%). Then, 1.0 g tetraethyl orthosilicate was added and the reaction was carried out in the ultrasonic bath for 1.5 h at 50°C. Again, the MNPs were separated with the help of a magnet and washed until the dispersion was pH-level neutral.

Preparing the hollow spheres, the magnetite was solved by concentrated hydrochloric acid. Once the particles fell to the ground, the supernatant was carefully removed with a pipette. This procedure was repeated until the dispersion was pH-level neutral.

Filling the hollow spheres, a dispersion of the hollow spheres was mixed with a saturated solution of (NH4)2CO3 and stored for 10 days with periodic redispersion but without stirring.

Preparation of core-shell particles

In this synthesis line, magnetite was used as core again. Now, carboxylic acids and amino acids were used as a shell. There were two different synthesis routes for both classes of acids.

First, the synthesis route for functionalization of magnetite with amino acids is described. The synthesis is known in the literature with light variation. Magnetite precipitation and functionalization with amino acids were done in the same reaction step. Therefore, two solutions were prepared: for solution 1, 0.44 g of sodium chloride and 10 mmol of any amino acid were solved in 50 mL deionized water. For solution 2, 1.6 g of water-free iron(III) chloride and 1.4 g of iron(II) sulfate heptahydrate were solved in 37 mL of 2 M hydrochloric acid. The first solution was heated to 80°C, and then, it was mixed with solution 2. Afterward, 13 mL of concentrated ammonium solution (25%) was added at a rate of 0.5 mL/min via a syringe pump. Following this, the reaction mixture ripened at 80°C for 24 h. The functionalized MNPs were separated with the help of a magnet and washed with water until the suspension was pH-level neutral.

The second synthesis route was specific for the functionalization with carboxylic acids. Here, the MNPs were synthesized in a first step and functionalized in a second step. For this synthesis, the MNPs were synthesized in an ultrasonic bath. Therefore, 0.65 g of water-free iron(III) chloride, 0.56 g of iron(II) sulfate heptahydrate and 2 mL of concentrated hydrochloric acid were solved in 200 mL deionized water. This solution was heated in an ultrasonic bath to 55°C and a 4 M solution of sodium hydroxide was added at a rate of 0.5 mL/min with a syringe pump. Afterward, the MNPs were washed until the suspension was pH-level neutral.

In the second step, the MNPs were dispersed in 100 mL diethylene glycol and redispersed in the ultrasonic bath for 15 min. Then, 1 mmol of any carboxylic acid was solved in 10 mL of diethylene glycol and added. The reaction was carried out under strong stirring at 100°C for 1 h. Afterward, the particles were washed several times with water.

Up-scaling

For the experiments at the spinning machine, 15 g of the functionalized particles were needed. Therefore, a tubular rector was built for a continuous synthesis of the core-shell particles. The reactor was a tube reactor with a diameter of 8 mm and had a volume of about 100 mL. The tube was made of PVC. The hole reactor could proceed in an ultrasonic bath. The reaction was performed with different conditions in water at 50–60°C. Different synthesis parameters like flow rate, iron-ion concentration, or sodium hydroxide concentration were varied.

Fiber spinning

The spinning experiments were performed at a custom-made spinning system. The oven was set to 1015°C and the spinning speed was 470 m/min. The glass raw material was E-glass purchased from Compagnie de Saint Gobain AG, Courbevoie. The composition of the used glass is described in Table 1.

Table 1 Composition of utilized E-glass
Full size table

Sizings with varying filling levels were used. All of them contained 11 wt.% polymer dispersion and 0.9 wt.% (3-aminopropyl)triethoxysilane as an adhesion promoter. Particles were added in three different concentrations: 0.1, 0.5, and 1.0 wt.%. The dispersing additive DISPERBYK 194 N was added to some sizings, increasing the stability of the dispersion. The formulations of the sizings are shown in Table 2.

Table 2 Formulation of the used sizings for the spinning experiments
Full size table

Scanning electron microscopy (SEM) and energy dispersive X-ray spectroscopy (EDX)

SEM imaging and EDX was performed on a Neon 40 EsB CrossBeam from the Zeiss (Oberkochen, Germany). The images were taken with an acceleration voltage of 2 kV while the InLens detector was used. The EDX spectra were obtained at an acceleration voltage of 5 kV. Spectra of about 20–30 points were obtained on every sample. The calculated elemental distribution of these spectra was averaged.

Thermogravimetric analysis (TGA)

The TGA measurements were performed using a TGA 8000 from PerkinElmer (Waltham, MA, USA). Therefore, the following temperature profile was used: heating with 10 K/min up to 120°C, holding 120°C for 10 min, heating with 10 K/min up to 300°C, holding 300°C for 10 min, heating with 10 K/min up to 800°C, holding 800°C for 10 min. The measurements were performed in a nitrogen atmosphere.

For better comparability, the weight was shown relatively. Additionally, the weight was set to 100 wt.% after the isotherm at 120°C to compensate for the weight loss due to remaining solvents. In most cases, the weight loss after the 300°C isotherm, which represented the recycling temperature, was observed.

Dynamic light scattering (DLS) and zeta potential

To measure the particle size, dynamic light scattering was performed. Therefore, the ZetaSizer Ultra Red Label from Malvern Pananalytical (Malvern, Worcestershire, UK) including a laser with 633 nm was used. The size measurements were performed at a scattering angle of 173°, and the zeta potential was measured at 13°. In the measuring program, the size was measured three times. Directly afterward, the zeta potential was measured three times, too. Then, the size was measured again three times. The d90 diameter was calculated by the associated software, ZS Xplore.

Pull-out tests

The machines used for the pull-out tests were manufactured by Textechno Herbert Stein GmbH & Co. KG (Mönchengladbach, Germany). For embedding of the fibers, the embedding station FIMABOND was used. The embedding velocity was 500 mm/min; the embedding depth was 60 µm. The 2 K epoxy system was heated at 120°C for 5 min and tested afterward. The single fiber test device FAVIMAT + was used for the pull-out tests. Ten measurements per sample were carried out. The fibers were tested before and after heating (250°C, 15 min). During the analysis, the force was measured via displacement. This force was normalized with the embedding depth and the fiber diameter, and the apparent interlaminar shear strength was calculated.

Results

Conceptualization

DoD mechanisms have a large potential for a value-retaining recycling method. Because of the highly structured nature of GFRP and the strong prize limitation, a simple DoD mechanism, based on thermal expansion, is chosen. As gas-releasing, thermo-responsive substances (TRS), carbonates, carboxylic acids and amino acids are investigated because of their low price point and their ideal decarboxylation temperature of 250–300°C. This temperature is high enough above the usual operating temperatures, but also low enough to exclude thermal degradation of glass fiber (mechanical) properties.

Direct usage of the mentioned substances, however, would introduce small crystalline substances to the interfaces. These crystals are mobile, which would significantly reduce adhesion between fiber and matrix. Therefore, these small crystalline structures must be avoided which can be realized by encapsulation or immobilization on, e.g., particles. Depending on the chemistry of the particles or their shell, covalent bonds between particle and fiber can be formed. The particles are not mobile anymore and the adhesion is not lowered. To realize this concept, small particles are needed that can both be easily synthesized and functionalized. Magnetite fulfills these requirements as it can be easily prepared by alkaline precipitation of iron ions and can form covalent or noncovalent bonds.

Initial experiments on flat glass as proof of concept

As initial proof of concept, flat glass substrates are coated with the carboxylic and amino acids at distinct spots. The entire glass substrate is then coated by a commercial epoxy resin. The samples are subsequently heated, and the occurring delamination is evaluated optically.

The utilized TRS strongly decreases the adhesion of a coating to the substrate when used in pure form due to its crystalline structure. Consequently, the adhesion between coating and glass before heating is decreased. Therefore, it was not measured. After heating, the coatings are rated visually.

All tested substances show a gas development after heating (Fig. 1). The best result shows citric acid; here the topcoat is completely delaminated. The sample, which is treated by β-alanine, shows no complete delamination but a grand gas bubble beneath the coating. In consequence, the coating only has to be cut for a good delamination from the surface.

Fig. 1
figure 1

Glass substrates are treated with a solution of thermo-responsive substances before coating them (left). After heating (right), the coating is colored at the treated spots (marked in orange). Under the coating, gases are formed

Full size image

The samples treated with tartaric acid and malic acid show an inhomogeneous gas development. In both cases, there are places in the treated spot where the topcoat sticks to the glass. Consequently, the delamination is incomplete, and the glass substrate must be cleaned in a time-consuming process.

Nanocapsules

The crystalline TRS decrease the adhesion between glass substrate and the coating if they are in the interface. This would effectively result in a poor fiber/matrix adhesion and consequently in reduced mechanical properties of the GFRP. Therefore, the crystalline TRS must be encapsulated to use them in a GFRP. Usually, glass fibers for reinforcement have a diameter of about 15 µm. To localize the TRS as close as possible in the interface, the capsules should have a maximum size of 500 nm.

Due to this harsh limitation of size, it is not possible to encapsulate the TRS directly. As an alternative, hollow spheres are synthesized and filled with TRS afterward. The overall synthesis route includes four steps: First, magnetite nanoparticles (MNPs) are built as templates. Second, the MNPs are coated with silica. Third, hollow spheres are prepared by dissolving the magnetite. Fourth, the hollow spheres are filled with TRS. As first TRS, ammonium carbonate is used, due to its small molecular size, which should result in a higher diffusivity compared to larger carbonic or amino acids. This synthesis is shown schematically in Fig. 2a.

Fig. 2
figure 2

Scheme of the synthesis route to prepare nanocapsules (a). SEM images of pure magnetite (b), silica coated magnetite (core-shell particles, c), silica hollow spheres (d) and hollow spheres filled with ammonium carbonate (e). The particles are also analyzed by EDX (f) showing the elemental distribution of carbon (blue), nitrogen (red), silicon (green) and iron (orange). The size of the primary particles is shown tabularly (g). TGA was done at every step of the synthesis (h): pure magnetite (red), core-shell particles (light gray), hollow spheres (yellow) and hollow spheres filled with ammonium carbonate (dark gray)

Full size image

First, the pure magnetite particles are characterized, which will be used as a template for the silica layer. As shown in Fig. 2 b, the primary particles, analyzed via SEM microscopy, are very small and have an average diameter of about 16 nm. EDX analysis (Fig. 2 f) shows the absence of carbon, nitrogen and silicon. Via TGA (Fig. 2 h), a weight loss of about 2.9 wt.% is observed. This can be attributed to adhering oleic acid, which was used as a spacer and stabilizer in the synthesis to grow stable and large magnetite particles.

In the next step, the particles are coated with silica. The primary particle size increases to an average of 23 nm (Fig. 2c) which indicates that the synthesis was successful. In the EDX analysis (Fig. 2f), a strong increase of silicon is visible which further indicates the presence of a silica layer. In addition, the carbon content increases too. This might be caused by nonhydrolyzed sidechains of the used silane. The TGA (Fig. 2h) shows a weight loss of about 1.8 wt.%. This might be unhydrolyzed ethoxy groups of the TEOS or remaining water and the oleic acid adhered on the magnetite core.

After the treatment with concentrated hydrochloric acid, the particle structure is still intact (Fig. 2d). The primary particles have a size of about 21 nm. The EDX measurement proves the absence of iron and by this a complete removal of the magnetite core. The TGA shows a small weight loss of about 0.9 wt.%. This might still be unhydrolyzed ethoxy groups of the silane or just adsorbed water inside the cavity.

In the final step, the filling of hollow spheres is analyzed. SEM (Fig. 2e) shows still intact particles without significant change of size. Interestingly, no contaminations in the form of salt crystals are observed. A very small increase of the nitrogen and carbon in the EDX measurement might indicate a slight filling of the hollow spheres with the TRS, although the significance is not sufficient to claim this as proof. Also, TGA shows a nearly insignificant weight loss of only 2.2 wt.% which indicates poor diffusion of ammonium carbonate into the hollow spheres. The diffusion might be improved by a thinner silica shell. Even if the results might be due to the TRS, the amount is much too low to be of significance for the desired application. Consequently, an alternative method for integration of the TRS into glass fibers must be considered next.

Core-shell particles

As the second synthesis route, TRS are now directly immobilized on the surface of MNPs. This type of functionalization is known from literature. Due to its magnetic properties and the small particle sizes, the particles tend to agglomerate. Preventing this, molecules are grafted at the surface for stabilization. Therefore, carbonic or amino acids are used frequently. The stabilization works in a simple one-step24,25 or two-step26,27 synthesis for both acid classes, so it should be easy to produce in a large scale.

Another advantage of magnetite is its possibility for inductive heating.28,29 This might be an advantage in the future recycling processes, because the heat develops exactly where the debonding should happen, i.e., in the boundary phase. Unlike with conventional thermal treatment, heat must not be conducted through the bulk to the boundary surfaces. Consequently, the overall process should be more energy-efficient and less stressing for the bulk materials.

To immobilize as much substance as possible, the MNPs must be as small as possible. Carbonic and amino acids are used as TRS now, due to the ability of the carboxylic acid group to coordinate the iron oxide on the surface of the MNPs.24,30 For both TRS classes, carboxylic and amino acids, synthesis routes are known from literature.27,31 MNPs functionalized with amino acids are prepared in a one-step synthesis: iron salts and amino acids are solved in saline solution and precipitated by ammonia (schematically shown in Fig. 3a).31 This synthesis route cannot be adapted for functionalization with carboxylic acid because with some carboxylic acids (e.g., citric acid) complexes are formed instead of MNPs. Therefore, the functionalized particles are prepared via a two-step synthesis: First, MNPs are precipitated by ammonia. Second, the MNPs are dried in resolved in diethylene glycol and functionalized with carboxylic acid (Fig. 3b).27

Fig. 3
figure 3

Schematic representation of the two synthesis routes for amino acids (a) and carboxylic acids (CA), (b). The iron ions are displayed in orange, the TRS in red and the magnetite in black. The weight loss is measured after the isotherm at 300°C (c). Zeta potential of core-shell particles is measured in dependence of the pH (d). The d90 is measured in the most stable dispersion according to the zeta potential (e). SEM images of particles functionalized with malonic acid (f), citric acid (g), oxalic acid (h), β-alanine (i) and aspartic acid (j)

Full size image

As shown in Fig. 3c, both functionalization routes display a significant difference in weight loss compared to pure magnetite (e.g., Fig. 2g). It is obvious that not all TRS give the same result: the best result shows citric acid with a weight loss of 24 wt.%, and the smallest weight loss shows aspartic acid with 5 wt.%. The other three samples show a weight loss between 10 and 15 wt.%. These different results are possibly caused by different affinities of the TRS to magnetite as well as different thermal behavior of the pure TRS.

Next, the particles are characterized with DLS, shown in Fig. 3d. The zeta potential is measured to test the stability. This was done at three different pH values (3, 7 and 10) because the surface is protonated by acid and deprotonated by alkali. For further analysis, samples are measured only in dispersions with pH values, where the dispersion is at least moderately stable (i.e., the zeta potential is greater than 20 mV32).

All functionalized MNPs are at least moderately stable in neutral or alkaline dispersions. Only the particles functionalized with oxalic acid are most stable at pH 7, and all other particle dispersions are most stable at pH 10. This can be explained by the pKa values of the acids. Oxalic acid has the smallest pKa values of all used TRS. Therefore, it is completely deprotonated at pH 7. All the other TRS have higher pKa values, so they are only deprotonated at pH 10.

The amino acids (aspartic acid and β-alanine) show the smallest zeta potential of all samples. Again, this can be explained by the pKa values: aspartic acid has two carboxylic acid groups and one amino group. At pH 7, the carboxylic acid is deprotonated while the amino group is protonated. This zwitterion has no resulting charge and cannot stabilize the MNPs. At pH 10, the amino group is deprotonated as well and the MNPs are stabilized. According to this theory, β-alanine should exhibit positive zeta potentials at low pH values. However, it displays comparable behavior to other samples with exposed carboxylic acid groups on the surface. This implies the alanine-functionalized particles also have acidic groups on the surface, suggesting that alanine binds to the MNPs via the amino group.

All particles, except those functionalized with aspartic acid, have a d90 value smaller than the upper limit of 500 nm, like Fig. 3e shows. This can be explained by the zeta potential: aspartic acid stabilizes the particles the least and the particles agglomerate. Although the other four samples are all smaller than 500 nm, there are even two different size ranges. MNPs functionalized with citric acid or β-alanine have a much smaller d90 value (40–70 nm) than those particles functionalized with malonic or oxalic acid (130–260 nm). These samples have a similar zeta potential, but the amount of bound TRS is different. This indicates that the smaller particles are more stable because of additional steric stabilization.

According to the SEM pictures (Fig. 3f–j), even the particles with the smallest d90 value of about 40 nm are small agglomerates. All MNPs are spherical and are not crosslinked. The MNPs functionalized with carboxylic acids tend to be smaller than the ones functionalized with the amino acids. This can be easily explained by the different synthesis routes: in the two-step synthesis with carboxylic acids, the particles are precipitated in an ultrasonic bath, first. This results in smaller particles.

In summary, both synthesis routes work. SEM images show a tendency of smaller particle sizes with the two-step functionalization, which can be explained by the synthesis in an ultrasonic bath. In addition, there are two particle species which are the best choice: MNPs functionalized with citric acid or β-alanine. Both show the highest weight loss and the smallest d90 value. On the other hand, functionalization with aspartic acid does not result in a stable particle dispersion. Also, it is less TRS immobilized than on all other samples.

Up-scaling

To enable the concept transfer to actual fiber spinning machines, the particles must be added to a commercial polymer dispersion to create the new sizing mixture for the spinning process. Even at the smallest scale, however, at least 15 g of functionalized particles is needed, whereas previous batch syntheses yielded only 1-1.5 g. To synthesize the necessary amount of the particle dispersion, two tubular reactors (first reactor: diameter of the tube: 8 mm, volume: 100 mL; second reactor: diameter of the tube: 4 mm, volume: 50 mL) are thus designed to enable a continuous synthesis and shown in Fig. 4a. The process parameters are varied to analyze their influence on the product particles. In the first step, the synthesis of pure magnetite was considered.

Fig. 4
figure 4

A tubular reactor is manufactured to enable a continuous magnetite synthesis (a). The d90 diameter and the zeta potential (purple) of pure magnetite is analyzed in dependence of the iron-ion concentration and the flow rate (b). Additionally, the zeta potential is analyzed, too. Furthermore, the dependence of the NaOH concentration and the HCl concentration in the iron-ion solution on the diameter of the MNPs is tested (c). Finally, a second reactor with a smaller diameter of the tube and a smaller reaction volume is tested (d). Furthermore, the concentration of the stabilizing DEG is varied (d)

Full size image

As shown in Fig. 4, many parameters influence the MNPs’ size. It is obvious that huge agglomerates of the primary magnetite particles are formed. Due to the different surface chemistry, pure magnetite particles are most stable at pH 3. The positive zeta potential in acidic environment indicates protonated (hydroxy) groups at the particles’ surfaces. At this pH, nearly all samples show a zeta potential larger than 25 mV.

A strong parameter for size control is the iron-ion concentration while precipitation, displayed in Fig. 4b. A smaller concentration results in smaller particles due to nucleation: if less ions are near the nucleus, the particles grow smaller. It can be achieved by lowering the concentration of the precursor solution or by a smaller flow rate. The zeta potential is not changed because the surface chemistry does not change.

The concentration of hydroxide ions has no strong influence on the diameter (cf. Fig. 4c, pink). Still, the zeta potential decreases with decrease in hydroxide concentration because of missing oxygen on the surface. On the other hand, the size of magnetite particles strongly depends on the concentration of hydrochloride acid in the precursor solution (cf. Fig. 4c, blue). When less hydrochloric acid is used in the precursor solution, the iron ions form small clusters that can act as nuclei. The particle formation is faster at existing nuclei, so less but larger particles are formed. Also, the zeta potential is larger if a higher concentration of hydrochloric acid is used because more oxygen-containing groups on the surface are formed.

The reactor geometry has no significant influence on the particle size, like Fig. 4d shows. With the same recipe, the particles grow a little bit smaller in the reactor with a smaller tube diameter (4 mm instead of 8 mm) and also a smaller volume (50 mL instead of 100 mL). This can be explained by the smaller retention time in the reactor. The particles have less time to merge with each other and the agglomerates grow smaller. In addition, the variance of diethylene glycol has no significant dependence in the objected concentration.

Next, the large-scale functionalization with TRS is investigated. As shown in Fig. 5, two amino acids (β-alanine, aspartic acid) are tested in the one-step reaction and five carboxylic acids (citric acid, malonic acid, tartaric acid, oxalic acid, cinnamic acid) are tested in the two-step synthesis. Now, the particles are analyzed at pH 7 because they show the highest zeta potential at these pH value.

Fig. 5
figure 5

The functionalization of the magnetite particles was analyzed with the particle diameter (blue/pink), the zeta potential (purple) and TGA (green/orange). The DLS measurements were performed at pH 7 or pH 10 depending on where the particles are most stable. First, different thermo-responsive substances are tested (a, d). Second, the concentration of citric acid in the functionalization reaction is varied (b, e). Third, the synthesis parameters of the functionalization with citric acid are varied (c, f)

Full size image

The size of all MNPs is much larger than in the batch reactions. Again, aspartic acid stabilizes the MNPs the least (small zeta potential) and the particles agglomerate (Fig. 5a). The particles functionalized with citric and cinnamic acid are also larger than the aimed 500 nm, even though the citric acid functionalized particles have a high zeta potential. The particles functionalized with β-alanine, malonic, tartaric and oxalic acid have d90 values of 300 nm and smaller, so they can be used for spinning experiments. Interestingly, the amount of immobilized TRS, as determined by TGA, is higher for larger particles, like Fig. 5d shows. According to stabilization mechanisms, more TRS should stabilize the particles better and the particle size should decrease. This contradictory behavior of the particle size can originate in the much higher concentration. Most used TRS have two carboxylic acid groups or more. In higher concentrations, it is more likely that two carboxy groups of the same molecule coordinate to two different MNPs especially if they are agglomerated before.

Changing the concentration of the TRS affects neither the particle size nor the immobilization of TRS (cf. Fig. 5b and e). Varying the pH or turning off the ultrasonic bath does not relevantly affect the immobilization of TRS. Contrarily, the particle size differs greatly. If no ultrasound is used during the functionalization, the particle size depends on the pH value. The smallest particles occur if the precursor solution has pH 10 due to stronger electrostatic repulsion. If ultrasound is used, the particles functionalized in acid or alkaline have nearly the same size. Particles functionalized in a pH neutral buffer show a larger particle size. One hypothesis suggests that the buffer inhibits the coordination of TRS with the particles, while another proposes that the buffer, which is more heat-resistant, coordinates with the magnetite.

Although different TRS and synthesis routes are tested, the up-scaling to a continuous flow reactor was not successful. The amount of immobilized TRS is strongly decreased compared with the batch experiments while the size is increased in most experiments. Even if there are particles small enough to use in a sizing, all particles form much to less gases to trigger the debonding. Presumably, the decrease in immobilization can be attributed to the change of solvent from diethylene glycol to water and lowered temperature from 100°C to 60°C. These changes were done because of the material compatibility of the custom-made reactor. For the one-step reactions, a mass flow of about 14 g/h is possible. Due to their poor results according to the functionalization, the space-time output was not optimized for the two-step synthesis and is only about 1 g/h. For pure magnetite, the mass flow is about 6 g/h.

Experimental fiber spinning

To test the delamination and recycling process, glass fibers coated with TRS are spun. Therefore, sizings with different amount of functionalized magnetite nanoparticles (MNPs) are prepared. The spun fibers are analyzed via SEM imaging. To analyze the delamination behavior, single fibers are embedded in polymer and pull-out tests are performed. In these tests, the force required to pull the fiber out of the plastic is measured. The measured force is normalized with the fiber diameter and the embedding depth, so the interlaminar shear strength (ILSS) is calculated as the relevant parameter.

In a first spinning experiment, gained core-shell particles from MNPs with aspartic acid were incorporated into a sizing mixture and applied onto glass fibers during the spinning process. For further stabilization, a dispersing additive (DISPERBYK 194 N) is used. Aspartic acid was chosen, because the respective particles showed the highest weight loss of all compared samples in the previous experiments. Three particle volume concentrations were tested, namely 0.1 wt.%, 0.5 wt.% and 1.0 wt.%.

The SEM images (Fig. 6a–d) show the spun fibers with different amounts of particles on their surface. Fibers which are coated only by polymer without particles have a nearly perfectly flat surface without structures (Fig. 6a). If 0.1 wt.% particles are added to the sizing, isolated particles are found on the surface of the fibers (Fig. 6b). The particles do not form large agglomerates or cover a large part of the surface. If the particle concentration in the sizing is increased to 0.5 wt.%, the surface coverage increases, significantly (Fig. 6c). Additionally, first agglomerates are visible which is not favored. If the particle concentration is increased even further, the agglomeration increases as well. At the same time, the surface coverage decreases again (Fig. 6d).

Fig. 6
figure 6

SEM images of spun glass fibers with sizing containing 0.0 wt.% particles (a), 0.1 wt.% particles (b), 0.5 wt.% particles (c) and 1.0 wt.% particles (d). Additionally, the interlaminar shear strength is displayed (e)

Full size image

These microscopic results are plausibly observed at the macroscopic scale. The main purpose of a sizing is to bundle and fix the individual filaments into a roving. The degree of fixation of the filaments is referred to as roving integrity. If the roving integrity is small, the roving breaks down into the individual filaments which is not desired as it is difficult to process.

The obtained rovings from the experiments indeed show different integrities: The roving without particles or only with 0.1 wt.% particles has a very good roving integrity. If 0.5 wt.% particles are added to the sizing, the roving integrity is lowered but the roving can still be handled. However, if the particle concentration is increased to 1.0 wt.%, the roving loses integrity and breaks into its single filaments. This behavior of the roving integrity correlates nicely with the SEM images: Without particles, the polymer typically accumulates in that area, causing the integrity of the roving. When a high concentration of particles is used, they tend to form large agglomerates in the spaces between the fibers and the polymer cannot keep up the roving integrity.

To explain this result, it is necessary to look at the sizing process. In the state of the art, sizings do not contain any particles but mainly consist of about 10 wt.% of a solved or dispersed polymer in water. If the roving is coated with such a sizing in the spinning process, the single filaments adhere to each other because of capillary forces. If the roving is dried, the water evaporates, and polymer film remains. According to the small concentration, this polymer film is very thin, as shown in Fig. 6a. If particles are added to a sizing, it is possible that they disturb this thin coating. Consequently, the roving integrity would decrease. If more particles are added, their tendency toward agglomeration increases. Such agglomerates would mostly be located at the free spaces between fibers and consequently disrupt the polymer film. Consequently, the fibers do not adhere to each other anymore and the roving integrity decreases strongly. One idea to overcome this lowered roving integrity is to immobilize the particles on the surface of the fibers by covalent bonds. The particles are thus not movable anymore, and they do not agglomerate between the fibers during drying. This idea is currently a working hypothesis and would have to be investigated in future experiments.

Next, the mentioned pull-out tests were performed with embedded fibers. Contrary to the expectation, ILSS of all tested samples increases after heating, like Fig. 6e shows. This undesired result might be caused by incomplete crosslinking of the matrix. Once the samples are tempered while trying to trigger the debonding mechanism, further crosslinking occurs, resulting in increased ILSS. Comparing the samples with and without particles, there are no significant changes in the ILSS. This is possibly due to the small overall number of particles in the sizing and even less particles on the fibers. The few correctly aligned particles furthermore currently consist mostly of the inert core. These summarized effects give the plausible assumption that the available amount of released gases is by far insufficient to trigger the debonding mechanism. A much more thermolabile substance would have to be placed at the interfaces.

Even more peculiar, the particles even tend to decrease the ILSS slightly before heating. This might be caused by the nature of the particles: If the particles are not covalently bound to the fibers, the particles might retain some degree of mobility. Thus, the friction between fiber and matrix is decreased, and consequently, the ILSS is decreased, too. However, this effect is not significant, as all values remain within the margin of error.

Conclusions

The approach of filling hollow spheres to generate nanocapsules results in a very small weight loss (2.2 wt.%) while heating. This is caused by poor diffusion of the TRS into the capsules. A direct encapsulation is not possible due to the harsh limitation of size. Consequently, these nanocapsules are not suitable to trigger the DoD.

Therefore, immobilizing thermo-responsive substances (TRS) on the surface of magnetite nanoparticles (MNP) was tried. Two synthesis routes known by literature were tested, one specific for amino acids31 and another for carboxylic acids.27 In the batch reactor, the immobilization of TRS worked very well with an amount of organic compound with 15–20 wt.%. Also, most particles are smaller than the limit of 500 nm. Particles functionalized with β-alanine or citric acid show the best results, while particles functionalized with aspartic acid show the worst results.

In the materials from the continuous scale-up experiments, the number of organic compounds was maximum 15 wt.% which might be caused by the change of solvent or the increased temperature or both. Also, the particle size is increased in nearly all syntheses, and some particle species are even larger than 500 nm. This is explained by the TRS being less bound to the magnetite, which stabilizes the particles the worst. Interestingly, the particles which show the best results in the batch experiments (β-alanine or citric acid) are nearly the worst syntheses according to bound TRS and particle size. On the other hand, the particles functionalized with aspartic acid are here one of the best particle species. Therefore, they are used for spinning experiments testing the DoD.

Although the functionalization works, there is no observable effect on the interlaminar shear strength visible. Presumably, too little gas is formed to induce effective delamination. The used particles only contain about 3 wt.% organic compounds (the relative amount decreases due to the silica top layer). Combined with the small filling levels of the particles in the sizing, there is too little internal pressure forming micro cracks and delamination. Even more, the observed post-curing effects of the polymer matrix outweigh any potential effect, that might remain hidden, by far.

To overcome these mentioned drawbacks, the particles must be immobilized on the fibers’ surfaces to maintain the roving integrity. If the particles are covalently bonded to the surface, it should be possible to increase the particle concentration in the sizing without agglomeration of the particles between the fibers. On the other hand, the amount of immobilized TRS on the MNPs must be increased drastically to enable the desired recycling mechanism.