Sage Journals: Discover world-class research

Abstract

The automotive industry is engaged on the development of new vehicles with more sustainable and lightweight materials. Recent manufacturing technologies are being developed to respond to these challenges. Thermoplastic-Resin Transfer Moulding (T-RTM), a liquid moulding technology, emerged in this field has a promising approach for the manufacture of lightweight thermoplastic composites for structural applications. This process is based on the in-situ polymerisation of a thermoplastic monomer. This material is suitable for impregnation of continuous fibres and is prone to recycling, in opposition to conventional thermosetting composites. However, T-RTM remains a complex manufacturing technology mainly due to the instability of the resin in the presence of moisture or oxygen as well as thermal and pressure gradients within the mould. These gradients lead to non-uniform initial conditions for polymerisation along the mould cavity and, consequently, to a considerable variability in the T-RTM process. This study aims to contribute to the upscaling of this technology towards its industrial application, through the implementation of an innovative dynamic resin mixing and vibrational apparatuses designed to increase the homogeneity of the resin mixture after its injection into the mould. The parts’ homogeneity and process reproducibility were evaluated based on the statistical analysis of the mechanical and physical properties assessed in different areas of glass fibre-reinforced parts. This study has shown that both apparatuses can improve homogeneity regarding the mechanical behaviour and density of the parts. Furthermore, the dynamic mixing enhanced the polymerisation reactivity, assessed through dielectric analysis, compared to the standard T-RTM manufacturing method.

Keywords

Polyamide 6 matrix structural thermoplastic composite T-RTM process variability homogeneity dynamic mixing apparatus vibrational apparatus

Introduction

The increase in Global Greenhouse Gas (GHG) emissions can affect the world´s biodiversity and natural resources. Aware of this problem, the automotive has fostered innovation strategies to reduce GHG emissions, with one of the focuses being on new manufacturing technologies to produce vehicles with more sustainable and lightweight materials.¹ With this purpose, polymer-based materials have been increasingly used in road vehicles due to their low density and high resistance to fatigue and corrosion.^1,2 For structural applications fibre-reinforced composite materials based on thermosetting polymers have been manufactured by Liquid Composite Moulding (LCM) technologies, however, these materials cannot be easily recycled^3,4 which leads to end-of-life problems: landfilling thermoset materials is not a sustainable solution.^4,5

Thermoplastic-Resin Transfer Moulding (T-RTM) is an LCM technology under development to produce lightweight thermoplastic composites with continuous reinforcement fibres. One of the distinctive features of T-RTM is the injection of a monomer with lower viscosity suitable for the impregnation of the fibres and its subsequent polymerisation inside a mould. To produce polyamide 6 (PA6), an exotherm reaction occurs through the anionic ring-opening polymerisation of the ε-caprolactam monomer. In the context of LCM technologies for structural composites, T-RTM has the advantage of recyclability,⁶ reducing the environmental impact of vehicle disposal without compromising relevant mechanical properties.^7,8

Although promising, this technology is not implemented at an industrial level. In addition to the common problems of composite manufacturing technologies, such as the presence of voids,⁹ the instability of the resin in the presence of moisture or oxygen⁹ and the thermal and pressure gradients⁷ of the process can cause a significative variability in the mechanical behaviour and density of the manufactured parts. Humidity and oxygen affect the polymerisation of the resin by deactivating the catalytic system.^10,11 The reaction with water or oxygen molecules can lead to the formation of secondary products which decrease the monomer conversion degree and the mechanical behaviour of the parts.¹² It is generally reported that it is important to obtain a monomer conversion degree (M_CD) above 95%, to ensure compatibility and adhesion between the fibre and the matrix.¹³ Even when the raw materials are being transferred to the equipment container, a relative humidity of more than 45% can promote significant moisture absorption.¹⁰ Furthermore, partial polymerisation in the mixing and injection phase, due to a thermally uncontrolled process can lead to clogging issues and high maintenance and repair costs, particularly for complex technological subsystems such as mixing heads.^7,9

Although the T-RTM variability is one of the factors limiting the upscaling of this technology, a lack of research in this field remains. Although several articles on T-RTM can be found in the literature, few address or demonstrate the robustness of the manufacturing processes employed through statistical analysis. Considering that variability in the production of such composites can be significant, the assessment of process robustness is critical towards industrial application, specifically in the manufacturing of structural components for the automotive industry. In fact, statistical methods like parametric and non-parametric two sample tests, ANOVA, Dunnett’s T3 pairwise comparison test, among others, have been used by several authors to assess the significance of the effect of manufacturing parameters on the variability of material properties,^14–16 and thus, strengthen the conclusions. Most of these statistical techniques demand that a set of assumptions about the data are met for a correct and adequate interpretation of the corresponding results. For example, the ANOVA class of methods used to compare means across different conditions usually assume normality and homoscedasticity of the data.^17,18

However, sample means can have relatively large standard errors even for slight departures from normality.¹⁹ In fact, heavy tail or asymmetrical distributions increase the probability of sampling outliers, with direct impact in the sample means standard errors. This in turn can signify a lower power of the statistical method to detect differences. The presence of outliers is also a concern, as the sample estimates may be affected by these extreme values masking the fundamental differences and giving rise to unreliable conclusions.²⁰ An appealing statistical alternative is the class of robust statistical methods, specifically designed to cope with violations of the assumptions about the data distribution such as normality, homoscedasticity and sphericity¹⁸ or in the presence of outliers. These methods rely on robust estimates of the sample measures (location and scale) and bootstrap sampling (sampling with replacement) to alleviate the effects of outliers and deviations from the assumptions and include robust tests for two independent or dependent samples, robust one-way, two-way and three-way ANOVA or even robust tests for repeated measurements or mixed ANOVA designs.^19,20

This work aims to contribute to the development of a distinct single stream methodology for the T-RTM process. The study focused on validating two innovative resin homogenisation systems within the mould, in order to minimize process variability, primary arising from thermal and pressure gradients within the mould cavity after resin injection, which strongly influence in situ polymerisation. A dynamic resin mixing and vibrational apparatus were incorporated into the manufacturing process to promote a more uniform polymerisation of the material throughout the mould cavity. The dynamic resin mixing unit consists of an external device responsible for inducing the resin’s reciprocal linear movement within the mould. The vibrational device promotes the vibration of the resin inside the mould cavity. This work assesses the effect of the in situ homogenisation apparatuses on the variability of T-RTM based on the statistical analysis of the mechanical behaviour and density in several areas of the manufactured parts. Due to the presence of outliers in the collected data and the violation of the assumptions of the usual techniques to compare groups, a robust mixed ANOVA was considered in our analysis.

Material and methods

Materials

AP-NYLON® caprolactam flakes monomer (CL), BRUGGOLEN® C1 catalyst (C1) and BRUGGOLEN® C20P activator (C20P) (L. BRÜGGEMANN GmbH & Co. KG) were used in a ratio of 85:10:5 wt%.

For each produced part, two layers of non-crimp glass fibres fabric (GF) with 573 g/m² were used as reinforcements to obtain a ≈30% Fibre Volume Content (F_Vol%).

T-RTM prototype equipment and manufacturing systems

The parts were manufactured on a prototype T-RTM laboratory equipment developed in our lab. A schematic draw of the equipment is presented in Figure 1. In this equipment, the resin is driven exclusively by positive and negative nitrogen pressure. A vacuum pump is used to generate negative nitrogen pressure in the mould while positive pressure is supplied by a nitrogen pressurised bottle during injection and resin packing. A dynamic mixing unit was settled to promote the reciprocal back-and-forth movement of the resin within the mould. The system was placed in a controlled lab environment with a temperature of 26 ± 3°C and a relative air humidity below 45%.

Figure 1.

Scheme of the prototype T-RTM equipment with mould.

Since oxygen and moisture (≤55% of relative air humidity) can hinder the polymerisation,¹⁰ the first stage of the process consisted of purging and pressurising the equipment with nitrogen gas.

The CL, C1 and C20P flakes were placed in a single container and heated for 10 min at 95°C. After melting, the raw materials were mixed by mechanical stirring at 250 rpm for 5 min. The processing conditions were adjusted so that no significant variation in the resin’s viscosity was detected. The resin viscosity was evaluated using a dielectric sensor coupled to the tank.

A single-cavity mould was developed to manufacture parts with a plate-shaped geometry, with a cavity of 120 × 75 × 1.8 mm³ (Figure 2). The cavity had a U-shaped ledge geometry to clamp the woven fibres. The setpoint temperature of the mould was 160°C and it was heated using electric cartridge heaters, controlled by a PID RS PRO 798-3472. The mould was sealed with double O-rings for vacuum purposes and due to the low resin viscosity. Netzsch TMM10c dielectric sensors coupled to a Netzsch DEA 288 Ionic dielectric analyser were inserted into the mould cavity to monitor the polymerisation reaction. To produce the parts, two GF layers were placed inside the mould cavity, corresponding to roughly 30% of F_vol%.

Figure 2.

Image of the Netzsch DEA 288 Ionic dielectric analyser and the mould cavity with the U-shape geometry.

In the next stage, the mould was set under vacuum at 0.15 bar. Before the injection, the molten resin was transferred to the mould inlet through a hose. The injection phase started by opening the mould inlet valve at a nitrogen pressure of 2 bar, for 30 seconds.

A 30-minutes polymerisation stage was set for all the parts. During polymerisation, a nitrogen gas packing pressure of 3 bar was applied for 2 minutes at the mould inlet. A second packing pressure stage was implemented to counteract material shrinkage and hinder void formation. It consisted of applying a nitrogen gas packing pressure of 3.5 bar at the mould outlet until the end of the polymerisation stage.²¹

Table 1 displays the nomenclature assigned to the parts based on the three manufacturing systems used.

Table 1.

Parts nomenclature according to the manufacturing system used.

Nomenclature of the manufactured parts	Manufacturing system
SSPP	Second Stage Packing Pressure
DMA	SSPP + Dynamic Mixing Apparatus
VA	SSPP + Vibrational Apparatus

Dynamic Mixing Apparatus (DMA) and Vibrational Apparatus (VA) introduce a new stage to the manufacturing process between the first and second packing pressures.

For the manufacturing process using DMA, a dynamic mixing unit was used to create a reciprocal movement of the resin inside the mould cavity after injection (in-situ), to enhance physical and thermal homogenisation while improving the impregnation of the reinforcing fibres. Laboratory tests using dye traces to evaluate the mixing behaviour of the resin indicated that the apparatus effectively promoted resin homogenisation inside the mould. This effect was more pronounced in the presence of reinforcing fibres. It was therefore concluded that the presence of fibres, combined with the resin’s reciprocal movement, constitutes a positive factor for enhancing the homogenization of the resin. The reciprocal movement consisted of displacing the resin from the inlet to the outlet of the mould and vice versa (Figure 3). The resin flow was set to 105 mL/min for 5 minutes at 0.2 Hz.

Figure 3.

Scheme of the DMA resin reciprocal movement from (a) mould inlet to mould outlet and (b) Mould outlet to mould inlet (inverted movement).

The implementation of a VA within the process aims to increase the composite physical homogenisation by removing trapped nitrogen bubbles prior to the resin polymerisation. A frequency of 100 Hz was set at this stage for 10 seconds. The frequency data was acquired by a Kistler LabAmb 5165A charge amplifier with a Kistler 8763B050AB accelerometer.

The processing parameters were set based on preliminary trials and dielectric measurements. Five parts were produced for each manufacturing condition.

Materials characterization

The location of the specimens and samples that were used to characterise the manufactured parts is shown in Figure 4. For each part, six specimens were used for the determination of the maximum tensile strength (σ_m) Young´s modulus (E_t), tensile strain at break (ε_b) and geometric density (ρ_g). Archimedes’ density (ρ_a), resin volume content (R_Vol%), F_Vol%, void volume content (V_Vol%) and monomer conversion degree (M_CD) were assessed from three different areas of each part. Relative monomer conversion degree (rM_CD) was measured at the inlet area of the parts.

Figure 4.

Location of samples and specimens on composite parts.

A representative image of the cut specimens manufactured using DMA is in Figure 5.

Figure 5.

Example of DMA specimens after cutting.

Mechanical and physical behaviour of the specimens

The mechanical tensile tests were based on ISO 527 and performed on Shimadzu AG-IS 10 kN. A 2 mm/min speed was used for 70 × 7 × 1.8 mm³ specimens with a 14 mm gauge length.

During the test, a video extensometer was used to measure the elongation in the gauge length (ΔL₀). The ΔL₀ was used to determine E_t and ε_b.

The ρ_g was assessed by measuring the width and thickness of each specimen in three different places: on both edges and at their centre. Width and thickness were measured using a Schut Geometrische Meettechniek b.v. Filetta micrometre. The length was measured using an INSIZE Co., Ltd 1108-150 calliper.

The specimens were weighed on an OHAUS Europe GmbH Adventurer AX224 analytic balance and the following equation (1) was used for the ρ_g calculation:

ρ_{g} ({g / c m}^{3}) = \frac{\bar{w} \times \bar{t} \times l}{m}

(1)

In which.

$\bar{w}$ – mean width of the specimen;

$\bar{t}$ – mean thickness of the specimen;

$l$ – length of the specimen;

m – weight of the specimen.

Statistical analysis

A thorough statistical analysis is applied to the experimental results to look for differences between the three methods, SSPP, DMA and VA, with respect to their geometric density and mechanical behaviour. A robust mixed ANOVA is used to look for significant differences on mean, while variability is assessed both through homogeneity and reproducibility of the results using coefficients of variation.

Differences on mean between SSPP, DMA and VA

This study corresponds to a repeated measures design as several measurements were collected from different areas of the same part. It is expected some variability in measurement depending on the area so, mixed ANOVA type of designs is the most appropriate to compare the three manufacturing systems in our case.²² In mixed designs a factor of between-subjects variation (the manufacturing system) is analysed while controlling for the within-subjects variability (repeated measurements) and eventual interactions between both. The method has the following data assumptions: normality of the groups, homoscedasticity and sphericity.¹ The absence of (at least severe) outliers is also beneficial. Regarding normality of the groups, the Shapiro-Wilk test²³ was used for each group of observations defined by each pair (Manufacturing system, Area). Homoscedasticity between manufacturing system groups was assessed using the Levene test²⁴ per Area), while the sphericity assumption was tested using the Mauchly’s test.²⁵

The violation of the mixed ANOVA assumptions can negatively impact its application and the corresponding results. For example, slight departures from normality are known to increase the standard error of the sample estimates¹⁹; similarly, a lack of homoscedasticity can inflate the false positive rate,²⁶ or violations of sphericity inflate the Type I error rate, producing artificially small p-values^27,28. This means that Type I or Type II errors can be inflated, bringing lower power of the statistical method to detect true differences.²⁰ The presence of outliers also affects sample estimates, eventually masking the fundamental differences and giving rise to unreliable conclusions.²⁰

Robust ANOVA methods stand as an attractive alternative to alleviate the impacts of assumption violation or presence of outliers. Here, the tools from the WRS2 package of the R statistical software were used.²² The robust version of the mixed ANOVA design considers trimmed means as the robust measure of location (instead of the usual mean) as well as Winsorized versions of the variances and covariance matrices.²⁰ Essentially, a trimmed mean is simply a mean where a given proportion of observations from both tails of the ordered sample are discarded, while the “Winsorizing” process sets such observations to the smallest value not trimmed at the left or the largest value not trimmed at the right, respectively. The WRS2 package uses a 20% trimmed mean (10% in each tail, that is, data below the lower 10th percentile and above the upper 10th percentile of the sample are discarded) as it achieves nearly the same amount of power as the mean, when sampling from a normal distribution, as well as a substantially smaller standard error due to the elimination of outliers.²⁰ If differences are found in the mixed design, post-hoc tests are also provided independently for the between-subjects effect (manufacturing system), the within-subjects effect (repeated measurements) and the interaction effect. These tests use bootstrap-based approaches to compute M-estimators of the pairwise comparisons between groups (see 20 for details). The advantage of using bootstrap-based approaches is that there is no need to assume any a priori distribution for our data.

All tests were concluded at a significance level of 5%.

Variability assessment

A part is considered homogeneous for a specific mechanical or physical property when there is little or no variation, across different areas of the part being analysed. In this work, the homogeneity of each property was evaluated using the mean coefficient of variation (from areas 1-6), considering 5 parts produced by the same manufacturing system.

A processing manufacturing system can be considered reproducible for a specific mechanical or physical property, for a given region, when there is little or no variation, regardless of the part being analysed. For each property, the process reproducibility was evaluated through the mean coefficient of variation (from 5 parts produced by the same manufacturing system), considering the 6 areas analysed in each part.

Figure 6 summarises the method used to assess the homogeneity of the parts and the reproducibility of the process by considering the following mechanical and physical properties: σ_m, E_t, ε_b and ρ_g.

Figure 6.

Scheme of the methodology used to assess the homogeneity and reproducibility of the parts.

Let $M_{i j}$ represent the specific property (σ_m, E_t, ε_b or ρ_g) measured at area $A_{j}$ of part $P_{i}$ . To assess the homogeneity of the manufactured parts, for each manufacturing system (SSPP, DMA and VA) the mean coefficient of variation across plates (sP) was computed through equations (2)–(4).

s P (%) = \frac{\sum_{i = 1}^{N P} ({s P}_{i} (%))}{N P}

(2)

{s P}_{i} (%) = \frac{\sqrt{\frac{\sum_{j = 1}^{N A} {(M_{i j} - u_{P_{i}})}^{2}}{N A - 1}}}{u_{P_{i}}} \times 100

(3)

u_{P_{i}} = \frac{\sum_{j = 1}^{N A} M_{i j}}{N A}

(4)

in which;

N P

– total number of parts;

${sP}_{i}$ – coefficient of variation of the property measured in part $P_{i}$ ;

$NA$ - total number of areas;

$u_{P_{i}}$ - mean property measured in part $P_{i}$ .

Using a similar approach, the mean coefficient of variation across areas (sA) was used to assess the part-to-part reproducibility²⁹ of T-RTM. It was computed through equations (5)–(7).

s A (%) = \frac{\sum_{j = 1}^{N A} ({s A}_{j} (%))}{N A}

(5)

{s A}_{j} (%) = \frac{\sqrt{\frac{\sum_{i = 1}^{N P} {(M_{i j} - u_{A_{j}})}^{2}}{N P - 1}}}{u_{A_{j}}} \times 100

(6)

u_{A_{j}} = \frac{\sum_{i = 1}^{N P} ({s A}_{j} (%))}{N P}

(7)

in which;

$s A_{j}$ – coefficient of variation of the property measured in area $A_{j}$ ;

$u_{A_{j}}$ - mean property measured in area $A_{j}$ .

Archimedes’ density, resin, fibres and void volume content

The ρ_a of specimens were measured based on ISO 1183, through immersion in deionised water. An Ohaus Adventurer AX224 analytic balance was used with its density determination kit. The AG was measured in the three different areas of each part: areas A, B and C (Figure 4).

The parts’ resin, fibres and void volume content were determined by the ISO 7822:1999 – Method A standard, also known as the burn-off technique. The samples were placed in a Termolab MLM furnace at room temperature, heated at 10°C/min to 560°C with a 2-hour plateau to eliminate the organic phase. The samples were then naturally cooled to room temperature and weighed with an Ohaus Adventurer AX224 analytic balance. The fibre weight content (F_W%) was calculated using the ratio between the final and initial weight of the samples. F_Vol% was calculated according to equation (8), where ρ_measured is the composite density and ρ_GF is the density of GF.

F_{V o l %} = \frac{{F_{W %} \times ρ}_{m e a s u r e d}}{ρ_{G F}} \times 100

(8)

Equation (9) determined the V_Vol%. The ρ_theoretical is the theoretical density estimated with F_vol% and the R_vol%.

V_{V o l %} = (\frac{ρ_{t h e o r e t i c a l} - ρ_{m e a s u r e d}}{ρ_{t h e o r e t i c a l}}) \times 100

(9)

The R_Vol% was then calculated using equation (10).

R_{V o l %} = 100 - F_{V o l %} - V_{V o l %}

(10)

Monomer conversion degree

Thermogravimetric analysis was performed to obtain information about the M_CD. The analysis was done on a Hitachi STA300 equipment with open aluminium pans. The temperature was set between 25°C and 550°C using a 10°C/min heating rate. A nitrogen gas flow of 200 mL/min was used to provide an inert atmosphere. The monomer conversion degree was calculated according to equation (11).³⁰

M_{C D} (%) = [1 - (\frac{{w l}_{240 ℃} - {w l}_{100 ℃}}{w_{i n i t i a l} - {w l}_{100 ℃}})] \times 100

(11)

where wl_100°C and wl_240°C are the sample weight loss at 100°C and 240°C, respectively. The m_initial is the initial weight of each sample.

The rM_CD data was obtained using a Netzsch TMM10c dielectric sensor mounted at the mould inlet using a 10 kHz frequency. The sensor was coupled to a DEA 288 Ionic dielectric analyser to monitor the polymerisation reaction. The sensor also has an integrated J-type thermocouple. The parts manufactured using the VA manufacturing system were not subjected to dielectric analysis due to the possibility of the vibration damaging the dielectric sensors. With the dielectric sensor, it was possible to determine.

• The mould temperature at injection (time = 0 min);

• The maximum resin temperature after injection;

• The difference between the referred temperatures (ΔT);

• The minimum and maximum log. Ionic Viscosity (IV) after injection³¹;

• The difference between the referred log. IV;

• The time (from injection) at which polymerisation started, set as the polymerisation time at which the log. IV is lower;

• The time (from injection) at which the reactivity of the resin was higher;

• The time (from injection) at which the polymerisation finished, based on the analysis of the smoothed first derivative curve of the log. IV;

• The maximum reactivity, calculated based on the maximum value of the smoothed first derivative curve of the log. IV;

• The mean viscosity, calculated based on the mean slope of the log. IV curve.

The minimum and maximum log. IV values correspond to a rM_CD of 0 and 100%, respectively. For clarity, before the minimum log. IV time the rM_CD value was considered 0%.

Results and discussion

The mechanical behaviour (σ_m, E_t and ε_b), density (ρ_g and ρ_a), polymer, fibres and void volume content (R_vol%, F_vol% and V_vol%) and monomer conversion degree (M_CD and rM_CD) of the parts manufactured by T-RTM are going to be evaluated in this section.

Mechanical and physical behaviour of the specimens

There is always some variability in the data collected from the mechanical tests and density measurements. This can be due to operator´s measurement errors, as well to some raw materials inhomogeneities that always exist, even within the same batch of material. Glass woven fibres are also known for having defects along the fabric.³²

Some heterogeneities were visible in the GF fabric used (Figure 7).

Figure 7.

Example of visible heterogeneities in the GF used.

Difficulties in GF fabric impregnation, void formation and deformation of the reinforcing fibres can also occur, increasing the variability of manufactured composite parts.¹³

Figure 8 shows the σ_m, E_t, ε_b and ρ_g boxplots for the three different manufacturing systems, categorised by area. Although each property, on average, tends to be similar across the three manufacturing systems, it was clear that DMA and VA yield a reduction in variability in most cases, demanding a more focused analysis to assess if those differences are statistically significative. We also observed several outliers and skewed distributions, which does not line up with the ANOVA assumption of normality within groups.

Figure 8.

Mechanical and physical properties of composite parts for each manufacturing method: (a) tensile strength (σ_m), (b) Young´s modulus (E_t), (c) tensile strain at break (ε_b) and (d) geometric density (ρ_d).

Statistical analysis

We divide our analysis in two different aspects. First, we look for statistical differences in the mean between the three manufacturing systems using ANOVA methods. Second, we assess more thoroughly the differences in variability found in Figure 8.

Differences in the mean Between SSPP, DMA and VA

We started to analyse the possibility of applying a parametric mixed ANOVA method to our data to look for differences between the three manufacturing systems (SSPP, DMA and VA).

Existence of outliers (Figure 8) and skewed distributions found in our data are known to generally affect further statistical analysis, therefore we relied on robust alternatives. Data assumptions were also violated in most situations, as Table 2 demonstrates.

Table 2.

Mixed ANOVA assumptions verification.

	σ_m	E_t	ε_b	ρ_g
Normality	✓	✗	✗	✗
Homoscedasticity	✓	✓	✓	✓
Sphericity	✓	✗	✗	✗

We see that only σ_m satisfied all the assumptions of a mixed ANOVA (in fact, data from the σ_m of composites is generally known to follow a normal distribution³³). Interestingly, homoscedasticity (analysed by considering the effect of the manufacturing system and area as demanded by the mixed ANOVA model) was verified in all the variables, but normality and sphericity failed for most cases. Given the lack of assumption validation, the existence of outliers and skewed distributions, testing for differences between the three manufacturing systems was performed using a robust version of the mixed ANOVA method, as described in statistical analysis section. The results are shown in Table 3.

Table 3.

Robust mixed ANOVA p-values for the main effects of manufacturing system and area and for the interaction manufacturing system: area.

Effect	σ_m	E_t	ε_b	ρ_g
Manufacturing system	0.4996	0.8671	0.2662	0.0273*
Area	0.005*	0.0185*	0.0046*	0.0021*
Manufacturing system:area	0.0258*	0.3846	0.0249*	0.0129*

Significant differences at the 5% level are marked with *. Supplemental Information 1.

The results from the robust mixed ANOVA indicate that no significant differences between the manufacturing systems were found for σ_m, E_t, ε_b mean values, but a significant difference was found for ρ_g. For this case, we performed post-hoc tests to look for specific differences. The WRS2 package provides a p-value for the test with the null hypothesis $H_{0} : θ_{S S P P - D M A} = θ_{S S P P - V A} = θ_{D M A - V A}$ where $θ$ represents some robust estimator of the difference in ρ_g between the manufacturing systems. The estimates ${\hat{θ}}_{S S P P - D M A} = 0.042$ , ${\hat{θ}}_{S S P P - V A} = 0.041$ and ${\hat{θ}}_{D M A - V A} = - 0.001$ and the p-value = .018 show evidence of a significant difference between SSPP and the other two systems, the former presenting higher values of ρ_g.

Significant differences between the measurement sites (area) were found for every variable. Significant interactions between the manufacturing system and the area were found for σ_m, ε_b and ρ_g. These results are according to Figure 8.

Although it is not being the central focus of this paper, it was interesting to observe that differences in the Area were statistically confirmed, while interaction terms also occur (meaning that manufacturing systems behave differently across the plate).

Variability assessment

The variability can be caused by the lack of parts homogeneity or the lack of process reproducibility. Thus, sP and sA values for each processing condition are presented in Figure 9. The results suggest that the DMA and VA can particularly benefit the homogeneity of the parts since, in those cases, the sP tended to be lower.

Figure 9.

Mean coefficient of variation for mechanical and physical properties of the composite parts: (a) in each part (sP) and (b) from part to part (sA) considering tensile strength (σ_m), Young’s Modulus (E_t), tensile strain at break (ε_b) and geometric density (ρ_g).

The differences in the homogeneity of the parts are also visible in Table 4 and Table 5. Analysing the results in the different areas, the parts tend to have a more heterogeneous mechanical and physical behaviour with the SSPP system.

Table 4.

Mean (M) and standard deviation (SD) for tensile strength (σ_m) and Young’s modulus (E_t) by part areas.

	σ_m (MPa)			E_t (GPa)
Area	SSPP	DMA	VA	SSPP	DMA	VA
1	225	261	256	9	12	14
2	223	270	290	9	12	13
3	317	319	298	16	14	15
4	424	333	299	19	16	15
5	358	350	324	18	16	16
6	334	366	343	17	15	18
Av ± SD	313 ± 78	317 ± 43	302 ± 30	15 ± 4	14 ± 2	15 ± 2

Table 5.

Mean (M) and standard deviation (SD) for tensile strain at break (ε_b) and geometric density (ρ_g) by part areas.

	ε_b (%)			ρ_g (g/cm³)
Area	SSPP	DMA	VA	SSPP	DMA	VA
1	4.0	3.9	3.0	1.50	1.50	1.51
2	4.1	4.5	3.9	1.51	1.52	1.52
3	3.0	3.4	3.8	1.62	1.56	1.55
4	3.8	3.5	3.4	1.67	1.58	1.56
5	2.4	3.3	3.3	1.58	1.56	1.57
6	2.6	3.4	3.0	1.59	1.56	1.57
M ± SD	3.3 ± 0.7	3.7 ± 0.5	3.4 ± 0.4	1.58 ± 0.06	1.55 ± 0.03	1.55 ± 0.03

As already analysed, the data in Figure 8 showed a tendency for the reduction in data variability when applying the DMA and VA manufacturing systems.

Archimedes’ density, resin, fibres and void volume content

The ρ_a of the composites is presented in Figure 10. The mean density values and higher ρ_a variability for the SSPP manufacturing system is in agreement with the ρ_g results.

Figure 10.

Archimedes’ densities (ρ_a) of the composite parts.

In Table 6, the two distinct density results can be compared by part areas.

Table 6.

Mean (M) and standard deviation (SD) for Archimedes’ density (ρ_a) and geometric density (ρ_g) by part areas.

Type of density (g/cm³)	Area	SSPP	DMA	VA
Archimedes’	A	1.51	1.53	1.53
Geometric	1	1.50	1.50	1.51
Geometric	2	1.51	1.52	1.52
Archimedes’	B	1.56	1.57	1.56
Geometric	3	1.62	1.56	1.55
Geometric	4	1.67	1.58	1.56
Archimedes’	C	1.68	1.61	1.58
Geometric	5	1.58	1.56	1.57
Geometric	6	1.59	1.56	1.57
Archimedes’	M ± SD	1.58 ± 0.08	1.57 ± 0.04	1.56 ± 0.04
Geometric	M ± SD	1.58 ± 0.07	1.55 ± 0.04	1.55 ± 0.04

When comparing the composite densities by area, it was possible to see differences depending on the manufacturing process used. As expected, a higher standard deviation was particularly noticeable for the SSPP parts.

The results reveal a trend towards a reduction of density in the inlet areas (areas A, 1 and 2) for all the manufactured parts. The density data suggests that the DMA and VA were able to homogenise the density in the centre and outlet areas of the parts. This can explain the higher homogeneity in the σ_m and ρ_g for DMA and VA.

The parts tend to exhibit a lower F_vol% in area A (Figure 4) compared to the remaining areas (Table 7). The assumption of GF fabric constant areal weight can no longer be valid during injection. A localised fibre distortion effect was likely to occur at the mould inlet due to the perpendicularity of the resin flow in relation to the alignment of the fabric. It should also be noted that the DMA and VA do not seem to correct this issue. The lower inlet F_vol% (area A in Table 7) and thus inlet densities (areas A, 1 and 2 in Table 6) can explain the lower mean σ_m and E_t in the part’s inlet (areas 1 and 2 in Table 4).

Table 7.

Mean (M) and standard deviation (SD) for Resin Volume Content (R_Vol%) and Fibre Volume Content (F_Vol%) by part areas.

Area	R_Vol%			F_Vol%
Area	SSPP	DMA	VA	SSPP	DMA	VA
A	75 ± 1	74 ± 1	74 ± 1	25 ± 1	26 ± 1	26 ± 1
B	72 ± 1	71 ± 1	72 ± 3	28 ± 1	29 ± 1	28 ± 3
C	63 ± 2	68 ± 2	70 ± 2	36 ± 2	32 ± 2	30 ± 2
M ± SD	70 ± 6	71 ± 3	72 ± 2	30 ± 6	29 ± 3	28 ± 2

It was also possible to identify a higher F_vol% in area C (Table 7) for SSPP parts. The distortion of the fabric fibres at the mould inlet may have resulted in a higher concentration of fibres in this area. This also caused a localised increase in density (areas 4 and C; Table 6) and σ_m (area 4; Table 4). However, for DMA and VA, an improvement in F_vol% SD within the part was noticeable (Table 7), thus a lower sP, density and mechanical behaviour variability was achieved (Table 6 and Table 4). The results indicate that DMA and VA contributed to mitigate this effect and to produce more homogeneous parts.

V_Vol% was below 1% for all the manufacturing systems (Table 8) due to the packing pressure stages applied during polymerisation.²¹ Depending on the application, the maximum recommended V_Vol% values should vary between 1 and 5%,^34–36 which supports the effectiveness of the process parameters and apparatus studied in this work.

Table 8.

Mean (M) and standard deviation (SD) void volume content (V_Vol%) in the analysed areas of the parts.

Area	V_Vol%
Area	SSPP	DMA	VA
A	0.4 ± 0.4	0.3 ± 0.2	0.3 ± 0.2
B	0.5 ± 0.2	0.3 ± 0.3	0.3 ± 0.1
C	0.5 ± 0.3	0.5 ± 0.3	0.4 ± 0.1
M ± SD	0.4 ± 0.1	0.4 ± 0.1	0.4 ± 0.1

Monomer conversion degree

All the samples in this study had a monomer conversion degree of around 98% with standard deviation values lower than 1% (Figure 11). It is generally reported in the literature that it is important to obtain an M_CD above 95% to achieve a suitable mechanical performance.¹³ These results show that the DMA and VA can be used in T-RTM.

Figure 11.

Monomer conversion degree (M_CD) of composites (a) by manufacturing system and (b) by part area.

The results obtained from the dielectric analysis suggest that the maximum and minimum log. IV values and the maximum reactivity time were similar between the SSPP and DMA (Table 9). However, differences were found in the time the polymerisation started and in the maximum and average reactivity of the resin. As mentioned, the parts manufactured using the VA were not subjected to dielectric analysis due to the possibility of damaging the dielectric sensors with the induced vibration.

Table 9.

Dielectric data of the resin after injection.

	SSPP	DMA
Mould Temperature at Injection (°C)	164.4 ± 0.5	161.7 ± 0.3
Max. Resin Temperature (°C)	166.8 ± 0.9	164.6 ± 0.2
ΔT (°C)	2.4 ± 0.6	2.9 ± 0.4
Min. Log IV (Ohm.cm)	5.1 ± 0.0	5.1 ± 0.0
Max. Log IV (Ohm.cm)	6.0 ± 0.0	6.1 ± 0.1
Δ Log IV (Ohm.cm)	0.9 ± 0.0	0.9 ± 0.1
Time at which the polymerisation started (min)	2.9 ± 0.2	3.7 ± 0.3
Max. Reactivity Time (min)	5.6 ± 1.2	6.2 ± 0.8
Time at which the polymerisation ended (min)	8.4 ± 1.2	7.5 ± 1.1
Max. Resin Reactivity (Ohm.cm/min)	0.40 ± 0.15	0.79 ± 0.13
Average Resin Reactivity (Ohm.cm/min)	0.17 ± 0.03	0.25 ± 0.04

In Figure 12 the representative curves of the rM_CD for SSPP and DMA as a function of the time after injection are shown. It is visible that the DMA delays the beginning of polymerisation by a minute. However, once the dynamic mixing stage was complete, the resin’s reactivity increased and the polymerisation ended more quickly likely due to improved physical and thermal homogenisation of the resin within the mould. Thus, it is possible to have shorter cycle times and a higher processing window with DMA.

Figure 12.

Representative curves of the relative monomer conversion degree (rM_CD) after injection.

Conclusions and future work

The tensile strength, Young’s modulus and tensile strain at break data suggested that although the mean values between the different manufacturing systems tend to be similar, there were differences in the variability of the results.

The results of the mean coefficient of variation analysis indicated that the dynamic mixing and vibrational apparatuses increased the manufacturing parts homogeneity, particularly regarding to their tensile strength, Young’s modulus and tensile strain at break.

Using the dynamic resin mixing and vibrational apparatus a more uniform resin and fibre volume contents were found. An improvement in the homogeneity of the mechanical tensile performance of the parts was particularly noticeable. All the manufactured parts had a void volume content below 1%.

All the manufacturing systems achieved a monomer conversion degree above 97%. This is particularly important because although the dynamic resin mixing and vibration occurred during polymerisation, the results showed that they do not negatively affect the monomer conversion degree. Furthermore, the dielectric data showed that the average and maximum reactivities were higher using the dynamic mixing apparatus when compared to the second-stage packing pressure alone.

In future work, a dynamic mixing apparatus should be used to study its variability effect in T-RTM with higher fibre volume content. As the fibre volume content increases, the more difficult it is to maintain the same resin flow. The use of parts with 3D geometry can also be a challenge for the incorporation of this manufacturing methodologies. An ultrasound vibrational apparatus should also be used to study the effect of higher frequencies on the variability of the parts.

Supplemental Material

Supplemental Material - Enhancing thermoplastic resin transfer moulding using in-situ homogenisation

Supplemental Material for Enhancing thermoplastic resin transfer moulding using in-situ homogenisation by Filipe P Martins, Paulo S Lima, Luís M Silva, Ricardo Torcato and José M Oliveira in Journal of Thermoplastic Composite Materials

Footnotes

Acknowledgements

The authors are grateful for the support of Simoldes Group, namely, IMA - Indústria de Moldes de Azeméis, SA.

Declaration of conflicting interests

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

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was developed within the scope of the project CICECO – Aveiro Institute of Materials, UID/50011/2025 & LA/P/0006/2020 (DOI 10.54499/LA/P/0006/2020), financed by national funds through the FCT/MCTES (PIDDAC). The project was also supported within the scope of TEMA-Center for Mechanical Technology and Automation, by the projects UIDB/00481/2020 and UIDP/00481/2020—Fundação para a Ciência e a Tecnologia, DOI 10.54499/UIDB/00481/2020 and DOI 10.54499/UIDP/00481/2020; and CENTRO-01-0145-FEDER-022083—Centro Portugal Regional Operational Programme (Centro2020), under the PORTUGAL 2020 Partnership Agreement, through the European Regional Development Fund. This work is also supported by CIDMA under the Portuguese Foundation for Science and Technology (FCT, ) Multi-Annual Financing Program for R&D Units, grants UID/4106/2025 and UID/PRR/4106/2025.

ORCID iDs

Filipe P. Martins

Paulo S. Lima

Luís M. Silva

Ricardo Torcato

J. M. Oliveira

Supplemental Material

Supplemental material for this article is available online.

Note

References

Cischino

Di Paolo

Mangino

, et al. An advanced technological lightweighted solution for a body in white. Transp Res Procedia 2016; 14: 1021–1030.

United Nations Environment Programme . Emissions gap report 2023: broken record – temperatures hit new highs, yet world fails to cut emissions (again). Nairobi: United Nations Environment Programme.

Zhang

Chevali

Wang

, et al. Current status of carbon fibre and carbon fibre composites recycling. Compos B Eng 2020; 193: 108053.

Rybicka

Tiwari

Leeke

. Technology readiness level assessment of composites recycling technologies. J Clean Prod 2016; 112: 1001–1012.

Oliveux

Dandy

Leeke

. Current status of recycling of fibre reinforced polymers: review of technologies, reuse and resulting properties. Prog Mater Sci 2015; 72: 61–99.

Bernatas

Dagreou

Despax-Ferreres

, et al. Recycling of fiber reinforced composites with a focus on thermoplastic composites. Clean Eng Technol; 5: 100272.

Ageyeva

Sibikin

Kovács

. A review of thermoplastic resin transfer molding: process modeling and simulation. Polymers 2019; 11: 1555.

van Rijswijk

Bersee

HEN

Beukers

, et al. Optimisation of anionic polyamide-6 for vacuum infusion of thermoplastic composites: influence of polymerisation temperature on matrix properties. Polym Test 2006; 25: 392–404.

Miranda Campos

Bourbigot

Fontaine

, et al. Thermoplastic matrix-based composites produced by resin transfer molding: a review. Polym Compos 2022; 43: 2485–2506.

10.

Herzog

Wendel

Weidler

, et al. Moisture adsorption and desorption behavior of raw materials for the t-rtm process. J Compos Sci 2021; 5: 12.

11.

Wendel

Thoma

Henning

. Influence of water during manufacturing of APA6 in the thermoplastic RTM process. In: 33rd International Conference of the Polymer Processing Society. Cancún, México, December 2017.

12.

Wilhelm

Wendel

Aust

, et al. Compensation of water influence on anionic polymerization of ε-Caprolactam: 1. Chemistry and experiments. J Compos Sci 2020; 4: 7.

13.

van Rijswijk

Bersee

HEN

Jager

, et al. Optimisation of anionic polyamide-6 for vacuum infusion of thermoplastic composites: choice of activator and initiator. Composites Part A 2006; 37: 949–956.

14.

Odrobina

Deák

Székely

, et al. The effect of crystallinity on the toughness of cast polyamide 6 rods with different diameters. Polymers 2020; 12: 293.

15.

Dönmez Çavdar

Yel

Kalaycıoğlu

, et al. Effect of waste melamine impregnated paper on properties of oriented strand board. Mater Des 2013; 51: 751–755.

16.

Joseph

Leramo

, et al. Effect of heat treatment on microstructure and mechanical properties of SAE 1025 steel: analysis by one-way ANOVA. J Mater Environ Sci 2015; 6: 101–106.

17.

Montgomery

. Design and analysis of experiments. 10th ed. Hoboken, NJ: Wiley, 2020.

18.

Andy

Miles

Field

. Discovering statistics using R. Thousand Oaks, CA: Sage Publications, 2012.

19.

Wilcox

. Introduction to robust estimation and hypothesis testing. 3rd ed. Boston: Academic Press, 2012. Epub ahead of print.

20.

Mair

Wilcox

. Robust statistical methods in R using the WRS2 package. Behav Res Methods 2020; 52: 464–488.

21.

Martins

Santos

Torcato

, et al. Reproducibility study of the thermoplastic resin transfer molding process for glass fiber reinforced polyamide 6 composites. Materials 2023; 16: 4652.

22.

Giloni

Seshadri

Simonoff

. Robust analysis of variance: process design and quality improvement. Int J Prod Qual Manag 2005; 1: 306–319.

23.

Shapiro

Wilk

. An analysis of variance test for normality (complete samples). Biometrika 1965; 52: 591–611.

24.

Levene

. Robust tests for equality of variances. In: Olkin

Ghurye

Hoeffding

(eds). Contributions to Probability and Statistics: Essays in Honor of Harold Hotelling. Stanford: Stanford University Press, 1960, pp. 278–292.

25.

Mauchly

. Significance test for sphericity of a normal $n$-Variate distribution. Ann Math Stat 1940; 11: 204–209.

26.

Harwell

Rubinstein

Hayes

, et al. Summarizing monte Carlo results in methodological research: the one- and two-factor fixed effects ANOVA cases. J Educ Stat 1992; 17: 315–339.

27.

Girden

. Anova. Thousand Oaks, CA: Sage Publications, Inc., 1992.

28.

Keselman

Algina

Kowalchuk

. The analysis of repeated measures designs: a review. Br J Math Stat Psychol 2001; 54: 1–20.

29.

Kim

Witherell

, et al. Toward a digital thread and data package for metals-additive manufacturing. Smart Sustain Manuf Syst 2017; 1: 75–99.

30.

Semperger

Suplicz

. The effect of the parameters of T-RTM on the properties of polyamide 6 prepared by in situ polymerization. Materials 2020; 13: 4.

31.

Lee

. The handbook of dielectric analysis and cure monitoring. 2nd ed. Lambient Technologies, 2017.

32.

Callister

. Materials science and engineering - an introduction. 7th ed. Hoboken, NJ: John Wiley & Sons, Inc, 2007.

33.

Zhang

Zeng

. A probabilistic model of the unidirectional tensile strength of fiber‐reinforced polymers for structural design. Adv Civ Eng. 2021. https://onlinelibrary.wiley.com/doi/10.1155/2021/8476784.

34.

Liu

Chen

. A review of void formation and its effects on the mechanical performance of carbon fiber reinforced plastic. Eng Trans 2016; 64: 33–51.

35.

Saenz-Castillo

Martín

Calvo

, et al. Effect of processing parameters and void content on mechanical properties and NDI of thermoplastic composites. Composer Part A Appl Sci Manuf 2019; 121: 308–320.

36.

Mehdikhani

Gorbatikh

Verpoest

, et al. Voids in fiber-reinforced polymer composites: a review on their formation, characteristics, and effects on mechanical performance. J Compos Mater 2019; 53: 1579–1669.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.06 MB