1. Introduction
The landscape of industrial applications has evolved to become more diverse and demanding, presenting challenges for the effectiveness of single-material parts in complex scenarios [
1,
2]. Therefore, a novel type of composite material, termed functionally graded materials (FGMs), which comprise two or more materials, has emerged and is drawing significant academic interest. FGMs are characterized by their material compositions varying spatially through customized gradients, leading to exceptionally satisfactory histocompatibility. This strategic composition mitigates the issue of abrupt interfaces between materials, thus enabling the full exploitation of the inherent potential of each material [
3,
4,
5]. Specifically, ceramic-based FGMs, when integrated with flexibility-oriented additive manufacturing processes, are becoming increasingly crucial in a variety of critical fields, including military [
6,
7,
8], bioengineering [
9,
10,
11,
12], energy [
13,
14], and aerospace [
15,
16,
17,
18].
During the last two decades, the field of additive manufacturing (AM) technology has made significant advancements. Layer-based AM processes for the preparation of FGM parts primarily emphasize flexibility and customization. This approach enables the incorporation of both geometric gradients and varying material compositions. For most ceramic-based FGMs, the raw material exists primarily in a liquid phase. Direct ink writing (DIW), as a distinctive form of Directed Energy Deposition (DED), proves to be a potent method for preparing liquid-phase FGMs [
19]. In addition, DIW demonstrates high applicability across a wide range of liquid-phase materials and offers economic efficiency. Therefore, DIW has been widely adopted in the field of ceramic-based FGM manufacturing [
20,
21,
22].
The workflow of DIW involves raw material preparation, real-time mixing, and extrusion of mixed materials. Among these steps, rapidly achieving uniform mixing of multi-pastes is crucial when producing composite materials with gradients. Ceramic pastes possess highly viscous properties, which makes mixing challenging. The mixing of highly viscous fluids primarily occurs in a laminar flow in the mixing mechanism, which significantly impedes mixing efficiency. Simultaneously, in the context of the continuous extrusion DIW process, the time required for various proportions of multi-materials to achieve homogeneous mixing in the mixing chamber results in a delay in the transition of printed material composites. This directly impacts the printing resolution [
23]. Therefore, it is imperative to optimize the efficiency of the mixing process, minimize the transition delay distance, and enhance print resolution to achieve intricate gradient variations in compact components. It is worth noting that some researchers have successfully prepared ceramic-based FGM parts using static mixers [
24,
25] or dynamic mixers [
26]. Additionally, Computational Fluid Dynamics (CFD) methods have been employed to analyze the fluid dynamics in the mixing chamber. Despite the increased control demands associated with active mixers, they demonstrate superior mixing capabilities in comparison to static mixers. This results in enhanced print resolution and a more streamlined printing system.
The screw mixer represents an active mixing mechanism known for its exceptional back-mixing and transportation capabilities. It has found widespread utilization in multi-material mixing and extrusion processes due to its simple design and low-maintenance advantages [
27,
28]. A typical screw serves to prolong the material’s residence time in the mixing chamber through back-mixing, thereby enhancing the homogeneity of mixed multi-materials. Nevertheless, when preparing FGMs using the DIW process and needing to change material components to improve print resolution while ensuring homogeneous mixing, it becomes crucial to reduce the required mixing time. Therefore, standard screw mixers are unsuitable for processing FGM parts.
Chaotic mixing proves to be the most effective method for enhancing the mixing efficiency of high-viscosity fluids [
29]. The emergence of chaotic mixing effects enhances the radial mixing capacity of the screw while reducing axial mixing along the extrusion direction, facilitating the swift completion of material composite transformation [
30,
31,
32]. Uncomplicated screw mixers excel in axial mixing abilities but are insufficient in generating chaotic mixing effects, which are essential in FGM preparation. In response to this constraint, some researchers have employed more complex differential twin-screw or tri-screw structures to induce chaotic mixing and thereby enhance mixing efficiency [
33,
34,
35,
36,
37,
38,
39]. Nevertheless, the resulting complexity in the mixing mechanism poses challenges in the accurate control of printed material components and leads to increased assembly requirements and an unwieldy printing system [
40]. In an effort to induce chaotic mixing in a simple single-screw structure, Kim and Wiggins et al. developed a novel single-screw mixer with rectangular pins. The existence of chaotic mixing effects induced by this pin-type single screw was confirmed through experiments and numerical analysis [
41,
42,
43,
44]. Nonetheless, the optimal pin profiling remains an area of exploration, and there is untapped potential in both the regular rectangular pins and structural parameters of the single-screw to further enhance the chaotic mixing effect, thus improving print resolution.
The Response Surface Method (RSM) represents a valuable statistical method ideally suited to address complex multivariate issues. In this context, the Central Composite Design (CCD) approach proves highly effective in analyzing complex factor interactions, thereby facilitating the derivation of a comprehensive and efficient response function. For instance, Park utilized CFD in the framework of the RSM method to optimize screw parameters, enhancing both drying efficiency and self-cleaning capabilities [
45]. Likewise, in the domain of food engineering, optimization of relevant screw parameters resulted in improved extrusion efficiency [
46]. Additionally, the CCD method was utilized to construct a significant response function for fundamental screw parameters, with the primary objective of evaluating the uniformity of grass seed mixing [
47]. These instances highlight the remarkable applicability of the RSM method in addressing the optimization challenges presented by multiple factors in the structural parameters of tailored screw mixers.
In response to these challenges, the primary objective of the present work is to design a specialized pin-type single-screw mixer tailored for use in DIW processes. This mixer aims to reduce the transition delay distance in material composites’ transformation and enhance print resolution. Contemporary computer-based optimization design methods offer an alternative to the traditional trial-and-error approach, which heavily relies on designer experience and is time-intensive. Hence, this research employs a combination of data-driven and simulation approaches to develop an efficient pin pattern. Initially, an active mixing chamber’s digital model, based on CFD in ANSYS FLUENT, is created. This model simulates the mixing time of an active online flow of FGMs. The transition time for gradient changes is assessed by monitoring the volume fraction change of the mixed fluid at the outlet of the mixing chamber, thus obtaining the gradient material’s transition regularity at various ratios during the DIW process. Thereafter, the pin’s shape is parametrically defined using a quadratic B-spline, and the optimal pin morphology is determined by integrating RSM with a genetic algorithm (GA), with the aim of achieving the shortest transition time. Additionally, the degree of chaotic mixing in the customized pin-type screw can be assessed in the digital model by tracking particle traces and calculating the Lyapunov exponent [
48]. The final stage of this study involves experimental validation using a self-developed FGM-printing prototype equipped with dual extruders operating at different feed rates. Two pastes are extruded at distinct feed rates into the dynamic mixing chamber, where they are continuously blended by the customized pin-type screw mixer and subsequently delivered to the extrusion needle. This process is supported by a movable platform to create FGM samples. The effectiveness of the optimized pinned mixer is confirmed through digital image processing methods, demonstrating that the tailored pin-type single-screw mixer can achieve shorter transition distances in the printed samples.
2. Materials and Methods
In this section, the modeling of the mixing chamber was detailed using ANSYS FLUENT Vision 2020R2 (fluid simulation software) as the initial step. Thereafter, the volume fraction of the mixed material at the outlet and the uniformity of mixing under a specific feed rate ratio for two input materials were monitored to assess the screw mixing performance. Following this, the response function corresponding to the transition delay time was derived with the RSM in conjunction with the feasible position of the screw pin control point. Finally, to determine the optimal pin pattern, the response function was optimized through a genetic algorithm. The effectiveness of this optimized pattern was then verified by comparing the results with simulation data, with a particular focus on identifying discrepancies.
2.1. Governing Equations
In kinetic analysis, a fluid can be regarded as a continuous medium, and its motion adheres to the principles of conservation of mass, momentum, and energy. Among these principles, the conservation of energy is commonly employed for calculations in systems involving heat exchange flows. For the mixing and extrusion processes conducted at room temperature in this study, the mixing pastes were considered incompressible, and heat transfer in the mixing chamber was considered negligible. In addition, the two pastes utilized in this study are classified as non-Newtonian fluids due to their high viscosity. Thus, it was assumed that heat transfer could be disregarded under isothermal conditions, while the influence of gravity was taken into account. The flow of incompressible multicomponent viscous slurries was described using the simplified Navier–Stokes equations [
49]. Finally, the continuity equation and momentum conservation equation governing fluid motion are depicted in Equations (1) and (2), respectively.
where
represents the fluid density with unit of kg/m
3,
is fluid velocity.
stands for static pressure on the fluid, and
is time.
indicates the stress tensor, and
is the gravity.
2.2. Finite Element Model
As illustrated in
Figure 1a, a dual-extruder system was employed to combine varying feed volume flow rates, using a motor-driven active mixer, to realize the DIW process for printing FGM parts. The primary component in the mixing chamber, the pin-type screw mixer, plays a pivotal role in this process. It possesses specific dimensions, including a length of 60 mm, a minor screw diameter of 6 mm, a pitch of 8 mm, and a flight width of 1 mm. To ensure effective mixing, a shallow screw groove with a depth of 1 mm was incorporated [
50]. The customized pin shape was derived using a quadratic B-spline curve featuring three control points. These control points were defined based on their respective coordinates in the local map shown in
Figure 1a. To prevent the co-linearity of the three control points, the coordinates of the first control point were established in relation to the third control point. The ranges of these three control points are detailed in
Table 1. It is crucial to emphasize that, to guarantee the uniqueness of the quadratic B-spline generated from these three control points, this research utilized an interpolator in the Unigraphics NX 12.0 software, incorporating neither slope nor curvature constraints. Therefore, the sequence of control point insertion ensured the uniqueness of the generated spline curve. The pin height was set at 1 mm, matching the screw flight height. In addition, we evenly distributed thirty pins around the screw to enhance mixing quality [
51].
As illustrated in
Figure 1b, we depicted the numerical simulation model for the mixing component in a mixing chamber equipped with a custom pin-type single screw structure. The screw is configured to rotate clockwise, with a counterclockwise direction of operation. In addition, two parameters are defined in the model to simulate varying material ratios, while a single outlet is designated to emulate the extruded mixing fluid. Details regarding the structural parameters of the finite element model for the mixing chamber can be found in
Table 2.
During the meshing process, the screw mixer structure is treated as a solid medium, and the remaining portion of the mixing chamber structure is designated as a fluid domain. Notably, the flow of the medium is significantly more vigorous in this region compared to other sections of the mixing chamber, owing to the narrow gap between the chosen screw structure and the inner wall. An unstructured tetrahedral mesh tailored to the irregular spatial characteristics is employed for meshing the fluid domain of the mixing chamber to enhance the accuracy of the calculations. Specifically, the boundary layer mesh is optimized, with local densification observed at the inputs, the boundaries of the rotational domain, and the contraction channel. Finally, a mesh size of 0.2 mm is applied to the rotational domain, while the rest of the region employs a mesh size of 0.3 mm. This results in approximately 1.753 million elements and 0.372 million nodes. It should be noted that the number of elements and nodes may slightly vary due to the different pin control points defined.
2.3. Simulation Settings and Paste Properties
To obtain a definitive solution for the flow field, it is necessary to establish the initial boundary conditions in the simulation domain. When preparing FGM parts with extrusion technology, a pressure-based solver was chosen to analyze the low-velocity, incompressible flow field. The boundary conditions were set separately for the velocity inlet and pressure outlet. The combined feed rate for the two inputs was held constant at 0.5 mm/s. Therefore, the total volumetric flow rate could be determined by calculating the cross-sectional area of the inlet, resulting in a value of 2.4 mm
3/s. To simulate the rotating flow field in the mixing zone of the pin-type screw, the transient simulation employed the Rotating Reference Frame (RRF) method. This involved designating Fluid-1 and Fluid-2 (as depicted in
Figure 1b) as static and dynamically rotational domains, respectively, with both domains being connected through an interface. In addition, the screw wall was assigned a no-slip boundary condition and was configured as a moving wall to simulate the adjustable operating speed of the screw, ranging from 10 to 50 rpm.
In this research, we employed two readily available types of calcium carbonate-based toothpaste, designated as Material A (white color) and Material B (green color), with similar non-Newtonian rheological properties to simulate the mixing extrusion process. The densities of Materials A and B were determined to be 1120 kg/m
3 and 1285 kg/m
3, respectively, using the specific gravity method. Both pastes exhibit shear-thinning rheological characteristics. Therefore, the study employed the power-law function, as shown in Equation (3), which is a commonly utilized tool in non-Newtonian fluid modeling, to describe their rheological properties. The rheological parameters for both pastes were assessed using a rotational rheometer (MCR 302, Anton Paar, Graz, Austria) at room temperature. The test results are depicted in
Figure 2, and the corresponding rheological parameters were derived through the power-law function, as presented in
Table 3. Finally, in the confines of a narrow mixing channel, in conjunction with the aforementioned pin-type screw speed, the viscous behavior of the mixing paste was characterized as laminar flow.
where is the shear stress with unit of
,
represents the flow index with a unit of
, and
symbolizes the dimensionless flow behavior index, which is less than 1 for shear thinning paste.
is the shear rate with a unit of s
−1.
2.4. Response Surface Method and Optimization
We employed a second-order RSM design with rotational center composites to achieve significant and highly fitting and accurate statistical results with a limited number of experiments. This design aimed to establish the response function of six coordination variables concerning transition delay time. Thereafter, global optimization was conducted using a genetic algorithm in conjunction with a robust predictive model to identify the optimal pinning pattern. The optimization process is illustrated in
Figure 3. Considering the feasible range of the three control points on the surface of the screw minor diameter, their variable boundaries were determined based on star points. Therefore, following the methodology of the rotatable CCD, a six-factor half-quantity experiment was conducted to calculate the normalized distance
from the design centroid to the boundary points of the design space, as described in Equation (4).
Table 4 provides information on the range of coordination variable levels for the three control points.
where
is the number of factors.
The response variable for the design experiment was set as the delay time corresponding to scenarios with the most challenging material ratio changes, with the aim of investigating the beneficial impact of varying pin patterns on reducing transition delay time [
24]. This scenario represents the highest print resolution achievable. Initially, the material ratio in the mixer chamber was defined as 9:1 for material A to material B, with input feed ratios set at 1:9 while maintaining a total feed rate of 0.5 mm/s. The simulation process maintained a constant screw speed of 25 rpm, consistent with the subsequent experimental phase. The volume fraction of material A at the outlet was continuously monitored and recorded. The tolerance for the component transformation spanned ±5%. In other words, the time interval corresponding to a change in the material A component from 0.86 to 0.14 represented the transition delay time.
Figure 4 presents a simulation result for the Nr.35 experimental design condition, illustrating the transition delay time for material mixing transformations and the evolving mixing conditions in the chamber at different time points. In addition, the software Design-Expert 13.0 generated a total of 52 conditional values for the six-factor half-quantity design points, including eight center points. These values were established based on the aforementioned simulation setup to obtain the response values, specifically the transition delay time. The design experiment points and response results are detailed in
Table 5.
The significant modified quadratic response regression model was derived through polynomial regression analysis of the experimental data provided in
Table 5, yielding the response function for transition delay time as shown in Equation (5).
where
is transition delay time with a unit of
. The
and
correspond to the coordinate values, respectively, and the foot index represents the serial numbers of the control points.
As seen in Equation (5), it involves all of the first terms of the six factors, along with several interaction and squared terms. Finally, the resulting RSM model fit effect had a coefficient of determination
of 93.35%, an adjusted coefficient of determination
of 91.52%, and a predicted determination coefficient
of 87.5%. This indicates that the model exhibits decent fitting results, and its predictive capability is considered satisfactory.
Figure 5 and
Figure 6 depict the response surfaces. The remaining four factors were set to intermediate values to generate the response surfaces. In
Figure 5a, the x- and y-values represent relative coordinates concerning the third control point. It is evident that changes in the y-value have a more pronounced impact on the response results, with the middle y-values corresponding to a shorter transition delay time.
Figure 5b illustrates the response surface for the x- and y-values of the second control point in relation to the transition delay time. Increasing both the x- and y-values results in a lower response value.
Figure 5c displays the influence of the x- and y-axis coordinate values of the third control point on the transition delay time. As the x-value increases and the y-value decreases, the response value decreases accordingly. Additionally,
Figure 6 represents the effect of changing the two relative coordinates of the first control point with respect to the third point on the transition delay time. It is evident that changes in the y-value exert a more significant impact on the response value compared to changes in the x-value. The minimum response value occurs when the y-value of the third point is at its minimum, and the y-value at the first point is at its middle value. These response surfaces provide valuable insights into the influence of factor variations on response values. Therefore, it is necessary to employ a global optimization approach for the obtained response function to attain specific optimization results.
A genetic algorithm (GA) is a powerful heuristic algorithm used for global search, especially in tackling multivariate and nonlinear optimization problems. A GA emulates the concept of population evolution, employing continuous iterative processes such as selection, crossover, and mutation to gradually transform individuals until the global optimal solution of the problem is achieved.
In this section, Equation (5) is utilized as the response function and optimized using a GA to determine the coordinates of the three control points, aiming to minimize transition delay time. Firstly, the six factors are encoded in binary with a length of 20 bits, establishing the connection between coordinate parameters and the chromosome bit strings’ structure in the genetic algorithm. The GA continuously selects and retains individuals with high fitness values for evolutionary processes until convergence is reached. The fitness value is defined based on Equation (5)’s minimum value. We employed a roulette wheel approach to prevent getting trapped in local optimal solutions during the global search. This approach takes into account that individuals with higher fitness values have a greater probability of being selected for further evolution. Additionally, the population size is set to 30, with a maximum of 200 generations. The crossover rate is set to 0.7, and the mutation operator rate is 0.007, ensuring a robust search range and improved convergence speed. The GA execution process is depicted in
Figure 7. The results indicate a minimum transition delay time of 117.69 s. The coordinates of the three control points corresponding to this solution are (0.5, 0.56), (2.5, 4.4), and (2.5, 3), while the single pin pattern area is 4.42 mm
2.
Figure 8 displays the optimized pin type screw mixer. Following the optimization results, a simulation of the optimized pin type screw mixer is conducted using the previously defined setup, resulting in a transition delay time of 113 s, which exhibits an acceptable level of error. A detailed numerical verification is presented in the subsequent section.
4. Conclusions
The pin-type active screw mixer, developed utilizing RSM in conjunction with a global optimization approach, has proven its efficacy in substantially reducing the distance required for the transition of material components. When compared to two other similar screw types, the optimized pin-type screw mixer reduced the transition distance by approximately 28.5% and 33.1%, respectively, in scenarios where the ratio of material component changes ranged from 9:1 to 1:9. In addition, CFD models with a high degree of accuracy have been generated in this study. These models are unquestionably capable of efficiently and economically characterizing the transition delay time for material component changes and the uniformity of mixing in extruded materials. In conjunction with the simulation results, the response function of the three control points to the transition delay time was eventually determined through the half-composite center design. In the simulation model, the particle tracking method was employed to compute the Lyapunov exponent, which serves to assess the degree of chaotic mixing in the chamber. Additionally, this method evaluates the radial dispersive mixing capability of the mixer through the Poincaré map. Numerical validation results indicate that the Lyapunov exponent of the optimized pin-type screw mixer is 4.3 times higher than that of the mixer lacking pins and 3.5 times higher than that of the cylindrical pinned mixer. The Poincaré map further confirms that the optimized pin-type screw offers superior radial dispersion mixing capabilities. These findings provide evidence that the optimized pin-type screw mixer enhances the chaotic mixing effect by increasing radial mixing while decreasing axial mixing. Through experimental research, the double-extruder printing system, combined with the optimized screw mixer, has successfully prepared FGM parts with smooth variations. Moreover, the gradient change process can be effectively characterized through digital image processing.
In this work, the researchers extensively studied a singular instance of transition delay distance associated with a change in material components. In addition, the CFD model can be employed to explore additional transition delay distances arising from various changes in material components. To streamline the experimental data, this study utilized a quadratic B-spline to characterize the pin’s geometry. Subsequent research may involve the utilization of higher-degree B-splines to extend the scope of design possibilities. It is worth noting that nearly all paste-like materials are amenable to the DIW process. This presents exciting opportunities for further exploration into the regulation of printing parameters for pastes exhibiting diverse rheological properties, such as ceramics, polymers, and other non-Newtonian fluids. This exploration holds the potential to unlock a wider array of performance of FGM parts. Regarding the study of print path strategies, future research directions may consist of the consideration of the spatial deployment of the transition region. The objective would be to achieve a transition while minimizing any negative effects on material properties.