This study investigates the enhancement of mechanical properties of metal/polymer composites produced through fused deposition modeling and the prediction of the ultimate tensile strength (UTS) by machine learning using a Classification and Regression Tree (CART). The composites, comprising 80% acrylonitrile butadiene styrene matrix and 10% each of aluminum (Al) and copper (Cu) fillers, were subjected to a comprehensive exploration of printing parameters, including printing temperature, infill pattern, and infill density using the Taguchi method. The CART unveiled a hierarchical tree structure with four terminal nodes, each representing distinct subgroups of materials characterized by similar UTS properties. The predictors’ importance was assessed, highlighting their role in determining material strength. The model exhibited a high predictive power with an R-squared value of 0.9154 on the training data and 0.8922 on the test data, demonstrating its efficacy in capturing variability. The optimal combination of parameters for maximizing UTS was a zigzag infill pattern, a printing temperature of 245 °C, and an infill density of 10%, which is associated with the highest UTS of 680 N. The model’s reliability was confirmed through a paired t-test and test and confidence interval for two variances, revealing no significant difference between the observed and predicted UTS values. This research contributes to advancing additive manufacturing processes by leveraging CART analysis to optimize printing parameters and predict material strength. The identified optimal conditions and subgroup characteristics pave the way for developing robust and predictable metal/polymer composites, offering valuable insights for material design in the era of advanced manufacturing technologies.

The world is progressively adopting contemporary procedures following the Fourth Industrial Revolution. The latest manufacturing methods in the modern world are additive manufacturing (3D printing) and rapid prototyping. The application of additive manufacturing has been growing steadily in recent decades.1 Industries employ it, and researchers are doing countless experiments in this area to improve it and continually improve it. Additive manufacturing involves printing 3D items. It is similar to translating a designer’s creativity and computer-generated realism into actual objects. It provides a range of printing materials, including those made of plastic, ceramics, metals, and even concrete.2 The manufacturing process begins with a computer-aided design (CAD) model. The model is then transformed into the STL format and loaded with G-codes for printing on a 3D printer. The required parameters are then generated using a 3D printer and layer-by-layer printing.3 This printing offers superb accuracy while managing material waste following customary manufacturing procedures.4 Fused deposition modeling (FDM) is the simplest of all the AM processes. FDM provides several benefits, including material waste recovery, flexible printing, and affordable labor. This is explained by using the extruded thermoplastic filament, which combines and solidifies layer by layer to create 3D structures. Because it prints with a filament-like polymer component, this technology is often called the extrusion-based manufacturing process. Thermoplastic polymers are growing because of their multiple benefits, including toughness, corrosion resistance, flexibility, and light weight, which are necessary for numerous technical applications in today’s industries.5 The characteristics of polymer materials are enhanced for additive manufacturing uses by different fillers. Then, polymer-based composites are created; when metal fillers are added to the polymer as reinforcement, the composites are known as metal–plastic composites (MPCs). Numerous sectors, including those in the transportation, aviation, electrical, athletics, and medicine, need these metal–plastic composites significantly. Numerous researchers studying this area are looking at how the addition of reinforcement to polymers and plastics affects their physical, electrical, mechanical, and thermal characteristics.6 

Additive manufacturing has transformed several areas, including biomedical and engineering. In biomedical applications, 3D printing has unique possibilities for creating patient-specific implants, prosthetics, and tissue scaffolds.7 The technique offers exact personalization, resulting in improved patient outcomes and fewer surgical complications. Furthermore, 3D printing is essential for quick prototyping of medical devices and equipment, allowing for greater innovation and faster product development.8 3D printing has changed traditional engineering manufacturing techniques, allowing complicated designs, lightweight constructions, and on-demand production.9 Industries such as aerospace, automotive, and consumer products benefit from 3D printing’s ability to produce sophisticated designs with great precision and material economy. The technology allows for quick iteration and optimization of designs, lowering time-to-market and total production costs. 3D printing has emerged as a disruptive force in biomedical engineering, providing unparalleled flexibility, customization, and efficiency in product development and manufacture. Its continuous developments are expected to promote more innovation and address issues across various applications, cementing its position as a cornerstone of contemporary industrial and healthcare practices.

Due to their characteristics and increasing demand in car manufacturing, architecture, electronics, and defense industries, metal-reinforced polymers have recently gained popularity. As a result, researchers are continually working in this domain to improve the properties of polymers while also addressing the limits of FDM. However, in this case, we are talking about the thermoplastic composites with metal reinforcement used in FDM. Researchers are studying the mechanical, physical, morphological, electrical, thermal, and tribological properties of metal/polymer composites for various purposes.10 For this reason, metal-reinforced polymers are lighter and have better characteristics. In terms of managing solid waste, it is also a key factor. Since plastics and metal powders from industrial waste are recyclable, employing them in worthwhile projects that will help with trash management is preferable.11 

Due to its appropriate fluidity, advantageous solidity, and toughness, Acrylonitrile Butadiene Styrene (ABS) is a commonly used thermoplastic polymer in FDM. Compared to components manufactured via molding, FDM items made of ABS have lower malleability, sturdiness, and fracture toughness.

Many researchers incorporated Cu (copper) and Fe (iron) particles to increase its tensile characteristics. Nikzad et al.12 studied the thermal and mechanical characteristics of copper-reinforced ABS and iron-reinforced ABS composites, finding that the heat transfer of the composite improved with the addition of filler percentage. There is an enhancement in the electromechanical characteristics of the composites. Castles et al.13 improved the dielectric permeability of the ABS polymer by adding BaTiO3 particles. At 70 wt. % of filler, the relative permeability of the ABS composite increased by 240% compared to the ABS polymer. Introducing reinforcement particles in polymeric materials has improved the structural and compositional characteristics while addressing the difficulties encountered throughout this printing process. Because of the rapid thermal expansion of polymers, one of the most pressing problems is free gripping between both layers while printing. Its expansion of metallic particles reduces the overall thermal effects of the composite and helps the layers stay together. The combination of Cu (copper) and Fe (iron) particles in ABS significantly reduces the thermal expansion of a three-dimensional composite. Singh et al.14 investigated the melt flow rate and tensile, thermal, and electrical characteristics of the copper-reinforced ABS composite, which shows the highest electrical and thermal conductivity at 10 wt. % of copper particles. Improvements in MFI and tensile strength with small particles of copper were achieved. Hamzah et al.15 studied ABS and copper ferrite composites and examined their thermal, mechanical, and electrical characteristics. With varying wt. % of the filler, the greatest tensile improvement is observed with the addition of 14 wt. % of the filler into the composite (17.57 MPa), which is 135% higher compared to the pure ABS, followed by 15.94 MPa with 11 wt. % addition and 15.76 MPa with 8 wt. % addition, which are 113% and 110% higher compared to ABS, respectively. It also improved the thermal and electrical properties. Torrado et al.16 performed fracture surface analysis and investigated the tensile strength of the ABS/TiO2 composite with 5 wt. % of filler TiO2. They achieved a tensile strength of 32.2 MPa. The surface finish of the composite pieces is smoother than the ABS polymer. Fracture strain is also lowered by 29%. The thermal and mechanical characteristics of Fe/ABS and Cu/ABS composites were examined, and it was revealed that with the addition of Cu and Fe particles, the tensile strength of the Cu/ABS and Fe/ABS composites was reduced to 26.5 and 36.2 MPa, respectively, when compared to the tensile strength of ABS plastic (45.7 MPa). With the addition of Cu (copper) particles to the ABS plastic, the heat transfer of the composite rose to 0.912 W/mK from 0.646 W/mK.17 Khatri et al.18 investigated how the inclusion of SS (stainless steel) particles changed the magnetic characteristics of the SS/ABS composite and how the tensile strength and modulus of elasticity decreased. Sa’ude et al.19 studied the MFI for the Cu/ABS composites and found that it rose with the addition of filler, reaching a value of 42.71 g/10 min. The BN/ABS composite’s thermal conductivity increased by 0.93 W/m K. The thermal conductivity is five times that of pure ABS. This also demonstrates the composite’s declining elastic modulus and toughness properties.20 Çantı et al.21 determined the increase in the strain of Al/ABS and ZrB2/ABS composites. The aggregate and individual scores were 108% and 82.5%, respectively. A complete literature review of metal-reinforced ABS composites is shown in Table I. Table I presents vital infromation on metal-reinforced ABS composites.

TABLE I.

An overview of relevant literature on metal-reinforced ABS composites.

CompositesMatrixReinforcementConclusionRef.
ABS/Cu ABS (99%–90%) Cu (1%–10%) Improvement in MFR (melt flow rate) and tensile strength with small copper particles. A copper density of 10% shows the highest electrical and thermal conductivity. 14  
Iron/ABS Copper/ABS ABS (95%, 90%, 80%, 70%, and 60%) Fe/Cu (5%, 10%, 20%, 30%, and 40%) Increase in heat transfer as the filler proportion increases. Cu/ABS and Fe/ABS attain their storage modulus at room temperature. At room temperature, 3.5–4 and 2.5–3 GPa are reached. Adding filler particles improved the thermal, electrical, and mechanical properties. 12  
ABS/BaTiO3 ABS (90%–30%) BaTiO3 (10%–70%) Improved dielectric permeability. 13  
ABS/CuFe2O4 ABS (rest) CuFe2O4 (8%, 11%, and 14%) An addition of 14% of filler into the ABS composite provided the highest tensile strength (17.57 MPa), which is 135% of the pure ABS. Good adhesion between the filler and ABS is observed. Increased thermal, electrical, and hardness properties. 15  
ABS/TiO2 ABS 95% TiO2 5% The UTS was 13.2% higher than that of regular ABS. It has lower roughness compared to normal ABS. Fracture strain was reduced by 29%. 16  
ABS/iron ABS/copper ABS (90%, 70%, and 60%) Fe/Cu (10%, 30%, and 40%) Decrease in the tensile strength with the increase in infill density. The tensile stress and strain decreased by increasing the metal particles’ percentage. The thermal conductivity of the ABS/Cu composites increases and the thermal coefficient of expansion decreases with the addition of copper particles. 17  
ABS/stainless steel ABS 60% Stainless steel 40% Compared to pure polymers, the structural performance falls as the number of filler particles increases while the functional magnetic reaction increases and reduces Young’s modulus. 18  
ABS/Cu ABS (10%–45%) Cu (45%–82%) Mechanical properties are affected by the increase in the amount of copper particles. Flexural strength and MFI rate increase with copper particle addition. 19  
ABS/BN ABS (95%, 85%, 75%, and 65%) Boron nitride (BN) (5%, 15%, 25%, and 35%) The thermal conductivity of the composite increased by five times that of the original ABS, which is 0.93 W/m-K. Flexural strength and impact toughness decrease with the increase in filler percentage. 20  
ABS/Al ABS/ZrB2 ABS (rest) Al (1.5%) ZrB2 (1.5%) With the addition of the filler, stress and strain increased to 82.5%. 21  
ABS/SnBi ABS (95% and 90%) Tin–bismuth (SnBi) (5% and 10%) A better dispersion of SnBi particles within the ABS was achieved. A better tensile strength compared to the ABS. 22  
ABS/420 SS ABS (rest) 420 SS (10%, 15%, 20, and 23%) 15% of infill preserved mechanical properties. Ultimate tensile strength reduced at 23% of infill. 23  
ABS/Al ABS (rest) Al (4.8%) Suitable for FDM manufacturing. Surface roughening of the 3D printed pieces is achieved by dissolving Al particles in caustic soda. 24  
CompositesMatrixReinforcementConclusionRef.
ABS/Cu ABS (99%–90%) Cu (1%–10%) Improvement in MFR (melt flow rate) and tensile strength with small copper particles. A copper density of 10% shows the highest electrical and thermal conductivity. 14  
Iron/ABS Copper/ABS ABS (95%, 90%, 80%, 70%, and 60%) Fe/Cu (5%, 10%, 20%, 30%, and 40%) Increase in heat transfer as the filler proportion increases. Cu/ABS and Fe/ABS attain their storage modulus at room temperature. At room temperature, 3.5–4 and 2.5–3 GPa are reached. Adding filler particles improved the thermal, electrical, and mechanical properties. 12  
ABS/BaTiO3 ABS (90%–30%) BaTiO3 (10%–70%) Improved dielectric permeability. 13  
ABS/CuFe2O4 ABS (rest) CuFe2O4 (8%, 11%, and 14%) An addition of 14% of filler into the ABS composite provided the highest tensile strength (17.57 MPa), which is 135% of the pure ABS. Good adhesion between the filler and ABS is observed. Increased thermal, electrical, and hardness properties. 15  
ABS/TiO2 ABS 95% TiO2 5% The UTS was 13.2% higher than that of regular ABS. It has lower roughness compared to normal ABS. Fracture strain was reduced by 29%. 16  
ABS/iron ABS/copper ABS (90%, 70%, and 60%) Fe/Cu (10%, 30%, and 40%) Decrease in the tensile strength with the increase in infill density. The tensile stress and strain decreased by increasing the metal particles’ percentage. The thermal conductivity of the ABS/Cu composites increases and the thermal coefficient of expansion decreases with the addition of copper particles. 17  
ABS/stainless steel ABS 60% Stainless steel 40% Compared to pure polymers, the structural performance falls as the number of filler particles increases while the functional magnetic reaction increases and reduces Young’s modulus. 18  
ABS/Cu ABS (10%–45%) Cu (45%–82%) Mechanical properties are affected by the increase in the amount of copper particles. Flexural strength and MFI rate increase with copper particle addition. 19  
ABS/BN ABS (95%, 85%, 75%, and 65%) Boron nitride (BN) (5%, 15%, 25%, and 35%) The thermal conductivity of the composite increased by five times that of the original ABS, which is 0.93 W/m-K. Flexural strength and impact toughness decrease with the increase in filler percentage. 20  
ABS/Al ABS/ZrB2 ABS (rest) Al (1.5%) ZrB2 (1.5%) With the addition of the filler, stress and strain increased to 82.5%. 21  
ABS/SnBi ABS (95% and 90%) Tin–bismuth (SnBi) (5% and 10%) A better dispersion of SnBi particles within the ABS was achieved. A better tensile strength compared to the ABS. 22  
ABS/420 SS ABS (rest) 420 SS (10%, 15%, 20, and 23%) 15% of infill preserved mechanical properties. Ultimate tensile strength reduced at 23% of infill. 23  
ABS/Al ABS (rest) Al (4.8%) Suitable for FDM manufacturing. Surface roughening of the 3D printed pieces is achieved by dissolving Al particles in caustic soda. 24  

Achieving consistent and optimal mechanical characteristics is a key problem when producing metal/polymer composites using FDM. Uncertainty in the final material strength is introduced by the intrinsic complexity of additive manufacturing processes and the complex interactions among different printing factors. The combined effect of these parameters, including printing temperature, infill pattern, and infill density, on the composite’s UTS is still poorly understood. This information gap results from the inability to systematically manage and improve the mechanical performance of metal/polymer composites. Therefore, exploring, evaluating, and simulating the interdependency of various printing factors is imperative to offer practical guidance for attaining a higher UTS in metal/polymer composites produced by FDM. Manufacturers face the dilemma of selecting optimal printing conditions to achieve the desired mechanical characteristics. The absence of clear guidelines or predictive models hampers the ability to produce metal/polymer composites with an enhanced UTS consistently.

Machine Learning (ML) has emerged as a powerful tool in materials science, revolutionizing how materials are designed, characterized, and optimized.25 Various ML applications within materials science focus on its relevance to the investigation of dual metal-reinforced polymer composites for 3D printing. Predicting material characteristics is one of the main uses of ML in materials science. Complex interactions between the composition of the material, the processing parameters, and the final qualities can be analyzed using ML algorithms.26 Machine learning models may be developed to predict critical attributes, including tensile strength, modulus, and thermal conductivity depending on the composition and processing parameters in the context of dual metal-reinforced polymer composites. ML speeds up the materials unearthing process by making it possible to identify novel materials with desirable features. Large datasets of material attributes may be analyzed by algorithms, which aids researchers in finding patterns, correlations, and possibly novel combinations. This skill is handy for creating cutting-edge materials for 3D printing applications, as customized qualities are frequently needed. ML algorithms may enhance manufacturing procedures by examining big datasets to determine ideal settings. ML may help determine the optimal printing temperature, layer thickness, and metal reinforcement distribution for 3D printing polymer composites, improving the manufacturing process’s overall efficiency and quality. In industrial processes, quality control is achieved through ML algorithms. They can find flaws, anomalies, or changes in the printed composites by analyzing real-time sensor data and imagery. By doing this, the creation of materials with uniform qualities is ensured.

Within the particular context of this study, UTS prediction is modeled using ML, especially the CART technique. The CART method can forecast the UTS of the composite materials and offer insights into the interaction between various factors by training the model on experimental data relating to the melt flow index and tensile characteristics of dual metal reinforced polymer composites. Combining ML with materials science presents a comprehensive strategy for comprehending, creating, and refining materials. ML is a useful predictive modeling and analysis technique for studying dual metal-reinforced polymer composites for 3D printing. This helps develop the fields of materials engineering and additive manufacturing technologies. Consequently, this research addresses the nuanced relationships between printing parameters and UTS, leveraging CART’s sophisticated analytical ML tool. By doing so, the study aims to unravel the intricacies of the additive manufacturing process, enabling the development of precise guidelines for parameter selection to attain a higher UTS in metal/polymer composites. This research seeks to bridge the existing gap in knowledge by formulating a systematic and data-driven approach to optimize FDM parameters for enhanced mechanical performance.

A literature review documented investigations into various thermoplastic polymers reinforced with bronze, copper, BNT, multi-walled carbon nanotubes, iron powder, and other metallic powders. However, the outcomes of this research in thermoplastic composite matrices with double metal reinforcement remain primarily unknown. The research aims to create a dual metal reinforced-based polymer composite filament wire for FDM. Following that, a tensile sample is created by utilizing the printing conditions. To get the best results for UTS, the tensile characteristics of the sample are next examined using a UTM and the Taguchi L9 orthogonal array approach. The objectives of this study are as follows:

  • To analyze the impact of key printing parameters, printing temperature, infill pattern, and infill density on the UTS of metal/polymer composites.

  • To determine the flow rate of polymer composites, which is essential for their applicability in FDM, and to determine the melt flow index (MFI).

  • To develop a predictive model using the ML-CART approach to optimize printing parameters for maximizing UTS.

  • To identify the optimal combination of printing parameters that yields the highest UTS in metal/polymer composites.

Furthermore, this research is significant for advancing the understanding of additive manufacturing processes, explicitly optimizing the metal/polymer composite strength. The study’s outcomes will contribute to developing guidelines for precise parameter selection, fostering the production of robust and predictable materials through FDM.

The ABS pellets were employed as the investigation’s matrix (Goyal Poly Products, Chandigarh, India). Pallav Chemicals in Mumbai and Chandigarh University supplied reinforcing aluminum (Al) metal powder and 400 mesh copper (Cu) powder, which are illustrated in Fig. 1.

FIG. 1.

(a) ABS pellets. (b) Al powder. (c) Cu powder.

FIG. 1.

(a) ABS pellets. (b) Al powder. (c) Cu powder.

Close modal

The melt flow rate (MFR) and melt volume rate (MVR) of melted polymers are measured by a melt flow test. The MFI was established to measure the flow properties of polymer composites, which is essential for their applicability in FDM. Running the weight of the polymer through melt flow testing apparatus for 10 min at a constant temperature is required to measure the MFI. ASTM D1238 and ISO 1133 provide specifics on the defined process for this evaluation.24 

A PACORR Melt Indexer Machine (MIM) was used for this investigation; refer to Fig. 2(a). The dual metal reinforced composite with a composition of 10% aluminum and 10% copper powder as reinforcement into the matrix ABS (80%) is first weighed using a Wensar weighing scale. Amla oil was used as a binder to appropriately blend the ABS pellets, aluminum powder, and copper powder into the composite. A temperature of 230 °C and 3.8 kg of weight were used in this experiment to calculate MFR for the ABS polymer using the rules of the ISO and ASTM.27 A die is first introduced into the MIM and heated to the requisite temperature of 230 °C. Once the machine has reached the required temperature, the barrel and die nozzle are cleaned by inserting an ABS pellet into the machine and extracting it with a piston and plunger. A 2-min timer is set after the machine has been cleaned. Next, the composite is placed within the barrel, and with the aid of weight provided to the piston, the molten polymer is forced out via the die for a total of five 2-min cycles, as shown in Fig. 2. To calculate the MFR,28 the following equation is used:
MFR=Wt×600sec,
(1)
where w = the extruded material’s weight and t = the extruded time in seconds. The unit of MFR is grams/10 min.
FIG. 2.

(a) Wensar weighing machine. (b) MIM (melt indexer machine).

FIG. 2.

(a) Wensar weighing machine. (b) MIM (melt indexer machine).

Close modal

During the current experimental research, the MIM [Model: PCMFIT1; see Fig. 2(b)] used to measure the MFR of polymer composite was offered by Chandigarh University, Mohali, Punjab, with a heating range of up to 400 °C, a resolution of 0.1 °C, a microprocessor-based digital timer, and an accuracy of ±0.1% FSD.

After producing the filament using the MIM, we ran all of the filaments through a Felfil Plastic Shredder to convert them into little pieces. By repeating this process four to five times, we created fine, minute fragments of the composite as shown in Fig. 3(a). Figure 3(b) displays a single screw extruder machine. A Felfil filament extruder with the following specifications was utilized in this experiment to fabricate the filament for FDM of a double metal reinforced polymer composite and was provided by Chandigarh University, Mohali, Punjab:

  • The highest temperature range is 250 °C.

  • It is compatible with materials such as PLA, ABS, HIPS, PETG, PA (6, 12), PMMA, HDPE, LDPE, TPU, TPE, and PVA.

  • The rpm range is 0–9.

  • It has a bronze nozzle extruder with interchangeable nozzles ranging from 1.75 to 2.85 mm.

  • With a diameter sensor precision of 10 μm, the filament’s diameter ranges from 0.5 to 3 mm.

FIG. 3.

(a) (A) Extruding filament from MIM, (B) extruded filament inserted into the shredder, (C) shredded into fine pieces, and (D) fine shredded polymer composite fragments. (b) Single screw filament extruder machine. (c) (Al–Cu–ABS) composite filament wire.

FIG. 3.

(a) (A) Extruding filament from MIM, (B) extruded filament inserted into the shredder, (C) shredded into fine pieces, and (D) fine shredded polymer composite fragments. (b) Single screw filament extruder machine. (c) (Al–Cu–ABS) composite filament wire.

Close modal

In order to get the barrel temperature between 180 and 200 °C, the extruder machine must first be heated for 10–20 min. Then, a particular amount of the (Al–Cu–ABS) composite mixture pellet was loaded into the hopper, and the material flowed from top to bottom using gravity. The extruder screw will compress the (Al–Cu–ABS) composite pellet material as it travels from the feeding region to the center, front, and die areas. For proper mixing of fresh PMC materials, the material was burnt inside the barrel screw.

The barrel temperature was raised gradually from low to high to avoid the melt from adhering to the screw. Four temperature zones and the extruder screw speed will be manually altered to keep the feedstock filament diameter. The feedstock filament’s target diameter was around 1.7–1.75 mm, while the die’s diameter was 2.85 mm. To manage the diameter of the filament during manufacture, the screw speed was gradually reduced. Without a waterbed cooling system, the feedstock filament will flow at room temperature. The FDM machine will employ this (Al–Cu-ABS) composite filament wire in layer manufacture. Figure 3(c) depicts the (Al–Cu–ABS) composite filament produced using a single screw extruder machine.

After producing the (Al–Cu–ABS) composite filament wire, we used a Creality Ender-3 3D printer to generate the tensile sample for the tensile test. Utilizing the CAD software, the dog bone structure for the tensile sample was first created. We then converted it to the STL file format and sent that file to the 3D printer. The sample was printed layer by layer using the printing settings, as shown in Fig. 4(a).

FIG. 4.

(a) (A) Preparing sample in the CAD software, (B) converting sample into the STL format, (C) applying printing parameters and slicing, (D) using the filament wire in printer, (E) printing the sample layer-by-layer, and (F) sample after print. (b) Specimen drawing and dimensions. (c) (A) Line, (B) triangular, and (C) zigzag.

FIG. 4.

(a) (A) Preparing sample in the CAD software, (B) converting sample into the STL format, (C) applying printing parameters and slicing, (D) using the filament wire in printer, (E) printing the sample layer-by-layer, and (F) sample after print. (b) Specimen drawing and dimensions. (c) (A) Line, (B) triangular, and (C) zigzag.

Close modal

These nine samples were created with a 3D printer, each under a distinct printing condition for tensile testing. Three factors were chosen to be changeable, and the others were fixed using the Taguchi method (L9 orthogonal array) using the Minitab application. The three parameters, printing temperature, infill density, and infill pattern, were continuously modified into three distinct levels using the Minitab application and a Taguchi L9 orthogonal array design, and the table was constructed with various combinations. Table II displays all three parameters utilized at various levels to generate nine specimens using the Taguchi technique (L9 orthogonal array), as indicated in Table III.

TABLE II.

Printing parameters used for sample preparation.

Levels
S. No.Printing parameters123
Infill pattern Line Triangular Zigzag 
Printing temperature (°C) 240 245 250 
Infill density (%) 10 15 20 
Levels
S. No.Printing parameters123
Infill pattern Line Triangular Zigzag 
Printing temperature (°C) 240 245 250 
Infill density (%) 10 15 20 
TABLE III.

L9 orthogonal array using Minitab.

S. No.Infill patternPrinting temperature (°C)Infill density (%)
Line (1) 240 10 
Line (1) 245 15 
Line (1) 250 20 
Triangular (2) 240 15 
Triangular (2) 245 20 
Triangular (2) 250 10 
Zigzag (3) 240 20 
Zigzag (3) 245 10 
Zigzag (3) 250 15 
S. No.Infill patternPrinting temperature (°C)Infill density (%)
Line (1) 240 10 
Line (1) 245 15 
Line (1) 250 20 
Triangular (2) 240 15 
Triangular (2) 245 20 
Triangular (2) 250 10 
Zigzag (3) 240 20 
Zigzag (3) 245 10 
Zigzag (3) 250 15 

All samples are made in compliance with ASTM D638-14 Standard Test Method for Tensile Properties of Plastics. Figures 4(b) and 4(c) depict the specimen’s drawing and dimensions and the nine samples produced using Minitab and the Taguchi approach (L9 orthogonal array), respectively, as indicated in Table III.

Several criteria were considered for determining the Taguchi method’s printing temperature, infill density, and pattern levels. The study attempted to address various realistic situations prevalent in additive manufacturing. It included varying the printing temperature to see how it affected material bonding and strength, adjusting the infill density to see how it affected the internal structure, and experimenting with different infill patterns to analyze material distribution and mechanical characteristics.29 The study used the Taguchi technique to efficiently explore a large parameter space with few experiments since the L9 orthogonal array enabled us to evaluate nine combinations while retaining statistical robustness. It allowed us to capture the interactions between printing settings and their effects on the UTS of metal/polymer composites.

Furthermore, picking three factors with three levels each was strategic since it allowed for a more balanced evaluation of the primary impacts and potential interactions among the variables. By methodically changing these factors and evaluating the resulting mechanical characteristics, optimal printing settings that enhance UTS while reducing material waste and processing time can be determined. The decision to employ the Taguchi technique and specified levels of printing temperature, infill density, and pattern was motivated by the necessity for complete parameter exploration, statistical rigor, and practical applicability to additive manufacturing processes.

The chosen parameter ranges represented practical manufacturing conditions through a combination of methods. The study conducted a thorough literature review on FDM and additive manufacturing processes. This review helped us to understand the typical parameter ranges used in industrial settings, providing a baseline for selecting initial ranges. The empirical knowledge and expertise in additive manufacturing involved considering material compatibility, printer capabilities, and material handling constraints. By leveraging experience, the study refined the parameter ranges to ensure that they were feasible and realistic for production scenarios. The preliminary testing and pilot run before the main experiments were iterative processes that involved testing a range of parameter values to identify each parameter’s upper and lower limits while achieving successful printing and acceptable material properties.

With the use of FTIR analysis, a sample’s reception of waves or particles with various infrared wavelengths is identified. It is accomplished by transmitting infrared light (IR) to component specimens. The material’s absorption spectra of infrared light emission at various frequencies are calculated to assess the molecule’s size and form. FTIR analysis may be used to identify a wide range of chemicals, including unknown substances, additives found in composites, and surface flaws on specific specimen types. It uses an interferometer, a fundamental tool that generates a wave with the majority of the IR wavelengths represented in it, to measure objects.

In addition, FTIR testing may be used to identify impurities in components and assess the oxidation or recovery levels in various composites. The material was inserted into the FTIR spectrometer. The spectrometer calculates the quantity of infrared beams and the frequencies at which the specimen can absorb them by focusing infrared light onto the sample. The best specimens for refraction include contaminants, stains, or coatings on generally smooth, clear materials and moderately flexible substances that are thin enough to fit under inspection using an absorbance spectrometer attached to a magnifying glass. A Perkins Spectrum Two FT-NIR Spectrometer was utilized in the analysis for the current study. It is depicted in Fig. 5(a).

FIG. 5.

(a) Perkins Spectrum Two FT-NIR Spectrometer. (b) Universal testing machine (UTM).

FIG. 5.

(a) Perkins Spectrum Two FT-NIR Spectrometer. (b) Universal testing machine (UTM).

Close modal

The UTM was used in this investigation to determine the tensile parameters of polymer composites at IIT Ropar. These devices use grips that firmly hold either end to retain the specimens being evaluated. The tensile specimen is pulled until it breaks with the help of the other grasp, which holds the material in place, as seen in Fig. 5(b). The UTM used for this test is the Tinius Olsen Model H50KS, which has a capacity of 50 kN. The specimen is initially correctly fastened into the holder’s jaws, and after that, force is applied. The sample will continue to lengthen gradually and steadily during the test. The data collection program will record the material’s test circumstances and any changes to the gauge length. The program monitors the force applied to the specimen and shows the stress–strain curve, which is valuable for analyzing the specimen’s behavior throughout the test.

The surface of a sample is scanned by a focused electron beam using an electron microscope called a scanning electron microscope, which produces pictures of the material. The sample’s surface topography and chemical makeup are revealed by the signals produced by the interaction of the electrons with the atoms in the sample. A picture is produced when an electron beam’s position and the strength of a signal are coupled in a raster scan pattern. The most popular SEM mode uses secondary electron detectors to find secondary electrons generated by excited atoms by the electron beam. The specimen’s topography is one factor that affects the quantity of secondary electrons that may be revealed and, as a result, the strength of the signal.

Taguchi’s technique is frequently applied in the engineering design domain. This strategy, which aims to create a reliable process and produce products of the highest quality, combines tolerance design, parameter design, and system design techniques. Parameter design, an engineering technique for product or process design that emphasizes discovering parameter (factor) settings capable of achieving optimal levels of a quality characteristic (performance measure) with slight fluctuations, is the fundamental component of Taguchi’s methodology.30 It has achieved widespread acceptance in the engineering and scientific communities because Taguchi’s experiment design method is simple to acquire and apply for users with no statistical background. The present study employs three factors to evaluate their influence on tensile characteristics, with the remaining parameters being fixed. As with FDM technology, the qualities of the printed item are substantially impacted by printing settings. Table II displays all three parameters utilized at various levels to generate nine specimens using the Taguchi technique (L9 orthogonal array), as indicated in Table III.

Taguchi distinguished three scenarios, which are as follows:

  • Larger is better.

  • Smaller is better.

  • On-target, minimum-variation.

Taguchi chose the signal-to-noise (S/N) ratio as the preferred quality criterion. Because the standard deviation reduces as the mean drops, the S/N ratio is employed as a quantifiable measure instead of the standard deviation.

The Signal to Noise Ratio (SNRA) and mean values, which represent the three input parameters, are used as the details of the measurement of the output parameter in Table VI. In the current study, a larger signal-to-noise ratio was required to obtain a maximum response for which the data’s features were positive. The formula to determine the SNRA for the larger is better scenario is as follows:
SN=10Log1Y2n,
(2)
where Y = outcomes for the given factor level combination and n = the number of outcomes for the factor level combination.

The MFI of ABS without reinforcement was examined and found to be 12.252 g/10 min according to the ASTM D1238 (230 °C, 3.8 kg load) standard.23 In this research, the composite is inserted inside the MIM. After a 1-min cycle, the machine’s extruded filament is collected and weighted, and then, the MFI for the composite is obtained using Eq. (1). During this test, using the same ASTM D1238 standard, the MFI is calculated to be 12.437 g/10 min at (10–10) wt. % reinforcement of both Al and Cu with ABS (230 °C, 3.8 kg load), as shown in Table IV.

TABLE IV.

Melt flow index of the (Al–Cu–ABS) composite.

MFI
Exp. No.Extruded time (s)(10% Al, 10% Cu, and 80% ABS)
60 1.129 
60 1.188 
60 1.320 
60 1.171 
60 1.165 
60 1.350 
60 1.282 
60 1.332 
60 1.228 
10 60 1.272 
Total 600 12.437 g/10 min 
MFI
Exp. No.Extruded time (s)(10% Al, 10% Cu, and 80% ABS)
60 1.129 
60 1.188 
60 1.320 
60 1.171 
60 1.165 
60 1.350 
60 1.282 
60 1.332 
60 1.228 
10 60 1.272 
Total 600 12.437 g/10 min 

It has been noted that the MFI of the Al–Cu–ABS composite has increased compared to ABS with the addition of copper and aluminum powder to the ABS. In addition, we noticed several physical differences between the (Al–Cu–ABS) composite filament and the ABS filament. Compared to ABS filament, the composite filament created by mixing aluminum and copper powder into ABS is more brittle and darker gray.

The increase in MFI observed in the (Al–Cu–ABS) composite compared to pure ABS can be attributed to the introduction of metal reinforcements, specifically Cu and Al powders. The MFI, determined according to the ASTM D1238 standard, indicates the composite’s melt processability. The higher MFI of 12.437 g/10 min of the composite, in contrast to the 12.252 g/10 min for pure ABS, suggests that adding metal reinforcements alters the flow characteristics during the extrusion process. The physical differences noted between the (Al–Cu–ABS) composite filament and pure ABS filament provide insights into the effects of metal incorporation. The increased brittleness in the composite filament can be attributed to the nature of metal fillers, which may interrupt the polymer matrix’s structural integrity, leading to a more fragile material. The darker gray color of the composite filament further indicates the presence of metal powders, influencing the visual appearance of the final product.

The justification for these observations lies in the changes occurring at the molecular level. Metal fillers, such as Al and Cu, possess different thermal and mechanical properties than the polymer matrix (ABS). Introducing these metals affects the overall composite structure, influencing parameters such as viscosity, polymer chain interactions, and cooling rates during extrusion. It is crucial to explore the implications of the increased MFI and altered physical properties on the printability and mechanical performance of the composite. The brittleness may impact the structural integrity of printed parts, while the color change might be significant for certain applications.

After printing the composite material on a 3D printer, tensile tests were performed on all nine tensile samples of the (Al–Cu–ABS) composite utilizing the UTM without the use of an extensometer to determine the UTS of the composite material we created with the composition of aluminum, copper, and ABS. In addition, we determined peak elongation (mm), break load (N), break elongation (mm), strength at peak (MPa), strength at break (MPa), % elongation at peak, and % elongation at break, as given in Table V. After the test, it was determined from the results that the specimen printed with the printing parameters of infill pattern type of line, printing temperature of 250 °C, and infill density of 20% had the minimum tensile strength of 456 N and the specimen printed with the printing parameters of infill pattern type of zigzag, printing temperature of 245 °C, and infill density of 10% had the maximum tensile strength of 680 N.

TABLE V.

Tensile test readings of all nine tensile samples. Boldface indicates optimal results.

S. No.Infill patternPrinting temperature (°C)Infill density (%)Ultimate tensile strength (N)Peak elongation (mm)Break load (N)Break elongation (mm)Strength at peak (MPa)Strength at break (MPa)% elongation at peak% elongation at break
1. Line (1) 240 10 515 1.45 470 1.48 21.5 19.6 4.4 4.49 
2. Line (1) 245 15 524 1.58 490 2.09 21.8 20.8 4.8 6.33 
3. Line (1) 250 20 456 1.45 446 1.82 19.0 18.6 4.4 5.52 
4. Triangular (2) 240 15 587 1.54 542 1.98 24.5 22.6 4.67 6.01 
5. Triangular (2) 245 20 468 1.32 438 1.71 19.5 18.2 4.00 5.18 
6. Triangular (2) 250 10 553 1.24 460 2.00 23.0 19.2 3.77 6.08 
7. Zigzag (3) 240 20 599 1.51 507 2.33 25.0 21.1 4.57 7.06 
8. Zigzag (3) 245 10 680 1.45 606 2.02 28.3 25.3 4.38 6.11 
9. Zigzag (3) 250 15 642 1.34 565 2.40 26.7 23.5 4.07 7.29 
S. No.Infill patternPrinting temperature (°C)Infill density (%)Ultimate tensile strength (N)Peak elongation (mm)Break load (N)Break elongation (mm)Strength at peak (MPa)Strength at break (MPa)% elongation at peak% elongation at break
1. Line (1) 240 10 515 1.45 470 1.48 21.5 19.6 4.4 4.49 
2. Line (1) 245 15 524 1.58 490 2.09 21.8 20.8 4.8 6.33 
3. Line (1) 250 20 456 1.45 446 1.82 19.0 18.6 4.4 5.52 
4. Triangular (2) 240 15 587 1.54 542 1.98 24.5 22.6 4.67 6.01 
5. Triangular (2) 245 20 468 1.32 438 1.71 19.5 18.2 4.00 5.18 
6. Triangular (2) 250 10 553 1.24 460 2.00 23.0 19.2 3.77 6.08 
7. Zigzag (3) 240 20 599 1.51 507 2.33 25.0 21.1 4.57 7.06 
8. Zigzag (3) 245 10 680 1.45 606 2.02 28.3 25.3 4.38 6.11 
9. Zigzag (3) 250 15 642 1.34 565 2.40 26.7 23.5 4.07 7.29 

Stress vs strain graphs were obtained using the UTM for the nine tensile samples using the following printing parameters: infill pattern (line, triangular, and zigzag), printing temperature (240, 245, and 250 °C), and infill density (10%, 15%, and 20%) based on the L9 orthogonal array indicated in Table III and are shown in Figs. 6, 7, and 8. The tensile properties of the (Al–Cu–ABS) composite were systematically evaluated through a series of tests, providing insights into the material’s behavior under different printing parameters and leading to the following key findings:

  • Infill Pattern Influence: The infill pattern impacts the UTS significantly, as observed in specimens with line, triangular, and zigzag patterns. Notably, the zigzag pattern at 245 °C and 10% infill density exhibited the maximum UTS of 680 N, emphasizing the importance of pattern selection.

  • Printing Temperature Variation: Across different printing temperatures (240, 245, and 250 °C), variations in UTS were evident. For instance, the specimen with a zigzag pattern at 240 °C and 20% infill density showed a UTS of 599 N, indicating the sensitivity of the material to temperature changes.

  • Infill Density Impact: The influence of infill density on UTS is apparent, with variations observed in specimens with 10%, 15%, and 20% infill densities. Notably, the specimen with a line pattern at 250 °C and 20% infill density exhibited the minimum UTS of 456 N, suggesting the need to consider infill density carefully.

FIG. 6.

Stress vs strain graph for tensile samples: (a) infill pattern (line), printing temperature (240 °C), and infill density (10%); (b) infill pattern (line), printing temperature (245 °C), and infill density (15%); and (c) infill pattern (line), printing temperature (250 °C), and infill density (20%).

FIG. 6.

Stress vs strain graph for tensile samples: (a) infill pattern (line), printing temperature (240 °C), and infill density (10%); (b) infill pattern (line), printing temperature (245 °C), and infill density (15%); and (c) infill pattern (line), printing temperature (250 °C), and infill density (20%).

Close modal
FIG. 7.

Stress vs strain graph for tensile samples: (a) infill pattern (triangular), printing temperature (240 °C), and infill density (15%); (b) infill pattern (triangular), printing temperature (245 °C), and infill density (20%); and (c) infill pattern (triangular), printing temperature (250 °C), and infill density (10%).

FIG. 7.

Stress vs strain graph for tensile samples: (a) infill pattern (triangular), printing temperature (240 °C), and infill density (15%); (b) infill pattern (triangular), printing temperature (245 °C), and infill density (20%); and (c) infill pattern (triangular), printing temperature (250 °C), and infill density (10%).

Close modal
FIG. 8.

Stress vs strain graph for tensile samples: (a) infill pattern (zigzag), printing temperature (240 °C), and infill density (20%); (b) infill pattern (zigzag), printing temperature (245 °C), and infill density (10%); and (c) infill pattern (zigzag), printing temperature (250 °C), and infill density (15%).

FIG. 8.

Stress vs strain graph for tensile samples: (a) infill pattern (zigzag), printing temperature (240 °C), and infill density (20%); (b) infill pattern (zigzag), printing temperature (245 °C), and infill density (10%); and (c) infill pattern (zigzag), printing temperature (250 °C), and infill density (15%).

Close modal

The difference in UTS across infill patterns emphasizes how crucial pattern selection is to maximizing material strength. The interconnecting layers that make up the zigzag design proved to be stronger than the line and triangle patterns. The effect that printing temperature has on UTS emphasizes how important it is to regulate temperature precisely. The observed fluctuations suggest that certain patterns and temperatures influence the material’s mechanical performance. The impact of infill density on UTS indicates a trade-off between material strength and density. Though stronger, larger infill densities must be carefully considered not to compromise other desired features. The stress vs strain graphs obtained from the UTM further complement the numerical data, visually representing the material’s behavior under different printing parameters. These graphs offer a comprehensive view of the material’s response to the applied stress and strain, aiding in the interpretation of mechanical properties.

The results highlight the necessity for a customized approach to parameter selection and offer insightful information for improving the fabrication of (Al–Cu–ABS) composites in FDM. Future rounds of the manufacturing process are guided by the findings, making it easier to produce composite materials with improved mechanical qualities suited to certain applications.

The data obtained after using Eq. (2) for the signal-to-noise ratio for a better scenario and the mean values of the nine combinations of input parameters selected by the Taguchi L9 orthogonal array design approach are shown in Table VI.

TABLE VI.

Measuring ultimate tensile strength and SNRA. Boldface indicates optimal results.

S. No.A infill patternB printing temperature (°C)C infill density (%)Ultimate tensile strength (N)Signal to noise ratio (SNRA)Mean
1. Line (1) 240 10 515 54.2361 515 
2. Line (1) 245 15 524 54.3866 524 
3. Line (1) 250 20 456 53.1793 456 
4. Triangular (2) 240 15 587 55.3728 587 
5. Triangular (2) 245 20 468 53.4049 468 
6. Triangular (2) 250 10 553 54.8545 553 
7. Zigzag (3) 240 20 599 55.5485 599 
8. Zigzag (3) 245 10 680 56.6502 680 
9. Zigzag (3) 250 15 642 56.1507 642 
S. No.A infill patternB printing temperature (°C)C infill density (%)Ultimate tensile strength (N)Signal to noise ratio (SNRA)Mean
1. Line (1) 240 10 515 54.2361 515 
2. Line (1) 245 15 524 54.3866 524 
3. Line (1) 250 20 456 53.1793 456 
4. Triangular (2) 240 15 587 55.3728 587 
5. Triangular (2) 245 20 468 53.4049 468 
6. Triangular (2) 250 10 553 54.8545 553 
7. Zigzag (3) 240 20 599 55.5485 599 
8. Zigzag (3) 245 10 680 56.6502 680 
9. Zigzag (3) 250 15 642 56.1507 642 

Signal-to-noise ratios are shown in Table VII as a response. Larger is better for ultimate tensile strength regarding infill pattern, printing temperature, and infill density. Delta denotes evaluating the effect size while determining the difference between the factor’s highest and lowest characteristic average, and rank denotes the ability to determine which variables have the most impact swiftly. The factor with the largest delta value receives rank 1, followed by the second-largest delta, and so on. Table VIII shows the influence of each parameter and its levels, also known as average performance or major effect. The maximum S/N ratio for each factor was chosen to enhance tensile strength, which revealed the ideal parameters. The ideal set of parameters is A3B1C2, as indicated in Table VIII and Fig. 9. Thus, it is evident from Fig. 9 that the infill pattern substantially influences the change in the material’s ultimate tensile strength, while the printing temperature has little impact. Therefore, from Table VIII, the ideal parameter set we selected is A3B1C2, and from Fig. 9, it is evident that the infill pattern of zigzag, printing temperature of 240 °C, and infill density of 15% were the optimal combination for the printing parameters to reach the ultimate tensile strength.

TABLE VII.

Ranking: signal to noise ratios. Boldface indicates optimal results.

LevelA infill patternB printing temperature (°C)C infill density (%)
53.93 55.05 55.25 
54.54 54.81 55.30 
56.12 54.73 54.04 
Delta 2.18 0.32 1.26 
Rank 
LevelA infill patternB printing temperature (°C)C infill density (%)
53.93 55.05 55.25 
54.54 54.81 55.30 
56.12 54.73 54.04 
Delta 2.18 0.32 1.26 
Rank 
TABLE VIII.

Response for means. Boldface indicates optimal results.

LevelA infill patternB printing temperatureC infill density
498.3 567.0 582.7 
536.0 557.3 584.3 
640.3 550.3 507.7 
Delta 142.0 16.7 76.7 
Rank 
LevelA infill patternB printing temperatureC infill density
498.3 567.0 582.7 
536.0 557.3 584.3 
640.3 550.3 507.7 
Delta 142.0 16.7 76.7 
Rank 
FIG. 9.

Main effects plot for SN ratio using Minitab for change in UTM.

FIG. 9.

Main effects plot for SN ratio using Minitab for change in UTM.

Close modal

ANOVA is used to examine if a design parameter and its interaction significantly impact the UTS using the Minitab program. The percentage contribution can identify the major parameter that influences the performance characteristics.31,32 In addition, a 95% confidence level F-test, named after Fisher (1925), may be used to identify the parameter significantly affecting the quality features. The parameter typically substantially impacts the performance when the F value is high. The evidence is tested by the p-value, a probability measured against the null hypothesis.31,33 A lower probability provides a stronger argument against the null hypothesis.34,35 So, based on the values of F and P, as well as the percentage contribution for each parameter, the best results with the most influential parameter are infill pattern (21.70%, 0.044%, and 68.152%), followed by infill density (8.66%, 0.104%, and 27.188%), and finally with the least effect printing temperature (0.48%, 0.674%, and 1.517%), according to Table IX.

TABLE IX.

ANOVA for ultimate tensile strength.

SourceDFSeq. SSAdj. SSAdj. MSFP% Contribution (%)
Infill pattern 7.6077 7.6077 3.803 85 21.70 0.044 68.152 
Printing temperature 0.1694 0.1694 0.084 72 0.48 0.674 1.517 
Infill density 3.0350 3.0350 1.517 51 8.66 0.104 27.188 
Residual error 0.3506 0.3506 0.175 30   3.140 
Total 11.1628     100 
SourceDFSeq. SSAdj. SSAdj. MSFP% Contribution (%)
Infill pattern 7.6077 7.6077 3.803 85 21.70 0.044 68.152 
Printing temperature 0.1694 0.1694 0.084 72 0.48 0.674 1.517 
Infill density 3.0350 3.0350 1.517 51 8.66 0.104 27.188 
Residual error 0.3506 0.3506 0.175 30   3.140 
Total 11.1628     100 

Based on the Taguchi analysis, the best design parameter levels are chosen for the ultimate tensile strength. Utilizing the ideal level of a process parameter, the last stage is to forecast and confirm the improvement of the response. Confirmation tests are conducted to verify the experiments’ outcomes and assess the analysis’s precision. According to the SNRA, the combination of zigzag lines (infill pattern), 240 °C (printing temperature), and 15% (infill density) produced the highest ultimate tensile strength. By experimenting on a particular combination, we can use the Minitab software to determine the prediction values and the signal-to-noise ratio. The values that were predicted and tested are detailed in Table X.

TABLE X.

Confirmation test result for the ultimate tensile strength.

S. No.A infill patternB printing temperature (°C)C infill density (%)Predicted S/N ratioPredicted meanExperimented S/N ratio
1. Zigzag 240 15 56.7426 675.222 56.5889 
S. No.A infill patternB printing temperature (°C)C infill density (%)Predicted S/N ratioPredicted meanExperimented S/N ratio
1. Zigzag 240 15 56.7426 675.222 56.5889 

One popular and adaptable machine learning approach that may be used for classification and regression tasks is the Classification and Regression Tree (CART). The input space is divided into areas recursively, and each region is given a specified class or a predicted numerical value. Leo Breiman introduced CART, a decision tree algorithm-based system.36 CART constructs a tree structure in a classification job to categorize input examples into predetermined classes or categories. Recursively dividing the dataset according to feature values creates the tree. The dataset is split into subsets, and decisions are made at each tree node depending on certain features. This procedure continues until a certain threshold is reached, such as getting to a certain depth or having a certain number of instances in a node. The leaves of the tree represent the final class labels. CART predicts a continuous numerical output rather than discrete classifications in a regression problem. The algorithm builds a tree for every input instance that predicts the target variable. The recursive division of the dataset is based on feature values, just like the classification tree, but the predictions at the leaves are numerical values. The following are the key components of CART, and Fig. 10 shows its description.31,37

  • Splitting Criteria: A splitting criterion, often quantified by metrics such as mean squared error for regression tasks or Gini impurity for classification tasks, informs the choice to separate a node. During the tree-building process, CART seeks to reduce these requirements.

  • Splitting Nodes: A selected characteristic and a threshold value are used to divide nodes. The aim is to find the split that improves the selected impurity metric the most.

  • Stopping Criteria: The recursive splitting process keeps going until a stopping requirement is satisfied. Examples of common stopping conditions are achieving a particular degree of impurity reduction, having a certain number of instances in a node, or reaching a given tree depth.

  • Pruning (Optional): Pruning can be done after the tree is built to avoid overfitting. Pruning is cutting out branches that do not significantly affect how well the model performs with validation data.

FIG. 10.

CART description.

FIG. 10.

CART description.

Close modal

The prevalence of CART in machine learning applications can be attributed to their numerous advantages. Interpretability is one of CART’s main advantages. Decision trees such as CART are naturally simple to comprehend and illustrate. Because of this feature, they are handy for explaining model decisions to stakeholders and non-experts who might not be knowledgeable about machine learning techniques. Users can follow the decision-making process because of the tree structure’s openness, which improves understanding and confidence.

The application of CART analysis over other ML algorithms in the study was carefully considered for several reasons. CART offers high interpretability, crucial for understanding the complex relationships between printing parameters and UTS in metal/polymer composites produced through FDM. The hierarchical tree structure of CART enables us to visually understand how different factors interact and impact UTS, delivering practical insights for process optimization. CART excels in capturing non-linear connections and interaction effects and is ubiquitous in additive manufacturing methods. Furthermore, CART is quite good at identifying intricate and non-linear correlations in the data. CART can handle a wide range of complex patterns in real-world datasets because it can build elaborate decision boundaries, unlike linear models that presume a linear connection between features. Because of its adaptability, CART is a good fit for many applications with non-linear correlations or interactions between variables. CART is a useful tool in many disciplines because of its interpretability, ability to represent non-linear connections, and robustness to outliers. Its broad use may be attributed to its ease of use and capacity to handle complicated patterns, particularly when clear decision-making is essential for user approval and practical execution. CART’s ease of use, interpretability, and versatility have made it a popular choice in various industries, including banking, healthcare, industrial engineering, and natural language processing.10,11

The CART ML algorithm was applied to 27 observations attained by repeating the L9 orthogonal array thrice. Figure 11 shows UTS descriptive stats by the Anderson–Darling test generated by Minitab. The UTS mean of 558 N measures the central tendency, indicating the average UTS value in the dataset. The standard deviation of 72.4950 N and variance of 5255.53 N quantify the spread or dispersion of the UTS values around the mean. A more minor standard deviation suggests a more concentrated distribution. The quartiles and range of 227.676 N provide information about the distribution’s shape and the extent of variability in the UTS values. The minimum (453.036 N) and maximum (680.712 N) values give the range within which most UTS observations fall. The skewness of 0.165 75 N indicates the asymmetry of the UTS distribution. A positive skewness suggests a slightly right-skewed distribution, where the tail on the right side is longer or fatter than that on the left. These descriptive statistics are essential for understanding the central tendency, variability, and distribution shape of the UTS data.

FIG. 11.

Ultimate tensile strength descriptive stats by the Anderson–Darling test.

FIG. 11.

Ultimate tensile strength descriptive stats by the Anderson–Darling test.

Close modal

The CART used “node splitting,” which suggests that the algorithm employs a least squared error criterion to determine how the decision tree nodes should be divided. This criterion aims to minimize the sum of squared differences between predicted UTS and actual UTS at each node during the model-building process. The “model validation” used a tenfold cross-validation approach. The technique involves dividing the dataset into ten subsets, iteratively using nine subsets for training and the remaining one for testing. This process was repeated ten times, providing a robust assessment of the model’s generalization performance across different subsets of the data. The “optimal tree” indicates that the model selection involves choosing a tree complexity within one standard error of the maximum R-squared. This implies a trade-off between model goodness-of-fit (R-squared) and simplicity, favoring simpler models if their performance is close to the best-performing model within a certain level of uncertainty. Figure 12 shows variation in the terminal node from 2 to 4; the R-square value increases and reaches a maximum of 0.8922 at the terminal node of 4.

FIG. 12.

Ultimate tensile strength: change in terminal node.

FIG. 12.

Ultimate tensile strength: change in terminal node.

Close modal

Figure 13 depicts an indication of the decision-making process of a decision tree model by the CART for UTS prediction. “Count” indicates the number of trials in the four unique subgroups with comparable material properties identified in the “terminal node” section. “StDev” indicates the standard deviation of anticipated UTS values, showing dispersion around the mean within each subgroup, while “fit” shows the predicted UTS, averaged from actual values. Subjects are classified into terminal nodes based on specific criteria related to the features “infill pattern” and “infill density,” and each terminal node is associated with a fit value indicating the predicted outcome or performance measure for subjects falling into that node. The regression tree’s analysis of the terminal nodes reveals distinct patterns in predicted UTS values based on infill patterns. Terminal node 1, characterized by an infill pattern less than or equal to 2.5, exhibits an average UTS of 558.00 with a standard deviation of 71.1399, indicating a relatively consistent and homogeneous strength profile.

FIG. 13.

Ultimate tensile strength: optimal tree.

FIG. 13.

Ultimate tensile strength: optimal tree.

Close modal

In contrast, terminal node 4, representing materials with an infill pattern greater than or equal to 2.5, displays a higher average UTS of 639.667 with a lower standard deviation of 32.3859, suggesting tighter clustering around the mean and greater consistency. Terminal node 1 shows an average UTS of 517.167 and a standard deviation of 45.4404, suggesting a somewhat higher variability in UTS values than node 4. This comparative analysis underscores the influence of infill patterns on UTS predictions, highlighting how materials with specific infill characteristics can lead to varying levels of strength uniformity and predictability in the context of the generated regression tree.

Table XI overviews a UTS CART predictive model’s key statistics and performance metrics. The CART model utilizes three predictors, which are considered significant for the model, indicating that each predictor contributes meaningfully to the model’s predictive accuracy. The decision tree structure includes four terminal nodes representing distinct subgroups in the data. The minimum terminal node size is set at six, implying that each node contains at least six observations.

TABLE XI.

UTS: CART performance metric.

StatisticsTrainingTest
R-squared 0.9154 0.8922 
Root mean squared error (RMSE) 20.6965 23.3594 
Mean squared error (MSE) 428.3455 545.6631 
Mean absolute deviation (MAD) 15.2416 17.2801 
Mean absolute percent error (MAPE) 0.0258 0.0293 
StatisticsTrainingTest
R-squared 0.9154 0.8922 
Root mean squared error (RMSE) 20.6965 23.3594 
Mean squared error (MSE) 428.3455 545.6631 
Mean absolute deviation (MAD) 15.2416 17.2801 
Mean absolute percent error (MAPE) 0.0258 0.0293 

The R-squared value of 0.9154 for the training data suggests that the model explains ∼91.54% of the variability in the training data. It indicates a strong fit of the CART model to the observed outcomes of UTS. The RMSE of 20.6965 represents the average magnitude of errors between the predicted and actual UTS values. A lower RMSE values indicate a better CART model performance. The MSE of 428.3455 is the average squared difference between the predicted and actual UTS values, and its lower values indicate a better performance. The MAD of 15.2416 represents the average absolute difference between the predicted and actual UTS values. It offers insight into the dispersion of errors. The performance metrics for the test data provide an assessment of how well the model generalizes to new, unseen data. The R-squared value of 0.8922 indicates a high level of explanatory power on the test dataset. The slightly higher RMSE (23.3594) and MSE (545.6631) compared to the training data suggest a reduced accuracy on the test set, which is expected. The MAPE provides a percentage representation of the average magnitude of errors relative to the actual values. A lower MAPE indicates a better accuracy. The training MAPE of 0.0258 and test MAPE of 0.0293 suggest a relatively small percentage error, indicating a good predictive performance. The CART model summary comprehensively evaluates the regression model’s fit and predictive accuracy, detailing the importance of predictors, tree structure, and key performance metrics on both the training and test datasets.

One of the most critical metrics for evaluating each variable’s contribution to the predicted accuracy of a CART model is its relative relevance, as shown in Fig. 14(a). The CART algorithm continually chooses variables for dividing nodes throughout the construction of the decision tree. These criteria might include mean squared error for regression or Gini impurity for classification. A variable’s significance is determined by calculating its enhancement to the splitting criterion on all nodes in which it is utilized. However, it is essential to understand that variable significance reveals relative contributions rather than the direction of the connection between variables and the target variable. Infill pattern has 100% relative importance, followed by infill density of 28.0% and printing temperature with minor relative importance.

FIG. 14.

(a) Variables’ importance for ultimate tensile strength, (b) actual UTS and fit UTS: scatterplot, (c) box plot of residuals, and (d) box plot of UTS by terminal node.

FIG. 14.

(a) Variables’ importance for ultimate tensile strength, (b) actual UTS and fit UTS: scatterplot, (c) box plot of residuals, and (d) box plot of UTS by terminal node.

Close modal

A scatterplot of UTS fits vs actual UTS values visually examines how well a CART model’s predictions align with the observed or true values, as shown in Fig. 14(b). It is an intuitive way to assess the accuracy of a predictive model. The points form a diagonal line, indicating a perfect match between the predicted and actual UTS. Deviations from this diagonal line suggest discrepancies between the model’s predictions and the real outcomes. For UTS, CART is a well-fitted model. The scatterplot will exhibit a tight cluster of points around the diagonal line. A box plot of UTS residuals in a CART model in Fig. 14(c) is a graphical representation that provides insights into the distribution, spread, and characteristics of the model’s prediction errors. Analyzing the box plot of residuals indicates a well-fitted model has residuals evenly distributed around zero, with no distinct patterns or outliers. A box plot of UTS by a terminal node in Fig. 14(d) illustrates the distribution of UTS values within different subgroups or terminal nodes identified by a CART model. Each box in the plot corresponds to a specific terminal node, and the plot allows for a comparative analysis of the UTS distribution across these nodes.

1. Statistical validation of the ML models

Statistical validation of ML models is critical for determining accuracy, generalizability, and detecting biases. It entails comparing model predictions to actual data and assessing performance using measures such as R-squared and RMSE.38 Cross-validation is a validation approach that ensures models generalize well to new data while detecting overfitting or underfitting. This procedure helps with model selection, hyperparameter adjustment, and optimizing prediction accuracy. Statistical validation also validates assumptions, which improves model interpretability.39,40 It guarantees that ML models are dependable, durable, and capable of generating correct predictions on previously unknown data, making it a critical phase in model development.41 

Evaluating the performance and dependability of the CART model is a necessary step in the statistical validation process.42 The “paired t-test” and the “test and confidence interval for two variances,” two popular statistical tests, can be used for this.

  • Paired t-test: The means of two similar groups are compared using the paired t-test. The test may determine whether there is a substantial discrepancy between the model’s actual UTS values and predictions in the context of CART model validation. Usually, the alternative hypothesis proposes a considerable difference between the means, whereas the null hypothesis believes that there is no significant difference between the means. A result that is statistically significant might mean that the model’s predictions and the actual values are very different from one another.

  • Test and Confidence Interval for Two Variances: To determine if there is a substantial difference between the anticipated and actual values variability, the “test and confidence interval for two variances” is utilized. This test can assist in determining if the model correctly depicts the data distribution. A noteworthy result might suggest that the model’s predictions and the actual results are inconsistent. The confidence intervals provide the range that the genuine variance is most likely to fall inside for the variances.

In summary, these statistical tests offer valuable insights into the performance and reliability of the CART model. The paired t-test assesses differences in means, highlighting potential discrepancies between the predicted and actual values, while the test and confidence interval for two variances provides information about the consistency and spread of the model’s predictions compared to the observed data. These tests contribute to a robust validation process, helping ensure the model’s effectiveness in making accurate predictions and generalizing well to new data.

a. Paired t-test.

Table XII summarizes the outcomes of a paired t-test conducted to evaluate the agreement between the observed UTS values and the UTS values predicted by the CART model. Descriptive statistics for both sets of UTS values are presented, including the mean, standard deviation, and standard error of the mean. The estimation for the paired difference reveals that, on average, there is no systematic difference between the observed UTS values and the fitted UTS values, as the mean of the paired difference is close to zero. The standard deviation of the paired difference and the standard error of the mean provide measures of variability and precision, respectively. The 95% confidence interval (−8.34, 8.34) for the population mean difference further supports the notion that the true difference is not significantly different from zero. The null hypothesis, stating no significant difference between the observed and fitted UTS values, is reinforced by a high P-value (1), indicating a lack of evidence to reject the null hypothesis. Therefore, the results of the paired t-test suggest that the CART model’s UTS predictions align well with the observed UTS values, affirming the model’s reliability for the given dataset.

TABLE XII.

Results of the paired t-test.

Descriptive statistics
SampleNMeanStDevSE mean
Ultimate tensile strength (N) 27 558 72.5 14 
Ultimate tensile strength (N) (Fit) 27 558 69.4 13.3 
Estimation for paired difference
MeanStDevSE mean95% CI for μ˙difference
0.00 21.09 4.06 (−8.34, 8.34)  
μ˙difference: Population mean of (ultimate tensile strength (N) − ultimate tensile strength (fit)) 
Test
Null hypothesis H0: μ˙difference = 0    
Alternative hypothesis H1: μ˙difference ≠ 0    
T-valueP-value
0.00   
Descriptive statistics
SampleNMeanStDevSE mean
Ultimate tensile strength (N) 27 558 72.5 14 
Ultimate tensile strength (N) (Fit) 27 558 69.4 13.3 
Estimation for paired difference
MeanStDevSE mean95% CI for μ˙difference
0.00 21.09 4.06 (−8.34, 8.34)  
μ˙difference: Population mean of (ultimate tensile strength (N) − ultimate tensile strength (fit)) 
Test
Null hypothesis H0: μ˙difference = 0    
Alternative hypothesis H1: μ˙difference ≠ 0    
T-valueP-value
0.00   

Figure 15, which includes a histogram of differences in Fig. 15(a), an individual value plot of differences in Fig. 15(b), and a box plot of differences in Fig. 15(c), provides a comprehensive visual examination of the discrepancies between the observed UTS values and the values predicted by the CART model. The histogram in Fig. 15(a) provides a distributional perspective on the discrepancies, highlighting the frequency or density of different sizes of variances between the expected and actual UTS values. It makes comprehending the prediction errors’ general pattern and primary trend easier. In Figure 15(b), each point represents a specific observation, allowing for a detailed inspection of individual discrepancies. The box plot in Fig. 15(c) adds another layer of insight by presenting a summary of the distribution of differences. The box indicates the interquartile range, and the whiskers extend to show the range of most values. These visualizations help assess the overall pattern, distribution, and potential outliers in the differences between the observed and predicted UTS values. Such a holistic view is instrumental in comprehensively understanding the CART model’s predictive performance and identifying areas where adjustments or further investigation may be warranted.

FIG. 15.

(a) Histogram of differences, (b) individual value plot of differences, and (c) box plot differences.

FIG. 15.

(a) Histogram of differences, (b) individual value plot of differences, and (c) box plot differences.

Close modal
b. Test and Confidence Interval for Two Variances.

Table XIII presents the outcomes of the “test and confidence interval for two variances,” specifically focusing on the standard deviations (σ) of the UTS and the fitted UTS values, along with the ratio of these standard deviations. The descriptive statistics provide information about the variability of the observed and anticipated UTS values for both UTS and fitted UTS. The intervals estimate where genuine standard deviations are most likely to fall inside. Levene’s and Bonett’s approaches compute the 95% confidence intervals and the estimated ratio of standard deviations (σ12). The ratio measures the model’s ability to represent the variability found in the dataset accurately. The test aims to assess whether there is a significant difference in variability between the observed UTS values and the fitted UTS values. Two methods, Bonett’s and Levene’s methods, are employed to conduct the test. The test statistics, degrees of freedom (DF1 and DF2), and P-values are provided for each method. A low P-value typically indicates evidence to reject the null hypothesis in favor of unequal variances.

TABLE XIII.

Results of “test and confidence interval for two variances.”

σ1: Standard deviation of ultimate tensile strength (N);
σ2: Standard deviation of ultimate tensile strength (fit);
Ratio: σ12; Bonett’s and Levene’s methods are valid for continuous distribution.
Descriptive statistics
Variable StDev Variance 95% CI for σ 
Ultimate tensile strength (N) 27 72.495 5255.532 (60.867, 93.103) 
Ultimate tensile strength (fit) 27 69.359 4810.712 (60.418, 85.856) 
Ratio of standard deviations
Estimated ratio 95% CI for ratio using Bonett’s method 95% CI for ratio using Levene’s method   
1.045 21 (0.784, 1.356) (0.723, 1.482)   
Test
Null hypothesis H0: σ12 = 1    
Alternative hypothesis H1: σ12 ≠ 1    
Significance level α = 0.05    
Method Test statistic DF1 DF2 P-value 
Bonett 0.11  0.737 
Levene 0.04 52 0.846 
σ1: Standard deviation of ultimate tensile strength (N);
σ2: Standard deviation of ultimate tensile strength (fit);
Ratio: σ12; Bonett’s and Levene’s methods are valid for continuous distribution.
Descriptive statistics
Variable StDev Variance 95% CI for σ 
Ultimate tensile strength (N) 27 72.495 5255.532 (60.867, 93.103) 
Ultimate tensile strength (fit) 27 69.359 4810.712 (60.418, 85.856) 
Ratio of standard deviations
Estimated ratio 95% CI for ratio using Bonett’s method 95% CI for ratio using Levene’s method   
1.045 21 (0.784, 1.356) (0.723, 1.482)   
Test
Null hypothesis H0: σ12 = 1    
Alternative hypothesis H1: σ12 ≠ 1    
Significance level α = 0.05    
Method Test statistic DF1 DF2 P-value 
Bonett 0.11  0.737 
Levene 0.04 52 0.846 

The 95% confidence intervals for the standard deviations of UTS and fitted UTS values suggest a certain level of overlap, indicating potential similarity in the variances. The estimated ratio of standard deviations is close to 1 (1.045 21), and the corresponding confidence intervals, using Bonett’s and Levene’s methods, include 1. It implies no substantial evidence to reject the null hypothesis of equal variances. The test statistics (0.11 for Bonett’s method and 0.04 for Levene’s method) are small, and the associated P-values are considerably higher than the significance level (α = 0.05). These results further support the lack of significant evidence to reject the null hypothesis. Therefore, based on this test, the variability in the observed UTS values is not significantly different from the variability in the fitted UTS values, suggesting that the model adequately captures the variability observed in the dataset. A visual description is shows in Fig. 16.

FIG. 16.

(a) Test and confidence interval for two variances, (b) histogram, and (c) individual value plot.

FIG. 16.

(a) Test and confidence interval for two variances, (b) histogram, and (c) individual value plot.

Close modal

1. Practical implications

  • Enhanced Manufacturing Precision: The study results provide firms in the additive manufacturing sector with valuable recommendations and optimum printing parameters to improve the mechanical qualities of metal/polymer composites. This results in more consistent and dependable final products, decreasing material strength variability and improving production accuracy.

  • Cost Efficiency and Material Optimization: Efficient material usage is made possible by the research’s improved printing settings. Manufacturers with more control over the printing process might benefit from lower material waste. This helps increase production efficiency, which is important for the financial viability of additive manufacturing.43 

  • Tailored Material Properties: Manufacturers can customize the mechanical characteristics of metal/polymer composites to satisfy particular application needs by utilizing the determined ideal conditions. This degree of personalization is essential in sectors such as aerospace, automotive, and medical applications, where materials must meet strict performance requirements.

2. Theoretical implications

  • Advanced Understanding of Additive Manufacturing Dynamics: This work explores the intricate relationships between printing parameters and material characteristics, adding to the theoretical underpinnings of additive manufacturing. CART analysis contributes to the theoretical comprehension of various factors’ hierarchical impact on composites’ mechanical performance, providing valuable information for the next research projects.44 

  • Modeling Framework for Parameter Interactions: This study offers a modeling framework for comprehending the complex relationships between FDM printing parameters. This framework provides a systematic method for modeling and forecasting the effects of various factors on material characteristics, which benefits the larger fields of materials science and manufacturing.45,46

3. Educational, industry collaboration, innovation and economic implications

  • Curriculum Development: This study’s results can help shape the creation of educational institutions’ materials science and additive manufacturing curricula. Its technique and optimum parameter settings may be included in training programs to guarantee that future professionals possess the knowledge and skills required by the industry.

  • Industry-wide Best Practices: The additive manufacturing community may collaborate and standardize using the determined ideal printing settings as industry best practices. The knowledge exchange regarding parameter optimization facilitates progress and fosters creativity.47 

  • Competitive Advantage: Manufacturers with the best printing conditions to create metal/polymer composites with exceptional mechanical qualities have a competitive edge. This makes their goods more marketable and establishes them as leaders in the quickly changing field of additive manufacturing.48 

  • Reduced Material Waste: Through more effective material use brought about by the enhanced printing conditions, waste in the additive manufacturing process is decreased. This supports ecologically friendly production techniques and is consistent with sustainability standards.49 

  • Safer and More Reliable Products: This study’s conclusions have applications in the aerospace and medical device industries, which depend on metal/polymer composites. Goods produced with an increased UTS have a beneficial effect on end users and society as a whole by improving safety and dependability.50 

The advancements in 3D printing technology have profound implications for both biomedical and engineering fields, leading to transformative changes and innovative solutions.51 

1. Biomedical applications

  • Patient-Specific Medical Devices: 3D printing enables the creation of personalized implants, prosthetics, and orthopedic devices based on specific patient anatomy. This customization enhances patient comfort, functionality, and treatment results.52 

  • Tissue Engineering and Regenerative Medicine: Bioprinting methods enable the creation of sophisticated tissue scaffolds and organoids.53 This has important implications for regenerative medicine, drug testing, and personalized medicine techniques.54,55

  • Surgical Planning and Training: 3D-printed models of human anatomy help in preoperative planning, improve surgical precision, and save operative time. They are also useful resources for medical training and teaching.56 

2. Engineering applications

  • Complex Geometry and Lightweight Structures: 3D printing creates sophisticated shapes and lightweight structures that are difficult or impossible to construct using standard manufacturing processes. This is especially advantageous to the aerospace, automobile, and defense sectors.57 

  • Rapid Prototyping and Iteration: Engineers can swiftly prototype and iterate ideas with 3D printing, shortening the product development cycle and allowing for design optimization.58 

  • Customized Manufacturing: The technology enables on-demand and customized part and component production, lowering inventory costs and waste while increasing supply chain efficiency.59 

In general, the implications of 3D printing in biomedical and technical applications are extensive. They include improved patient care, better product functioning, shorter production lead times, and greater design flexibility. As technology advances, its effect on these industries will expand, fueling innovation and changing the future of industry and health care.60 To summarize, this study has practical, theoretical, educational, industrial, economic, environmental, and social implications. It also fosters improvements in materials science and additive manufacturing processes.

Incorporating metal fillers in ABS matrices through FDM presents a promising avenue for enhancing the mechanical properties of composites. This study showcased the successful fabrication of MPCs with 10% Al and 10% Cu, achieving a notable MFI of 12.437 g/10 min using a single screw extruder machine. The meticulous exploration of printing parameters demonstrated their pivotal role in influencing the UTS of the composites. Notably, the variation in UTS from 456 to 680 N underscores the sensitivity of mechanical properties to parameter adjustments. Applying the Taguchi L9 orthogonal array method facilitated the identification of optimal parameters, with the combination of a printing temperature of 240 °C, an infill density of 15%, and a zigzag infill pattern yielding the highest UTS. These findings contribute valuable insights for optimizing the fabrication of metal/polymer composites using FDM, guiding future endeavors in material design and additive manufacturing. The MFI for the composite is observed to rise compared to ABS polymer, and the color change is seen. The filament created by the MIM for the composite was revealed to be brittle compared to the ABS filament. The application of machine learning leads to the following significant conclusions:

  • The CART analysis was conducted to model and predict the UTS of metal/polymer composites fabricated through FDM. The tree structure was constructed with three predictors: printing temperature, infill pattern, and infill density. The constructed tree consists of four terminal nodes, representing distinct subgroups of materials with similar UTS characteristics. The tree was pruned to avoid overfitting, with a minimum terminal node size set at six.

  • The R-squared value of 0.9154 on the training data indicates a high level of explanatory power, suggesting that the selected predictors effectively capture the variability in UTS. The model’s ability to generalize to new data is supported by a respectable R-squared value of 0.8922 on the test data. All three predictors, printing temperature, infill pattern, and infill density, were important in predicting UTS. The tree structure highlights these parameters’ hierarchical importance and interaction effects in determining material strength.

  • Examination of terminal nodes revealed specific patterns. For instance, materials with an infill pattern greater than 2.5 exhibited an average predicted UTS of 639.667, with a relatively low standard deviation of 32.3859, indicating consistency in strength within this subgroup.

  • The paired t-test conducted on the observed and predicted UTS values revealed no significant difference, supporting the model’s reliability. In addition, the Test and Confidence Interval for Two Variances indicated no significant difference in variability between observed and predicted UTS values.

  • The CART analysis provided valuable insights into the complex relationships between printing parameters and UTS in metal/polymer composites manufactured through FDM. The identified optimal parameters and subgroup characteristics contribute to the advancement of process optimization in additive manufacturing, fostering the development of robust and predictable materials.

The study focused on a specific composition of metal/polymer composites (80% ABS, 10% Al, and 10% Cu), and the results may not be directly generalizable to other compositions. The research assumes a uniform distribution of fillers within the matrix, which might not capture variations in filler dispersion that could impact mechanical properties. The study is conducted under controlled laboratory conditions, and real-world variations in FDM processes may introduce additional complexities.

In the future, researchers can combine different metals to create novel composites and determine the melt flow characteristics of such composites for specific applications. Other qualities, such as electrical and thermal ones, can also be tested by researchers. The varied printing qualities can also be tested by adjusting other printing settings. The best test results may be revealed using a variety of optimization techniques. Furthermore, the study will be extended to explore the impact of additional parameters, such as layer thickness and print speed, on composite properties. Next, the applicability of the optimized printing parameters to different composite compositions can be investigated. More machine-learning techniques can be explored for predictive modeling to capture complex interactions among multiple parameters. By addressing these aspects, future research can further refine and broaden the applicability of the findings, advancing the field of additive manufacturing.

The authors extend their appreciation to the Deanship of Scientific Research at King Khalid University for funding this work through Large group Project RGP-2/529/44.

The authors declare no conflict of interest.

All authors listed have significantly contributed to the development and the writing of this article.

Mukesh Singh Manola: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Funding acquisition (equal); Investigation (equal); Methodology (equal); Project administration (equal); Resources (equal); Software (equal); Supervision (equal); Validation (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Balwant Singh: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Funding acquisition (equal); Investigation (equal); Methodology (equal); Project administration (equal); Resources (equal); Software (equal); Supervision (equal); Validation (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Manoj Kumar Singla: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Funding acquisition (equal); Investigation (equal); Methodology (equal); Project administration (equal); Resources (equal); Software (equal); Supervision (equal); Validation (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Jasgurpreet Singh Chohan: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Funding acquisition (equal); Investigation (equal); Methodology (equal); Project administration (equal); Resources (equal); Software (equal); Supervision (equal); Validation (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Raman Kumar: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Funding acquisition (equal); Investigation (equal); Methodology (equal); Project administration (equal); Resources (equal); Software (equal); Supervision (equal); Validation (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Yashwant Singh Bisht: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Funding acquisition (equal); Investigation (equal); Methodology (equal); Project administration (equal); Resources (equal); Software (equal); Supervision (equal); Validation (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Raman Kumar: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Funding acquisition (equal); Investigation (equal); Methodology (equal); Project administration (equal); Resources (equal); Software (equal); Supervision (equal); Validation (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Meshel Q. Alkahtani: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Funding acquisition (equal); Investigation (equal); Methodology (equal); Project administration (equal); Resources (equal); Software (equal); Supervision (equal); Validation (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Saiful Islam: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Funding acquisition (equal); Investigation (equal); Methodology (equal); Project administration (equal); Resources (equal); Software (equal); Supervision (equal); Validation (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Muhammad Imam Ammarullah: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Funding acquisition (equal); Investigation (equal); Methodology (equal); Project administration (equal); Resources (equal); Software (equal); Supervision (equal); Validation (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal).

The data generated during the research conducted in the present work are available within the article.

1.
C.
Zhao
,
J.
Kang
,
Y.
Li
,
Y.
Wang
,
X.
Tang
, and
Z.
Jiang
,
Cyborg Bionic Syst.
4
,
0022
(
2023
).
2.
B. N.
Turner
,
R.
Strong
, and
S. A.
Gold
,
Rapid Prototyping J.
20
(
3
),
192
204
(
2014
).
3.
J.
Gan
,
F.
Li
,
K.
Li
,
E.
Li
, and
B.
Li
,
Compos. Sci. Technol.
234
,
109928
(
2023
).
4.
G. N.
Levy
,
R.
Schindel
, and
J. P.
Kruth
,
CIRP Ann.
52
(
2
),
589
609
(
2003
).
5.
Y.
Zhang
,
Z.
Huang
,
H.
Wang
, and
J.
Li
,
ACS Appl. Mater. Interfaces
15
(
27
),
32984
32992
(
2023
).
6.
X.
Li
,
H.
Yu
,
H.
Feng
,
S.
Zhang
, and
Y.
Fu
,
Cyborg Bionic Syst.
4
,
0025
(
2023
).
7.
C.
Xiao
,
K.
Zheng
,
S.
Chen
,
N.
Li
,
X.
Shang
,
F.
Wang
,
J.
Liang
,
S. B.
Khan
,
Y.
Shen
,
B.
Lu
,
H.
Ma
, and
Z.
Chen
,
Addit. Manuf.
71
,
103607
(
2023
).
8.
A. T.
Prakoso
,
H.
Basri
,
D.
Adanta
,
I.
Yani
,
M. I.
Ammarullah
,
I.
Akbar
,
F. A.
Ghazali
,
A.
Syahrom
, and
T.
Kamarul
,
Biomedicines
11
,
427
(
2023
).
9.
A.
Zainal
,
M.
Yanis
,
M. Z.
Kadir
,
Astuti
,
P.
Akbar Teguh
,
S.
Edo
,
S.
Ardiyansyah
, and
B.
Hasan
, paper presented at the
Proceedings of the 4th Forum in Research, Science, and Technology (FIRST-T1-T2-2020)
,
2021
.
10.
K.
Badogu
,
V.
Thakur
,
R.
Kumar
,
R.
Kumar
, and
S.
Singh
, “
Acrylonitrile butadiene styrene-ZrO2 composites for roller burnishing as post-processing of 3D printed parts: Machine learning modeling using classification and regression trees
,”
J. Mater. Eng. Perform.
(published online)
(
2023
).
11.
V.
Thakur
,
R.
Kumar
,
R.
Kumar
,
R.
Singh
, and
V.
Kumar
,
J. Thermoplast. Compos. Mater.
37
(
2
),
466
492
(
2024
).
12.
M.
Nikzad
,
S. H.
Masood
, and
I.
Sbarski
,
Mater. Des.
32
(
6
),
3448
3456
(
2011
).
13.
F.
Castles
,
D.
Isakov
,
A.
Lui
,
Q.
Lei
,
C. E. J.
Dancer
,
Y.
Wang
,
J. M.
Janurudin
,
S. C.
Speller
,
C. R. M.
Grovenor
, and
P. S.
Grant
,
Sci. Rep.
6
(
1
),
22714
(
2016
).
14.
B.
Singh
,
R.
Kumar
,
J. S.
Chohan
,
S.
Singh
,
C. I.
Pruncu
,
M. L.
Scutaru
, and
R.
Muntean
,
Materials
14
,
3504
(
2021
).
15.
K. A.
Hamzah
,
C. K.
Yeoh
,
M. M.
Noor
,
P. L.
Teh
,
Y. Y.
Aw
,
S. A.
Sazali
, and
W. M. A.
Wan Ibrahim
,
J. Thermoplast. Compos. Mater.
35
(
1
),
3
16
(
2022
).
16.
A. R.
Torrado Perez
,
D. A.
Roberson
, and
R. B.
Wicker
,
J. Failure Anal. Prev.
14
(
3
),
343
353
(
2014
).
17.
S.
Hwang
,
E. I.
Reyes
,
K.-s.
Moon
,
R. C.
Rumpf
, and
N. S.
Kim
,
J. Electron. Mater.
44
(
3
),
771
777
(
2015
).
18.
B.
Khatri
,
K.
Lappe
,
D.
Noetzel
,
K.
Pursche
, and
T.
Hanemann
,
Materials
11
,
189
(
2018
).
19.
N.
Sa’ude
,
M.
Ibrahim
, and
M.
Ibrahim
,
ARPN J. Eng. Appl. Sci.
11
(
10
),
6562
6567
(
2016
), https://www.arpnjournals.org/jeas/research_papers/rp_2016/jeas_0516_4314.pdf.
20.
T. J.
Quill
,
M. K.
Smith
,
T.
Zhou
,
M. G. S.
Baioumy
,
J. P.
Berenguer
,
B. A.
Cola
,
K.
Kalaitzidou
, and
T. L.
Bougher
,
Appl. Compos. Mater.
25
(
5
),
1205
1217
(
2018
).
21.
B.
Çantı
,
M.
Aydin
, and
F.
Yıldırım
, Development of novel nanocomposite materials for 3D printing (
2016
).
22.
F. A.
Chávez
,
P. A.
Quiñonez
, and
D. A.
Roberson
,
J. Thermoplast. Compos. Mater.
34
(
9
),
1193
1212
(
2021
).
23.
M. A.
Ryder
,
D. A.
Lados
,
G. S.
Iannacchione
, and
A. M.
Peterson
,
Compos. Sci. Technol.
158
,
43
50
(
2018
).
24.
D.
Fischer
,
C.
Eßbach
,
P.
Neumann
,
D.
Dietrich
,
T.
Mehner
,
A.
Dittes
,
T.
Lampke
, and
D.
Nickel
, Novel filament materials for RoHS-compliant electroplating of FDM-derived plastic parts (
2021
).
25.
R. A. A.
Helmi
,
M. G. M.
Johar
, and
M. A. S. B. M.
Hafiz
, paper presented at the
1st International Conference in Advanced Innovation on Smart City, ICAISC 2023—Proceedings
,
2023
.
26.
K.
Dissanayake
,
M. G. M.
Johar
, and
J.
Indones
,
Electrical Eng. Comput. Sci.
32
(
1
),
381
391
(
2023
).
27.
L.
McKeen
, in
The Effect of Sterilization Methods on Plastics and Elastomers
, 4th ed., edited by
L.
McKeen
(
William Andrew Publishing
,
2018
), pp.
63
90
.
28.
S. K.
Gawali
,
G. C.
Pandey
, and
P. K.
Jain
,
Adv. Mater. Process. Technol.
9
(
2
),
391
401
(
2023
).
29.
I.
Akbar
,
M. L.
King
,
Z.
Fatoni
,
T. P. O.
Sianipar
,
Sukarmansyah
, and
A. T.
Prakoso
,
Int. J. Res. Vocat. Stud.
3
(
1
),
52
57
(
2023
).
30.
A. S.
Sidhu
,
R.
Kumar
,
S.
Singh
, and
H.
Kaur
, paper presented at the
Intelligent Manufacturing and Energy Sustainability
,
Singapore
,
2024
.
31.
K.
Dissanayake
and
M. G.
Md Johar
,
Appl. Comput. Intell. Soft Comput.
2021
,
5581806
.
32.
A. E.
Wan
,
M. S. B.
Khan
,
B. S. X.
Teo
,
J.
Khan
,
I.
Abdullah
,
M.
Kaleemullah
,
F.
Asmani
,
M.
Suofeiya
,
S.
Al Dhalli
,
Z.
Kasim
,
S.
Fattepur
, and
M. R. M.
Rasny
,
Int. J. Med. Toxicol. Legal Med.
23
(
3-4
),
169
178
(
2020
).
33.
Z.
Deldar
,
H.
Ekhtiari
,
H. R.
Pouretemad
, and
A.
Khatibi
,
Front. Psychiatry
10
,
776
(
2019
).
34.
N.
Al-Sharah
,
O.
Khoury
,
J. A.
Hamid
,
I. A.
Ariffin
, and
J.
Tham
,
Eurasian J. Appl. Linguist.
7
(
2
),
112
124
(
2021
), https://ejal.info/menuscript/index.php/ejal/article/view/110.
35.
S.
Fattepur
,
K. C.
Nilugal
,
R.
Rajendran
,
F.
Asmani
, and
E.
Yusuf
,
Asian J. Pharm. Clin. Res.
11
(
15
),
8
12
(
2018
).
36.
A.
Astuti
,
A.
Costa
,
A. T.
Prakoso
,
I.
Yani
, and
Y.
Resti
,
Indones. J. Eng. Sci.
4
(
1
),
075
081
(
2023
).
37.
Z.
Li
and
J.
Tham
,
Proc. SPIE
12941
,
129410H
(
2023
).
38.
M. I.
Abdullah
,
S. K.
Hao
,
I.
Abdullah
, and
S.
Faizah
, paper presented at the
2023 IEEE 14th Control and System Graduate Research Colloquium, ICSGRC 2023—Conference Proceeding
(
IEEE
,
2023
).
39.
R. H.
Elghanuni
,
R. A. A.
Helmi
, and
M. I.
Abdullah
,
AIP Conf. Proc.
2398
,
050001
(
2022
).
40.
R. A. A.
Helmi
,
A. T. L.
Lee
,
M. G.
Md Johar
,
A.
Jamal
, and
L. F.
Sim
, paper presented at the
ISCAIE 2021—IEEE 11th Symposium on Computer Applications and Industrial Electronics
(
IEEE
,
2021
).
41.
R. A. A.
Helmi
,
R. H.
Elghanuni
, and
M. I.
Abdullah
, paper presented at the
2021 IEEE 12th Control and System Graduate Research Colloquium, ICSGRC 2021—Proceedings
(
IEEE
,
2021
).
42.
S.
Pathirana
,
D.
Asirvatham
, and
M. G.
M'D Johar
, paper presented at the
Procedia Computer Science
,
2019
.
43.
J.
Wang
,
Z.
Xi
,
B.
Niu
,
R.
Gao
, and
Z.
Xu
,
Sustainability
16
,
3009
(
2024
).
44.
T.
Liu
,
P.
Feng
,
Y.
Bai
,
S.
Bai
,
J.-Q.
Yang
, and
F.
Zhao
,
Eng. Struct.
308
,
117962
(
2024
).
45.
D. K.
Pratiwi
,
Y.
Yudianto
,
A.
Mataram
, and
A. T.
Prakoso
,
J. Mech. Sci. Eng.
9
(
1
),
007
013
(
2022
).
46.
A.
Teguh Prakoso
,
A.
Rendiko Ichsan
,
A.
Syahrom
,
A.
Putra Md Saad
,
A.
Hadi Abdul Wahab
,
M. A.
Sulong
,
F. A.
Mohd Ghazali
, and
H.
Basri
,
J. Phys.: Conf. Ser.
1500
(
1
),
012023
(
2020
).
47.
J.
Xiang
,
J.
Chen
,
Y.
Zheng
,
P.
Li
,
J.
Huang
, and
Z.
Chen
,
Eng. Anal. Boundary Elem.
162
,
28
44
(
2024
).
48.
A. P.
Md Saad
,
A. T.
Prakoso
,
M. A.
Sulong
,
H.
Basri
,
D. A.
Wahjuningrum
, and
A.
Syahrom
, “
Impacts of dynamic degradation on the morphological and mechanical characterisation of porous magnesium scaffold
,”
Biomech. Model. Mechanobiol.
18
(
3
),
797
811
(
2019
).
49.
Q.
Zhu
,
J.
Chen
,
G.
Gou
,
H.
Chen
, and
P.
Li
,
J. Mater. Process. Technol.
246
,
267
275
(
2017
).
50.
H.
Basri
,
A.
Syahrom
,
A. P. M.
Saad
,
A. A.
Rabiatul
,
P.
Akbar Teguh
,
A.
Diansyah
, and
R. U.
Putra
,
E3S Web Conf.
68
,
01020
(
2018
).
51.
H.
Basri
,
A. T.
Prakoso
,
M. A.
Sulong
,
A. P.
Md Saad
,
M. H.
Ramlee
,
D.
Agustin Wahjuningrum
,
S.
Sipaun
,
A.
Öchsner
, and
A.
Syahrom
,
Proc. Inst. Mech. Eng., Part L
234
(
1
),
175
185
(
2020
).
52.
S.-J.
Kim
,
M.-G.
Kim
,
J.
Kim
,
J. S.
Jeon
,
J.
Park
, and
H.-G.
Yi
,
Cyborg Bionic Syst.
4
,
0043
(
2023
).
53.
C. K.
Cheng
,
H. A.
Bakar
,
M.
Gollasch
, and
Y.
Huang
,
Cardiovasc. Drugs Ther.
32
(
5
),
481
502
(
2018
).
54.
A. F.
Mohamed
,
M.
Isahak
,
M. Z.
Awg Isa
, and
R.
Nordin
,
Pertanika J. Sci. Technol.
30
(
3
),
2225
2252
(
2022
).
55.
H.
Basri
,
A. T.
Prakoso
,
Z.
Abidin
,
A.
Syahrom
,
I.
Akbar
, and
D.
Adanta
,
CFD Lett.
15
(
7
),
61
73
(
2023
).
56.
Y.
Xu
,
F.
Zhang
,
W.
Zhai
,
S.
Cheng
,
J.
Li
, and
Y.
Wang
,
Polymers
14
,
566
(
2022
).
57.
H.
Basri
,
J. D.
Nasution
,
A.
Syahrom
,
M. A.
Sulong
,
A. P.
Md Saad
,
A. T.
Prakoso
,
F.
Aminin
, and
M. J.
Fundam
,
Malays. J. Fundam. Appl. Sci.
13
,
546
552
(
2017
).
58.
S. U.
Pathiratne
,
A.
Khatibi
, and
M. G.
Md Johar
,
Int. J. Lean Six Sigma
9
(
4
),
543
561
(
2018
).
59.
M.
Joyce
,
T.
Hodgkinson
,
M.
Lemoine
,
A.
González-Vázquez
,
D. J.
Kelly
, and
F. J.
O’Brien
,
Eur. Cells Mater.
45
,
158
172
(
2023
).
60.
M. N.
Mohammed
,
H. R.
Hamid
,
S.
Al-Zubaidi
,
N. S.
Zamani
, and
M. I.
Abdullah
, paper presented at the
Proceeding—2019 IEEE 7th Conference on Systems, Process and Control, ICSPC 2019
(
IEEE
,
2019
).