⚡ electrical engineering

Multi-response optimization of 3D printing parameters by Taguchi method and machine learning

This study demonstrates that combining the Taguchi method with Random Forest Regression effectively optimizes FDM printing parameters for ABS PA-747H, significantly enhancing tensile and flexural mechanical properties while validating the improvements through statistical analysis and fractographic evidence.

Original authors: Mohd Nazri Ahmad, Mohamad Ridzwan Ishak

Published 2026-07-06

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Mohd Nazri Ahmad, Mohamad Ridzwan Ishak

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are a chef trying to bake the perfect loaf of bread. You have a recipe (the 3D printer), ingredients (plastic filament), and a few knobs you can turn: how thick you slice the dough, which way you stack the layers, how much dough you pack into the pan, and how fast you push it through the oven.

This paper is about a team of researchers who acted like master chefs for 3D printing. They wanted to figure out exactly how to turn those knobs to make the strongest, stiffest plastic parts possible using a material called ABS (a tough plastic often used in LEGO bricks and car bumpers).

Here is the story of their experiment, broken down into simple parts:

1. The Problem: The "Layer Cake" Weakness

When you 3D print something, it's not made in one solid piece like a stone statue. It's built layer by layer, like a cake.

The Issue: If the layers don't stick together perfectly, the cake falls apart easily when you pull on it or bend it.
The Goal: The researchers wanted to find the perfect "recipe" so the layers fused into a single, super-strong unit.

2. The Detective Work: The Taguchi Method

Instead of trying every single combination of settings (which would take forever and waste a lot of plastic), they used a smart detective tool called the Taguchi Method.

The Analogy: Imagine you have 4 ingredients (Layer Thickness, Orientation, Infill Density, Speed) and 3 choices for each. Instead of baking 81 different cakes to find the best one, the Taguchi method is like a magic map that tells you exactly which 9 cakes to bake to figure out the winner.
The Variables:
- Layer Thickness: How thin or thick each "slice" of plastic is.
- Orientation: Which way the part is sitting on the printer bed (flat, on its side, or standing up).
- Infill Density: How much plastic is inside. 100% means it's solid; 0% means it's hollow.
- Printing Speed: How fast the printer head moves.

3. The Big Discovery: The "Golden Recipe"

After baking their 9 test cakes and pulling on them until they broke, they found a clear winner. The "Golden Recipe" (A1B1C1D1) was:

Thin Layers: 0.2 mm (like a very thin slice of bread).
Flat Orientation: 0 degrees (lying flat).
Solid Fill: 100% density (no air pockets inside).
Slow Speed: 10 mm/s (moving very slowly to let the plastic melt and stick perfectly).

The Result:
When they used this recipe, the plastic got a massive upgrade:

Tensile Strength (Pulling power): Went from 30 MPa to 40 MPa. (Imagine a rope that could hold 30 people suddenly holding 40).
Flexural Strength (Bending power): Went from 69 MPa to 92 MPa. (It became much harder to snap).
Stiffness: The material became much more rigid, resisting bending like a steel beam rather than a rubber band.

4. The Crystal Ball: Machine Learning

The researchers didn't just stop at baking; they taught a computer (using a method called Random Forest Regression) to look at their results and predict what would happen next.

The Analogy: It's like showing a student 9 math problems and their answers, then asking them to solve a 10th one. The computer learned the "hidden rules" of how the printer works.
The Success: The computer was incredibly accurate. It could predict how strong a part would be just by knowing the settings, without needing to print it first. It confirmed that Orientation and Infill Density were the most important ingredients.

5. Looking Under the Hood: The "Crack" Detective

To understand why the best recipe worked, they looked at the broken pieces under a powerful microscope (SEM).

The Bad Cakes: The parts made with the wrong settings had voids (tiny air bubbles) and the layers peeled apart like a poorly glued book. The cracks traveled straight through the weak spots.
The Golden Cake: The parts made with the optimal settings had no gaps. The layers were fused so tightly that the cracks had a hard time moving. Instead of peeling apart, the material held together, proving the "glue" between the layers was strong.

The Bottom Line

This paper proves that you don't need a super-expensive printer to get great results; you just need the right settings. By slowing down, printing flat, filling the part completely, and using thin layers, you can turn a standard 3D printed plastic part into something significantly stronger and more reliable.

They also showed that combining old-school statistical detective work (Taguchi) with modern computer brains (Machine Learning) is a powerful way to solve engineering puzzles without wasting time or money.

Technical Summary: Multi-response Optimization of 3D Printing Parameters by Taguchi Method and Machine Learning

Problem Statement
Fused Deposition Modeling (FDM) is a widely adopted additive manufacturing technology, yet the mechanical performance of printed components often falls short of conventionally manufactured parts. This limitation is primarily attributed to process-induced defects such as poor interlayer adhesion, anisotropic behavior, internal porosity, and inconsistent material deposition. These issues significantly compromise tensile strength, flexural strength, stiffness, and structural reliability. While various studies have explored FDM parameters, there is a specific gap in the simultaneous optimization of multiple mechanical responses (tensile strength, flexural strength, tensile modulus, and flexural modulus) for the commercial-grade ABS PA-747H material. Furthermore, limited research integrates traditional statistical optimization with machine learning to predict these properties effectively.

Methodology
This study employed a hybrid approach combining the Taguchi method, Analysis of Variance (ANOVA), and Machine Learning (Random Forest Regression) to optimize FDM printing parameters for ABS PA-747H.

Material and Fabrication: ABS PA-747H granules were processed into filament using a hot-press compounding machine and a twin-screw extruder to ensure homogeneity.
Experimental Design: An L9 orthogonal array was utilized to evaluate four control factors at three levels each:
- Layer Thickness: 0.2, 0.3, 0.4 mm
- Build Orientation: 0°, 45°, 90°
- Infill Density: 100%, 50%, 0%
- Printing Speed: 10, 50, 100 mm/s
Mechanical Testing: Specimens were fabricated using a FlashForge Creator Pro 3D printer. Tensile and flexural properties were measured according to ASTM D638 and ASTM D790 standards, respectively, using a universal testing machine.
Statistical Analysis: The Signal-to-Noise (S/N) ratio (using the "larger-the-better" criterion) and ANOVA were used to identify optimal parameter combinations and determine the significance of each factor. Normality of data was verified using Anderson–Darling tests.
Machine Learning: A Random Forest Regression (RFR) model was developed to predict mechanical properties based on the input parameters. The model's performance was validated using 5-fold cross-validation, with metrics including $R^2$ , Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE).
Fractographic Analysis: Scanning Electron Microscopy (SEM) was conducted to observe fracture surfaces, crack propagation paths, and interlayer bonding quality.

Key Results

Optimal Parameters: The Taguchi analysis identified the optimal combination as A1B1C1D1: 0.2 mm layer thickness, 0° build orientation, 100% infill density, and 10 mm/s printing speed.
Performance Improvements: Under these optimized conditions, significant improvements in mechanical properties were observed compared to non-optimized specimens:
- Tensile strength increased from 30 MPa to 40 MPa.
- Flexural strength increased from 69 MPa to 92 MPa.
- Tensile modulus improved to 1.90 GPa.
- Flexural modulus showed the most dramatic improvement, rising from 0.60 GPa to 2.15 GPa (approximately 258% increase).
Factor Significance: ANOVA results indicated that infill density was the most significant factor affecting mechanical properties (lowest p-values), followed by build orientation. Layer thickness and printing speed also contributed but to a lesser extent.
Machine Learning Performance: The Random Forest Regression model demonstrated high predictive accuracy, successfully capturing the nonlinear relationships between parameters and mechanical responses. Feature importance analysis confirmed that build orientation and infill density were the dominant predictors.
Fractography: Optimized specimens exhibited fewer voids, stronger interlayer bonding, and shorter crack propagation paths. In contrast, non-optimized samples displayed extensive interlayer delamination, void-induced fracture initiation, and brittle transverse fractures.

Significance and Claims
The paper claims that integrating the Taguchi optimization method with Random Forest Regression provides a comprehensive framework for enhancing the structural integrity and reliability of FDM-printed ABS PA-747H components. The study demonstrates that:

Statistical Optimization: The Taguchi method effectively identifies parameter combinations that maximize mechanical performance while minimizing experimental trials.
Predictive Capability: Machine learning models, specifically Random Forest, can accurately predict mechanical properties, reducing the dependence on extensive physical experimentation.
Process Understanding: The correlation between optimized parameters and improved microstructural features (reduced voids, better bonding) validates the physical mechanisms behind the performance gains.

The authors conclude that this hybrid approach offers a robust strategy for improving process stability and part quality in additive manufacturing, specifically for engineering applications requiring high mechanical performance from ABS PA-747H.