A team of three graduate researchers has developed a machine learning-based approach that can predict the quality of 3D-printed parts before they are manufactured, potentially reducing waste, saving time, and improving efficiency across the manufacturing sector.

While 3D printing is often viewed as a straightforward process of turning a digital design into a physical object, the reality is far more complex. Manufacturers must determine numerous process parameters, including print speed, layer thickness, build orientation, and nozzle temperature. Incorrect settings can result in weak, rough, or unusable parts, while optimal settings can produce components that are strong, smooth, and ready for practical applications.

Traditionally, identifying the best printing parameters requires repeated trial-and-error testing. Engineers print a part, evaluate its performance, adjust settings, and repeat the process until satisfactory results are achieved. This approach can be time-consuming, costly, and wasteful.

Seeking a more efficient solution, graduate researchers Nur Mohammad Ali, Jayanta Bhusan Deb, and Shilpa Chowdhury conducted a study to determine whether machine learning could accurately predict part quality without the need for physical printing and testing.

The research team produced 27 carefully designed specimens using polylactic acid (PLA), one of the most widely used biodegradable materials in additive manufacturing. Each specimen was tested for two critical performance indicators: mechanical strength and surface quality.

The experimental data were then used to train multiple machine learning models capable of learning relationships between printing parameters and final product performance. The objective was to determine whether these models could accurately predict the strength and surface smoothness of a part based solely on its printing settings.

According to the findings, published in Elsevier’s Decision Analytics Journal, the answer was yes.

The researchers evaluated five different machine learning approaches and identified highly accurate predictive models for both performance measures. The best-performing model for strength prediction, an Extremely Randomized Tree model, achieved an average prediction error of only 2.20 percent. For surface quality prediction, a Random Forest model achieved an average prediction error of 9.28 percent. Both significantly outperformed conventional prediction methods.

The study also revealed that not all printing parameters contribute equally to final product quality. Build orientation and layer thickness emerged as the two most influential factors affecting performance. The researchers suggest that this insight can help manufacturers prioritize the parameters that matter most when optimizing production processes.

“Imagine knowing the quality of a part before it’s printed. That’s not science fiction anymore-it’s what our models can do,” the researchers noted.

The findings arrive at a time when additive manufacturing is becoming increasingly important to the United States’ industrial strategy. Aerospace companies use 3D printing to manufacture lightweight components, healthcare providers rely on it for customized medical devices, and defense contractors employ it to produce complex parts that are difficult or impossible to create through traditional manufacturing methods.

Despite its growing adoption, the industry continues to face challenges related to quality consistency, material waste, and production efficiency. The researchers believe that predictive machine learning models could help address these issues by enabling manufacturers to evaluate quality before production begins.

Although the current study focused on PLA-based components, the researchers emphasize that the same methodology can be extended to metals, composites, and other advanced materials used in critical industries. As a result, the work represents more than a single research achievement; it offers a scalable framework for smarter and more efficient manufacturing.

What began as a question posed by three graduate researchers may ultimately contribute to transforming how products are designed, tested, and manufactured in the future.

About the Research Team-

Nur Mohammad Ali

Originally from Bangladesh, Nur Mohammad Ali is a graduate researcher in Industrial and Systems Engineering. He contributed to data investigation, simulation analysis, and manuscript preparation for the study. His research interests include machine learning, smart manufacturing, and supply chain innovation.

Jayanta Bhusan Deb

Jayanta Bhusan Deb is a researcher in Mechanical and Aerospace Engineering. He led the conceptualization of the study and designed the machine learning framework. His research focuses on data-driven prediction modeling in additive manufacturing and composite materials.

Shilpa Chowdhury

Shilpa Chowdhury is a researcher in Information Systems and Analysis. She developed and optimized the traditional machine learning models used to validate the team’s ensemble approach and contributed to the analysis, review, and editing of the research paper.