AI Framework Predicts Hidden Defects that Weaken Metal 3D-Printed Parts

Key Takeaways

• DSML AI predicts yield strength with 9.51 MPa MAE.
• Symbolic regression provides interpretable equations linking porosity to strength.
• Four‑fold accuracy boost over traditional defect assessment methods.
• Enables defect‑aware design maps for aerospace and automotive parts.
• Cuts material development time from weeks to seconds.

Pulse Analysis

Metal additive manufacturing has reshaped how aerospace and automotive engineers approach lightweight design, but its promise is tempered by microscopic pores that form during laser powder‑bed fusion. These hidden defects act like internal voids, reducing load‑bearing area and jeopardizing safety‑critical components. Traditional qualification relies on extensive mechanical testing and costly microscopy, creating bottlenecks that slow certification and increase part cost. As supply chains demand faster iteration, the industry has turned to data‑driven methods that can infer material performance without destructive experiments.

The research team at POSTECH introduced a data‑selective machine learning (DSML) framework that isolates the most influential process variables—laser power, scan speed, layer thickness, and measured porosity—to predict yield strength of AlSi10Mg parts. By coupling DSML with symbolic regression, the model outputs human‑readable equations rather than opaque predictions, satisfying both accuracy and interpretability requirements. Validation on a diverse set of laser‑powder‑bed samples yielded a mean absolute error of just 9.5 MPa, a four‑fold improvement over conventional statistical approaches, and delivered results in seconds instead of weeks.

These capabilities translate into a defect‑aware design map that lets engineers select process windows that meet target strength before printing a single part. For aerospace OEMs, this reduces the risk of in‑service failure and accelerates the certification timeline; automotive manufacturers can shorten development cycles and lower inventory of trial builds. As the framework expands to other alloys and integrates with real‑time sensor data, it could become a core component of closed‑loop AM production, driving broader commercialization of metal 3D‑printing across safety‑critical markets.