• HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
Saturday, June 21, 2025
BIOENGINEER.ORG
No Result
View All Result
  • Login
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
No Result
View All Result
Bioengineer.org
No Result
View All Result
Home NEWS Science News Technology

Unsupervised Porosity Segmentation in Laser Powder Fusion

Bioengineer by Bioengineer
June 1, 2025
in Technology
Reading Time: 5 mins read
0
ADVERTISEMENT
Share on FacebookShare on TwitterShare on LinkedinShare on RedditShare on Telegram

blank

In the rapidly evolving landscape of additive manufacturing, ensuring the structural integrity and reliability of 3D printed components remains a paramount challenge. Among the various defects that compromise the quality and performance of these parts, porosity—tiny, often microscopic voids within the material—poses significant concerns. Addressing this intricate issue, a groundbreaking study has introduced an innovative methodology that harnesses the power of unsupervised machine learning to segment porosity in parts produced through laser powder bed fusion (LPBF). This approach leverages the state-of-the-art “Segment Anything” model, promising to revolutionize quality control protocols in additive manufacturing by enabling promptable and highly accurate porosity segmentation without the need for extensive labeled datasets.

Laser powder bed fusion, emblematic of precision in the additive manufacturing domain, involves selectively melting layers of metal powder to build complex geometries layer by layer. However, the process is inherently susceptible to porosity formation resulting from various factors including improper melting parameters, powder contamination, or gas entrapment. Traditional inspection techniques—ranging from destructive testing to X-ray computed tomography (CT)—though effective, are expensive, time-consuming, or require specialist interpretation. Moreover, conventional image analysis methods demand extensive manual labeling of defects, limiting scalability and real-time application. This research milestone circumvents these obstacles by deploying an unsupervised approach capable of prompting the segmentation algorithm to automatically and accurately identify porosity features, thus dramatically lowering the barriers for industrial adoption.

By integrating the novel Segment Anything framework, originally developed for broad and flexible image segmentation tasks, the authors have fine-tuned the model to specialize in the unique challenge of detecting voids in LPBF-manufactured parts. Their approach eschews reliance on supervised learning where models are trained on vast, meticulously annotated datasets. Instead, it leverages intrinsic patterns within the data to segment regions exhibiting void-like characteristics based on subtle contrasts and textural anomalies in scanned imagery such as CT data. This unsupervised paradigm marks a significant leap forward, particularly pertinent in scenarios where annotated porosity datasets are scarce or infeasible to obtain.

.adsslot_NdOD9SKjJp{ width:728px !important; height:90px !important; }
@media (max-width:1199px) { .adsslot_NdOD9SKjJp{ width:468px !important; height:60px !important; } }
@media (max-width:767px) { .adsslot_NdOD9SKjJp{ width:320px !important; height:50px !important; } }

ADVERTISEMENT

What truly sets this methodology apart is its promptability—the capacity for operators or quality assurance systems to guide the segmentation process with minimal input. Using intuitive prompts, the algorithm dynamically hones in on regions of interest, delivering segmentation masks that delineate porosity with remarkable precision. This interactive capacity not only enhances the adaptability of the system across different materials and printing parameters but also accelerates defect detection workflows by minimizing human-in-the-loop annotation overhead. The real-time feedback loop embedded within the approach stands to transform in-process monitoring and post-build inspection practices.

Intricately detailed in this study are the underlying technical innovations that enable such a transformative capability. The researchers enhanced feature extraction mechanisms within the Segment Anything architecture to develop high-sensitivity representations, capable of discriminating between true porosity voids and imaging artifacts. This was achieved by integrating advanced normalization layers and contrast enhancement techniques tailored to the grayscale volumetric data characteristic of LPBF scans. Additionally, multi-scale analysis empowered the model to capture porosity across a range of spatial resolutions, recognizing both micron-sized pores and larger, irregular void conglomerates.

The implications of this technology extend beyond mere defect identification. Accurate porosity mapping facilitates predictive maintenance by correlating porosity distribution patterns to mechanical failures, thereby informing adaptive control strategies for LPBF machines. Manufacturers can leverage detailed porosity segmentation to fine-tune process parameters, optimize powder reuse protocols, and implement targeted post-processing treatments such as hot isostatic pressing. This holistic feedback mechanism fosters the production of stronger, more reliable parts with reduced scrap rates and improved lifecycle performance.

Industrial stakeholders will find particular value in the model’s versatility. The unsupervised, promptable nature ensures rapid deployment across diverse LPBF platforms and materials, eliminating the laborious need to construct bespoke training datasets for each new application. Moreover, the model’s architecture is readily extensible to other additive manufacturing defects such as cracks or inclusions, setting the stage for a comprehensive defect surveillance ecosystem powered by advanced artificial intelligence.

The research team conducted rigorous validation experiments using both synthetic and real LPBF datasets, demonstrating superior segmentation accuracy compared to state-of-the-art supervised models. Importantly, their approach exhibited enhanced robustness to variations in scan resolution and noise—typical challenges in industrial nondestructive evaluation. Such resilience is critical for adoption in production environments, where imaging conditions can fluctuate and defect manifestations vary unpredictably.

From a computational perspective, the model’s efficient inference algorithms facilitate near real-time analysis, a feature that paves the way for integration with edge computing solutions directly on the manufacturing floor. This convergence of AI and Industry 4.0 paradigms promotes autonomous quality control whereby defects can be identified and addressed immediately, reducing lead times and operational costs.

The study not only presents a novel algorithmic contribution but also embodies a paradigm shift towards democratizing advanced defect detection in additive manufacturing. By dispensing with reliance on large annotated datasets and introducing a prompt-driven unsupervised mechanism, the path is laid for broader dissemination of cutting-edge AI tools in niche manufacturing segments. This democratization can catalyze innovation, increase competitiveness, and enhance safety standards across sectors dependent on high-integrity metal parts—from aerospace to biomedical implants.

Looking ahead, the researchers envision expanding the scope of their method to encompass multi-modal data inputs, integrating thermal, acoustic, and optical signals to enrich the defect detection signal space. Such fusion approaches promise even greater fidelity in identifying subtle and complex porosity patterns. Furthermore, coupling the segmentation outputs with machine learning models predicting mechanical properties could establish comprehensive digital twins of manufactured parts, bridging the gap between microstructural defects and macroscopic performance.

In summary, this pioneering work constitutes a landmark achievement in employing unsupervised, promptable segmentation techniques for porosity detection in laser powder bed fusion additive manufacturing. By seamlessly marrying advanced computer vision models with domain-specific adaptations, it paves the way for agile, scalable, and highly effective quality assurance solutions. As additive manufacturing ventures from prototyping into full-scale production, innovations like this will be indispensable for unlocking its true potential and assuring the manufacture of defect-resilient components critical to tomorrow’s technological landscape.

Subject of Research: Porosity segmentation in laser powder bed fusion additive manufacturing using unsupervised machine learning.

Article Title: An unsupervised approach towards promptable porosity segmentation in laser powder bed fusion by segment anything

Article References:
Era, I.Z., Ahmed, I., Liu, Z. et al. An unsupervised approach towards promptable porosity segmentation in laser powder bed fusion by segment anything. npj Adv. Manuf. 2, 10 (2025). https://doi.org/10.1038/s44334-025-00021-4

Image Credits: AI Generated

Tags: additive manufacturing defect detectionadvancements in quality assurance for additive manufacturingautomated porosity analysis in 3D printingchallenges in metal powder fusionefficiency in porosity detection methodsinnovative methodologies in LPBFlaser powder bed fusion quality controlnon-destructive testing techniques for porosityporosity in 3D printed componentsSegment Anything model for image analysisstructural integrity in additive manufacturingunsupervised machine learning for porosity segmentation

Share12Tweet8Share2ShareShareShare2

Related Posts

blank

Urban Trees Viewed More Negatively Post-COVID Lockdowns

June 21, 2025
High-Momentum 2D Emission Coupled to Surface Resonance

High-Momentum 2D Emission Coupled to Surface Resonance

June 21, 2025

Ultrathin Membranes Enable Fast, Selective Ion Transport

June 21, 2025

Phosphor-Free White LEDs Emit Vibrant Yellow-Green Light

June 20, 2025

POPULAR NEWS

  • Green brake lights in the front could reduce accidents

    Study from TU Graz Reveals Front Brake Lights Could Drastically Diminish Road Accident Rates

    161 shares
    Share 64 Tweet 40
  • New Study Uncovers Unexpected Side Effects of High-Dose Radiation Therapy

    76 shares
    Share 30 Tweet 19
  • Pancreatic Cancer Vaccines Eradicate Disease in Preclinical Studies

    71 shares
    Share 28 Tweet 18
  • How Scientists Unraveled the Mystery Behind the Gigantic Size of Extinct Ground Sloths—and What Led to Their Demise

    65 shares
    Share 26 Tweet 16

About

We bring you the latest biotechnology news from best research centers and universities around the world. Check our website.

Follow us

Recent News

New Translational Read-Through Drugs for Fanconi Anemia

Urban Trees Viewed More Negatively Post-COVID Lockdowns

Truncated LKB1 Mimics Smac to Boost Fas Apoptosis

  • Contact Us

Bioengineer.org © Copyright 2023 All Rights Reserved.

Welcome Back!

Login to your account below

Forgotten Password?

Retrieve your password

Please enter your username or email address to reset your password.

Log In
No Result
View All Result
  • Homepages
    • Home Page 1
    • Home Page 2
  • News
  • National
  • Business
  • Health
  • Lifestyle
  • Science

Bioengineer.org © Copyright 2023 All Rights Reserved.