A machine learning-enhanced ultrasonic monitoring system for real-time porosity detection in additive manufacturing

This technology uses machine learning and ultrasonic sensors beneath the build plate in metal 3D printing to detect and classify tiny pores in real time, enabling immediate quality control and defect correction during the manufacturing process.

Background

Additive manufacturing, particularly Laser Powder Bed Fusion (LPBF), has rapidly advanced the production of complex metal components, offering unprecedented design freedom and material efficiency. However, the field faces persistent challenges related to ensuring the structural integrity and reliability of manufactured parts. One of the most critical issues is the formation of porosity—microscopic voids or clusters of voids within the material—that can significantly compromise mechanical properties such as strength, fatigue life, and overall durability. As LPBF is increasingly adopted for high-performance applications in aerospace, biomedical, and automotive industries, the demand for effective, real-time quality assurance methods has grown. Reliable detection and control of porosity during the manufacturing process are essential to prevent costly failures and to meet stringent certification standards.

Current approaches to porosity detection in LPBF are largely inadequate for real-time, in-process monitoring. Traditional ex-situ techniques, such as X-ray computed tomography (XRCT) and scanning electron microscopy (SEM), provide high-resolution imaging but are expensive, time-consuming, and unsuitable for integration into the manufacturing workflow. In-situ alternatives, including synchrotron X-ray and thermal imaging, are either prohibitively costly or lack the necessary spatial resolution to detect small-scale defects. Conventional ultrasonic testing (UT), while promising for non-destructive evaluation, struggles with the unique challenges posed by LPBF: limited resolution at the micrometer scale, complex signal interpretation due to heterogeneous microstructures, the need for coupling agents, and practical difficulties in mounting sensors within the build environment. Additionally, higher-frequency UT signals, which could improve resolution, suffer from increased attenuation, making it difficult to extract meaningful information from the weakened signals. These limitations underscore the pressing need for a robust, high-resolution, and real-time porosity detection solution tailored to the demands of LPBF additive manufacturing.

Technology

This technology integrates an array of ultrasonic transducers beneath the build plate of a Laser Powder Bed Fusion (LPBF) additive manufacturing system to enable real-time, layer-by-layer detection of porosity defects. After each powder-melt cycle, the system emits ultrasonic pulses through the newly formed metal layer, capturing the reflected echoes as time-series data at multiple probe points. These acoustic signals are then processed using advanced machine learning models—trained on extensive finite element simulations of titanium alloy samples with known pore characteristics—to classify the average pore diameter, spatial distribution, and cluster density with micrometer-level precision. The methodology focuses on identifying clusters of voids, which generate more detectable signals, rather than individual pores, thus providing actionable insights for immediate process adjustments and in-process defect remediation.

What differentiates this solution is its unique combination of high-resolution ultrasonic testing and machine learning, overcoming the limitations of traditional ex-situ inspection methods like X-ray CT and SEM, which are costly, slow, and unsuitable for real-time monitoring. Unlike conventional ultrasonic testing, which is hampered by signal attenuation, resolution constraints, and complex interpretation, this approach leverages simulation-driven data and sophisticated algorithms—such as CNNs and HIVE-COTE V2—to achieve high classification accuracy for pore location, size, and density. The system’s ability to process data rapidly and adaptively during the build process enables unprecedented quality control and parameter optimization, making it a transformative tool for producing defect-free, high-performance metal components in critical applications.

Benefits

  • Enables real-time, in-process detection of porosity defects during Laser Powder Bed Fusion (LPBF) manufacturing.
  • Achieves high-resolution porosity characterization down to tens of micrometers, improving defect detection accuracy.
  • Utilizes machine learning models to classify pore size, spatial distribution, and cluster density with high accuracy.
  • Facilitates immediate corrective parameter adjustments to optimize build quality and reduce defects.
  • Offers a cost-effective and practical alternative to traditional ex-situ and expensive in-situ porosity detection methods.
  • Improves structural integrity, strength, and fatigue life of manufactured metal components by controlling porosity.
  • Supports scalable implementation with efficient training times suitable for real-time manufacturing environments.
  • Provides a foundation for automated quality assurance and process optimization in metal additive manufacturing.

Applications

  • Real-time LPBF quality assurance
  • Automated defect correction in 3D printing
  • In-process porosity monitoring
  • Adaptive process parameter optimization
  • Critical aerospace part manufacturing

Additional info

A system employs an array of ultrasonic transducers beneath an LPBF build plate. After each melt cycle, pulses propagate through the new layer, and reflected echoes are captured as time-series data. A machine learning model, trained on simulated pore clusters, processes this data to classify pore diameter, spatial distribution, and cluster density in real-time, enabling layer-by-layer porosity detection and corrective adjustments.