RPI ID: 2023-001
Innovation Summary: A real-time monitoring system for laser powder bed fusion (L-PBF) additive manufacturing processes leverages time-frequency domain analysis of melt pool imaging data to detect anomalies during fabrication. Sequential melt pool images, captured via near-infrared or thermal imaging, are compressed into a time series signal and transformed into a spectrogram capturing geometry-dependent frequency characteristics. A nominal performance baseline is established using statistical decomposition techniques such as principal component analysis (PCA), enabling characterization of expected machine behavior across varying raster scan patterns. During operation, reconstructed spectrograms are compared against the nominal basis, and deviations quantified via reconstruction error enable unsupervised, statistically driven anomaly detection with minimal latency and reduced data labeling requirements.
Challenges / Opportunities: Additive manufacturing processes, particularly L-PBF, are highly sensitive to variations in energy deposition, scan patterns, and thermal dynamics, leading to defects that degrade part quality. Existing monitoring approaches either lack real-time capability or rely on “black-box” machine learning models that are difficult to interpret and generalize across geometries. This innovation addresses these limitations by introducing an interpretable, geometry-aware detection framework using time-frequency analysis. It creates opportunities for improved process control, reduced material waste, and scalable quality assurance across different build configurations and machines.
Key Benefits / Advantages: ✔ Real-time, in-situ anomaly detection with low latency during printing ✔ Geometry-aware monitoring using spectrogram-based frequency signatures of scan patterns ✔ Unsupervised learning approach reduces need for labeled datasets ✔ Interpretable statistical detection method (e.g., PCA-based reconstruction error) ✔ Applicable across multiple raster patterns and part geometries ✔ Enables detection of both temporal and spatial defects (e.g., melt pool instability, voids, spatter)
Applications: • Monitoring and quality control in metal additive manufacturing (L-PBF systems) • Aerospace and automotive component fabrication • Industrial process control and smart manufacturing systems • Research and development in advanced manufacturing and materials processing
Keywords: Additive manufacturing, laser powder bed fusion, melt pool monitoring, spectrogram analysis, anomaly detection, PCA
Intellectual Property: Published U.S. Patent Application No. 18/668708