Powder Bed Image Segmentation for Real-Time Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing powder bed additive manufacturing technologies rely on open loop control and ex-situ inspection, which are inadequate for producing high-quality parts with low rejection rates, and current image analysis algorithms are limited by machine specificity, localization accuracy, and discriminatory power, lacking generalizability and adaptability.
Innovation Solution
A dynamic segmentation convolutional neural network (DSCNN) for pixel-wise anomaly classification operates on multi-scale input layers, providing real-time, machine-agnostic anomaly detection across various powder bed additive manufacturing machines and imaging systems, enabling efficient transfer of learned knowledge and reducing data collection burdens.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional image analysis algorithms are used for powder bed additive manufacturing anomaly detection, then the system is simple to implement, but the localization accuracy and discriminatory power are limited
Solution Approach 1:
The patent segments the anomaly detection task into multiple specialized neural network modules: a segmentation network for pixel-wise classification, a defect proposal network for candidate identification, and a classification network for defect type determination. This modular segmentation enables high localization accuracy while managing system complexity through functional decomposition.
Solution Approach 2:
The patent introduces an intermediary defect proposal module that bridges the gap between raw image data and final anomaly classification. This intermediate stage generates candidate defect regions that are then refined by subsequent networks, improving overall localization accuracy without requiring a single monolithic complex system.
2Adaptability or versatility
If machine-specific image analysis algorithms are used, then the algorithm can be optimized for a specific printer type, but the algorithm lacks generalizability to other powder bed processes
Solution Approach 1:
The patent implements a universal anomaly detection framework using deep learning networks that can process images from different powder bed additive manufacturing systems regardless of specific printer type or imaging system. The networks are trained on diverse datasets and can generalize across multiple machine types while maintaining high detection accuracy through transfer learning and multi-site training approaches.
3Reliability
If ex-situ inspection methods are used, then the inspection process is simple, but the rejection rate remains high and quality control is insufficient
Solution Approach 1:
The patent implements real-time feedback loops where anomaly detection results are immediately fed back to the additive manufacturing process control system. This enables in-situ quality monitoring and adaptive process adjustment during manufacturing, improving quality control while maintaining productivity through automated closed-loop control rather than post-process inspection.
Solution Approach 2:
The patent performs preliminary anomaly detection and classification during the additive manufacturing process itself, identifying defects before the build is complete. This preliminary action enables early termination of defective builds and prevents waste of resources on obviously flawed parts, improving both quality control and production efficiency.
4Measurement precision
If high-resolution anomaly detection is implemented, then the localization accuracy improves, but the computational time and processing complexity increase
Solution Approach 1:
The patent segments the image processing task into parallel specialized networks that operate simultaneously on different aspects of anomaly detection. This parallel segmentation enables high-resolution pixel-wise analysis without sequential processing bottlenecks, maintaining localization accuracy while reducing overall processing time through concurrent computation.
Data Source
AI summary
Detection and classification of anomalies for powder bed metal additive manufacturing. Anomalies, such as recoater blade impacts, binder deposition issues, spatter generation, and some porosities, are surface-visible at each layer of the building process. A multi-scaled parallel dynamic segmentation convolutional neural network architecture provides additive manufacturing machine and imaging system agnostic pixel-wise semantic segmentation of layer-wise powder bed image data. Learned knowledge is easily transferrable between different additive manufacturing machines. The anomaly detection can be conducted in real-time and provides accurate and generalizable results.