Powder-Bed Additive Manufacturing Anomaly Detection with Adaptive Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anomaly detection methods in powder-bed-based additive manufacturing are not robust enough to detect defects consistently, particularly due to the need for manual threshold adjustments and the generation of large amounts of unlabeled data, leading to inconsistencies and inefficiencies.
Innovation Solution
A semi-automated anomaly detection method using Principal Component Analysis (PCA) and clustering algorithms to compute adaptive thresholds, allowing for semi-supervised evaluation of image data, which adapts to the recorded dataset during process monitoring.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If manual threshold adjustments are used for anomaly detection, then detection can be performed, but consistency and reliability of defect detection deteriorate
Solution Approach 1:
The system automatically computes adaptive thresholds through PCA and clustering algorithms applied to the image dataset, eliminating the need for manual threshold adjustments. The method self-calibrates by analyzing the statistical properties of the image data and determining optimal thresholds without human intervention, thereby ensuring consistent and reliable anomaly detection across different datasets and imaging conditions.
Solution Approach 2:
The invention transforms static manual thresholds into dynamic adaptive thresholds that automatically adjust based on the characteristics of the input image dataset. By using PCA to identify the most significant variance components and clustering to group similar pixels, the system computes thresholds that adapt to each specific dataset, resolving the contradiction between ease of operation and detection reliability.
2Measurement precision
If supervised machine learning approaches are used, then anomaly detection capability is improved, but data labeling effort and complexity increase
Solution Approach 1:
The invention introduces PCA and clustering algorithms as intermediary steps between raw image data and anomaly detection. Instead of directly applying supervised machine learning that requires extensive labeled data, the system uses unsupervised PCA to reduce dimensionality and clustering to identify natural groupings in the data. This intermediary approach enables accurate anomaly detection without the need for manual data labeling, reducing complexity while maintaining precision.
Solution Approach 2:
The method replaces the mechanical process of manual data labeling with an automated computational approach using PCA and clustering. By substituting human labeling effort with algorithmic processing of image statistics and pixel groupings, the system achieves supervised-level detection accuracy without the associated complexity and time investment of data annotation.
3Productivity
If static thresholds are used for image processing, then processing speed is maintained, but detection robustness to image setting changes deteriorates
Solution Approach 1:
The invention transitions from static thresholds to dynamic adaptive thresholds that automatically adjust to different image settings and datasets. The PCA and clustering algorithms compute thresholds specific to each image dataset, making the detection system robust to variations in imaging conditions, lighting, and camera settings while maintaining processing efficiency through automated computation.
Solution Approach 2:
The system performs preliminary analysis using PCA to identify the most significant variance components in the image data before computing thresholds. This preliminary action of understanding the data structure through clustering and component analysis enables the subsequent threshold computation to be both rapid and robust, resolving the contradiction between processing speed and detection reliability.
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
A Computer-implemented anomaly detection method in powder- bed-based additive manufacture of a workpiece is described. The method comprises, (i), providing an image data set and applying a Principal Component Analysis to said image data to compute a number of image clusters, (ii), applying a clustering algorithm to the analysed image data and computing respective cluster centroids, (iii), comparing the computed cluster centroids with a set of reference anomaly centroids, wherein - based on a match of cluster centroids with the reference, the image data is segmented layerwise into cluster images of a specific anomaly, (iv), transforming the segmented images into a defined color space, such as a Lab color space or greyscale spectrum, and, (v), integrating a pixel information of the transformed segmented cluster images to compute a threshold value for the image data set in order to determine a respective anomaly. Moreover, a related additive manufacturing method, computer program product and a data processing apparatus are described.