Adaptive Weighted Uncertainty Sampling for Additive Manufacturing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In additive manufacturing processes like directed-energy-deposition (DED) and powder bed fusion (PBF), sub-optimal process parameters and camera settings lead to occlusion of melt-pool features, making it difficult to use images for further feature extraction and defect prediction, and existing active learning methods like uncertainty sampling can select redundant instances, increasing computational complexity.
Innovation Solution
Adaptive Weighted Uncertainty Sampling (AWUS) method that balances exploration and exploitation by adjusting sampling probabilities based on model changes, converging to equal sampling for large changes and focusing on uncertain instances when models are similar, reducing the need for manual annotation and improving model performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If uncertainty sampling is used to select instances for annotation, then the model performance gain is improved, but redundant instances or outliers are selected which do not add to model performance
Solution Approach 1:
The patent implements feedback by comparing model predictions across multiple iterations and using the change in predictions to guide instance selection. The system monitors how model outputs evolve and uses this feedback to identify instances that provide meaningful information gain, avoiding redundant selections.
Solution Approach 2:
The patent changes the selection criterion from static uncertainty-based sampling to dynamic sampling that considers the change in model predictions across iterations. By monitoring parameter changes in model outputs and using these changes to weight instance selection, the system avoids selecting redundant instances while maintaining performance gains.
2Quantity of substance
If manual annotation is performed on high-dimensional imaging data from additive manufacturing processes, then the quality and size of annotated training dataset is improved, but the annotation workload becomes labor-intensive and time-consuming
Solution Approach 1:
The patent implements self-service by enabling the system to automatically identify and select the most informative instances for annotation based on model prediction changes. This self-directed instance selection reduces reliance on random sampling and minimizes the total number of instances requiring manual annotation, thereby reducing annotation time while maintaining dataset quality.
Solution Approach 2:
The patent replaces the mechanical process of random or uniform manual annotation with an intelligent selection mechanism that uses model prediction analysis. This substitution automates the identification of valuable training instances, reducing the labor-intensive nature of manual annotation while improving the efficiency of data collection.
3Ease of manufacture
If sub-optimal process parameters and camera settings are used in additive manufacturing processes, then the imaging-based process monitoring is simpler, but the melt-pool features are occluded and cannot be used for feature extraction and defect prediction
Solution Approach 1:
The patent uses feedback from model prediction changes to identify instances where melt-pool features are properly visible and informative. By monitoring which instances provide meaningful prediction changes, the system can selectively use only those images with adequate feature visibility, compensating for sub-optimal imaging conditions through intelligent instance selection.
Data Source
AI summary
A method and system of active learning that includes receiving a set of data instances, passing the set of data instances through an adaptive weighted uncertainty sampling methodology to select a set of unlabeled data instances and the determining if any of the set of unlabeled data instances need to be further processed. The AWUS methodology assigns a weighting to each of the selected unlabeled data instances whereby the weighting may be used to determine which of the set of unlabeled data instances should be further processed.