Powder Bed Melting Quality Assessment Using Multi-Layer Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining component quality in powder bed-based melting, such as additive laser powder bed fusion (LPBF), often result in false alarms and unnecessary process interruptions due to the identification of non-critical anomalies, leading to extended build times and inefficient production.
Innovation Solution
A method that uses a data set and a machine learning-based modeling model to assess the quality of components produced by powder-bed melting, where the data set summarizes information from multiple successive layers, and the modeling model predicts anomalies based on layer data from measurements or simulations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If anomaly detection is performed based on individual layer data using threshold methods, then critical regions can be identified, but too many false alarms occur leading to unnecessary process interruptions
Solution Approach 1:
The patent combines data from multiple successive layers to assess anomaly criticality. Instead of evaluating each layer independently, the system aggregates layer data to determine whether anomalies persist across multiple layers, thereby reducing false alarms caused by transient variations in single-layer measurements and improving detection reliability without unnecessary process interruptions
Solution Approach 2:
The system implements a feedback mechanism where anomaly detection results from previous layers inform the assessment of current layer data. By continuously monitoring and comparing anomaly patterns across successive layers, the system can distinguish between isolated false positives and genuine critical anomalies, thereby reducing false alarms while maintaining high detection accuracy
2Extent of automation
If machine learning techniques are used for anomaly detection, then automated feature detection and pattern recognition improve, but large amounts of training data are required
Solution Approach 1:
The patent prepares and structures training data in advance by collecting layer data from multiple successive layers and pre-processing it into the required format. This preliminary preparation of multi-layer datasets enables the machine learning model to be trained efficiently with consolidated data, reducing the overall volume of training data needed while maintaining high automation capability for anomaly detection
Solution Approach 2:
The system transitions from analyzing single-layer data to analyzing multi-layer data by adding the temporal dimension of successive layers. This dimensional change allows the machine learning model to learn patterns across layers, improving automated feature detection while requiring less training data per individual layer since the multi-layer context provides additional information
3Reliability
If size threshold filtering is applied to reduce false positives, then small non-critical regions are filtered out, but some critical small anomalies may be missed
Solution Approach 1:
The patent merges data from multiple successive layers to assess the criticality of small anomalies. By combining information across layers, the system can distinguish between small non-critical regions that appear intermittently and small critical anomalies that persist across multiple layers, thereby reducing false positives without missing critical small defects through multi-layer correlation analysis
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
The invention relates to a method for determining a component quality of a component (10) to be produced by powder bed-based melting, which component is formed from several successive layers (12).