In-process monitoring techniques provide critical insights into components manufactured using laser powder bed fusion. However, one of the most significant challenges is handling the immense volume of data generated during a print, particularly for large components and high-volume production runs. While current state-of-the-art methods in academia are focused on the scan track level or for small parts, industrial applications require scalability. This work presents a new approach to analysing on-axis photodiode data by transforming it into a voxelised spatial structure. It uses statistical reduction functions to consolidate multiple sensor readings into a representative value, exposing characteristic thermal and melt pool stability behaviour. This reduces complexity while preserving key patterns. We demonstrate the framework’s effectiveness with two case studies: first, we validate the approach using a benchmark dataset from an overheating test part. Second, we investigate the correlation between photodiode data and spatter-induced porosity showing how the statistical features of the dataset can be used to highlight the location of defects. We found that using the mean of the photodiode response had a moderate negative correlation with porosity, up to 0.6, and the standard deviation had a moderate positive correlation, up to 0.45. Finally, we show how this correlation is improved by combining multiple statistical features into a single indicator, improving the correlation strength to up to 0.68.
In-process monitoring techniques provide critical insights into components manufactured using laser powder bed fusion. However, one of the most significant challenges is handling the immense volume of data generated during a print, particularly for large components and high-volume production runs. While current state-of-the-art methods in academia are focused on the scan track level or for small parts, industrial applications require scalability. This work presents a new approach to analysing on-axis photodiode data by transforming it into a voxelised spatial structure. It uses statistical reduction functions to consolidate multiple sensor readings into a representative value, exposing characteristic thermal and melt pool stability behaviour. This reduces complexity while preserving key patterns. We demonstrate the framework’s effectiveness with two case studies: first, we validate the approach using a benchmark dataset from an overheating test part. Second, we investigate the correlation between photodiode data and spatter-induced porosity showing how the statistical features of the dataset can be used to highlight the location of defects. We found that using the mean of the photodiode response had a moderate negative correlation with porosity, up to 0.6, and the standard deviation had a moderate positive correlation, up to 0.45. Finally, we show how this correlation is improved by combining multiple statistical features into a single indicator, improving the correlation strength to up to 0.68. Read More
Related posts:
Dynamic model to characterise sectors using machine learning techniques
Analysis of Dam Behavior by Statistical Models: Application of the Random Forest Approach
Fusing CNNs and statistical indicators to improve image classification
Thermal imaging from UAS for estimating crop water status in a Merlot vineyard in semi-arid conditions
Android malware detection through hybrid features fusion and ensemble classifiers: The AndroPyTool framework and the OmniDroid dataset
Carbonic: a framework for creating and visualizing complex compound graphs