ADVANCED MATERIALS & PROCESSES | JANUARY/FEBRUARY 2025 31 Most recently, iWP has been training beta versions of deep learning CT analysis software from Hexagon to further improve the accuracy of their data segmentation before subsequent porosity/inclusion analyses. The AI trainer software uses iWP’s pre-segmented models, for which accurate ROIs have been identified, comparing those known models against scans of unknown parts (Fig. 3). “The more datasets that you feed into the AI trainer, the smarter—or more accurate—your model will be,” says Chao Wang, deep learning specialist for iWP. Early training exercises he has conducted involved up to 50 datasets. “You have to invest quite a bit of time and energy in labeling all those datasets so you can use them to train your AI models,” he notes. “But then that saves you considerable time on the production line, enabling you to more quickly and automatically find the correct ROIs and segment them properly. The results of your porosity-inclusion analysis are thus much more telling about quality because it’s been run on accurate segments of real parts.” CONCLUSION Industrial CT scanners and other NDT tools can all be sources of data from which analysis software integrated with deep learning capabilities can deliver actionable, real-world insights across almost any manufacturing method (Fig. 4). Highly reliable image processing with AI can provide a foundation on which companies can build their own customized bridges to quality manufacturing. Statistical process control is the final step in quantifying the insights that deep learning can provide for monitoring a production process, finding correlations between manufacturing environmental variables, and arriving at better ways to make products competitive. ~AM&P For more information: Kai Winter, business enablement manager, NDE, and Patrick Fuchs, product owner AI, Hexagon’s Volume Graphics, Heidelberg, Germany, kai.winter@hexagon.com, www.volumegraphics.com. Fig. 3 — Chao Wang of iWP trains beta versions of deep learning CT analysis software from Hexagon to further improve the accuracy of data segmentation before subsequent porosity/ inclusion analyses. Fig. 4 — Simplified processing enables engineers to load CT data (here, battery anodes), select a segmentation model, calculate, and apply the resulting regions of interest for subsequent analyses—all in a few steps that can be incorporated into automated inspection workflows.
RkJQdWJsaXNoZXIy MTYyMzk3NQ==