VGTRAINER enables manufacturers to train deep-learning models for segmentation in-house on their data.
Volume Graphics GmbH (VG), owned by Swedish company Hexagon, has announced the release of version 2025.1 of its suite of CT analysis software, which includes a new VGTRAINER application. Designed for use by non-AI specialists across R&D and production, the application helps manufacturers apply automation and deep learning to support product design, development, and inspection.
Alongside the 2025.1 release, the Volume Graphics brand will officially merge with the Hexagon brand, following a five-year transition process. Operating under Hexagon’s VG software, the logo and product icons will be refreshed while the well-known product names remain: VGSTUDIO MAX, VGSTUDIO, VGMETROLOGY, VGinLINE, and myVGL. As the de facto standard for non-destructive evaluation of NDE and CT-data analysis, VG software provides clear, reliable insights across all data sources, from CT scans to optical and tactile measurements.
Future VG software releases will add data processing support for other NDT-source data to serve customers across Hexagon Manufacturing Intelligence division. This will help provide more comprehensive digital tools for quality inspection and product development.
Hexagon’s VGSTUDIO MAX 2025.1 supports the initial step in training a model. It allows users to label and prepare training datasets by setting up segmentation scenarios and marking only the necessary inclusions or regions of interest. From there, VGTRAINER offers:
AI-assisted segmentation models—Using a number (exact amount depending on application/data set complexity) of pre-segmented and labeled training data sets, VGTRAINER produces a machine-learning model that can be imported into VGSTUDIO MAX for subsequent use in accurately segmenting complex or noisy data sets quickly and accurately. This is ideal for cases in which there are hundreds of parts to inspect, or in- and at-line quality assurance scenarios.
Non-expert model training—VGTRAINER is easy enough to use that once a set of labeled data has been prepared by experts, other customer-based engineers can then simply import the training data into VGTRAINER and generate a model automatically, without in-depth AI or machine learning experience. This increases accuracy over time and continuously improves quality inspection models.
Speed in R&D and in-line inspection—one customer, ELEMCA—an independent laboratory in France servicing manufacturing customers across multiple industries—recently saw marked time savings when putting this new functionality to work. They fed 8 ideally segmented datasets of four materials – carbon, aluminum, glue, and porosity – into Hexagon’s VGTRAINER to “learn” the skills needed to generate a segmentation model. The manual segmentation, particularly of carbon vs glue, is time consuming as the gray values of both components are very similar. With just 10 datasets, VGTRAINER was able to create a machine-learning model that accurately segments these materials into two separate ROIs, allowing for the subsequent standard porosity detection within VGSTUDIO MAX. This full workflow (VGTRAINER model creation—AI-segmentation in VGSM + standard porosity detection) enabled ELEMCA to reduce time needed from 1 hour to just 10 minutes.
In addition to the new VGTRAINER application, version 2025.1 provides users with a host of other enhanced features that include automatic beam hardening, object-specific views, and 3D reporting.
For more information, visit hexagon.com/mi.
