Industry Applications

Advanced Featured Identification (AFid)

Advanced Feature IDentification (AFid) is a software plugin (SDK) produced by Fluid Dynamic Sciences LLC available for commercial use by third parties. AFid was originally developed as part of an automated mesh generation toolset to enhance the use of mesh spacing rules, and has a range of applicability across a wide range of sectors.

AFid uses machine vision technology, originally developed for self-driving cars, with modifications to enable easy use in industrial applications. Unlike other machine learning algorithms, AFid does not require extensive training, and only needs a single example of a feature in order to recognise it automatically in much larger datasets.

AFid can be integrated within existing software frameworks to provide feature recognition capability based on digital data sources (CAD) and / or in-service scanned data (LIDAR, MRI, CT). AFid runs quickly and efficiently on desktop computing equipment and is designed for easy portability. AFid technology is patented in the US and worldwide.

Fluid Dynamic Sciences is leveraging AI/neural network technology for rapid and accurate feature recognition in digital (CAD) and real-world scanned (LIDAR, MRI, CT) data sources, and CFD analysis accelerated by massively parallel GPU acceleration.

Our patented AI / Machine Vision technology requires no iterative training, providing immediate feature recognition on lightweight (edge) computing resources.

We are in the process of enhancing this technology with a hybrid architecture that augments machine vision with accelerated RCNN neural networks to improve performance and customization, either incrementally or via offline training.
This approach combines dynamically improving responses based on learned experience with low compute, zero-training machine vision.

Zero Training AI

AFid uses a breakthrough zero-training AI approach—also known as single-example learning—to automatically recognise complex geometric features in CAD models, point clouds, and other 3D datasets.

Unlike conventional AI systems, AFid requires no costly neural-network training or large labelled datasets. Instead, it learns from a single example and immediately generalises that feature across an entire model with exceptional accuracy.

This allows engineering teams to identify deviations, design inconsistencies, and critical structures faster and more reliably, all while avoiding the time, expense, and expertise typically required to build and train AI models. By integrating seamlessly into existing workflows, AFid delivers a powerful combination of speed, precision, and automation that elevates quality assurance and accelerates development.

OpenFOAM* Mesh Spacing Tool

OpenFOAM is leading software for computational fluid dynamics (CFD), licensed free and open source, with a user base spanning industry, government research and academia worldwide. We demonstrate here a GUI-based tool using AFid, for quickly and easily creating and replicating mesh spacing rules (for use in snappyHexMesh) based on matching ...

Target Vehicle Identification in Video / EO Sensor Data

AFid can find geometric library items in 3D point clouds constructed from raw video or EO sensor data. In this example, a 3D point cloud has been created using the SIFT algorithm. The VW vehicle represented by open source CAD geometry has been correctly found and oriented within the large ...

Integration with fTetWild Mesh Generator

fTetWild is a popular open-source mesh generator. AFid works well with its naturally tessellated geometry representation. In this example, the junction region on the fTetWild “impeller” test case is automatically identified and sourced for higher resolution. The screenshot shows the demo application that ships with the AFid software development kit ...

DrivAer Automotive

AFid can be used for application-specific rule application. In this example, the popular DrivAer automotive test case for computational fluid dynamics (CFD) benchmarking has particular features that require special treatment. The wing mirrors produce separated flow and so refinement regions are required in their wake. The wheels are in rotational ...

Aerospace Computational Engineering – Boeing 737

AFid is very useful for computational engineering processes where rules (such as mesh spacing) are associated to specific geometric shapes, even if they are not tagged or labelled in a model tree. In this example, over 700 mesh spacing rules (the colourful volumes of interest) are used repeatedly in a ...

Medical – MRI Human Skeleton

AFid can distinguish between features that are very similar, for example in medical scans (MRI, CT) or CAD. In this example, the MRI data from an entire human skeleton has been converted to a point cloud and a specific vertebra used as the target feature. AFid rapidly identifies the target, ...

ANSYS Meshing Integration

AFid can be easily integrated within third-party applications because of its modular nature and the convenient Python and C/C++ interfaces. In this example, a mesh generation workflow has been set up for a specific set of labels. New geometry is supplied with a different topology and geometric origin, requiring the ...

Printed Circuit Board

AFid uses a highly accurate local registration last step in feature recognition, quickly identifying non-conformity between similar objects. For supply-chain and production line checks, AFid can highlight (here in red) the points in the original printed circuit design that are either not present or changed in the new model. This ...

Industrial Factory (CAD or LIDAR data)

AFid can work with point clouds from real-world scanned LIDAR or digital 3D CAD to (i) Compare as-commissioned infrastructure to as-designed models, (ii) Quantify in-service changes from as-designed and through the lifecycle, (iii) Count the number of occurrence of specific parts or sub-assemblies, (iv) Determine changes in the orientation of ...