A new body part comes out of design, and before it can be manufactured, the painstaking work begins: Where do the clamping points for the fixture go? Where are the weld points? Which locations belong in the measurement plan? Up to 23% of an engineering team’s time goes into this kind of preliminary work – points that are effectively re-set by hand for every new part variant, even though the databases often already hold a very similar part from an earlier project for which they were long ago defined. This is exactly where we at inpro come in, with our expertise in Geometric Deep Learning.

Untapped engineering knowledge in the CAD archives

Legacy CAD data is an untapped treasure. A part is redesigned several times over the years – and each time, defining reference and inspection points starts over from scratch, even though a geometrically almost identical predecessor part already exists. Conventional search systems find only exact duplicates or parts with matching metadata, but no geometric similarity across different naming conventions and storage locations.

Our workflow: find a similar part, transfer the points, validate the result

Based on AI and digital twin technologies we developed a three-stage workflow together with Volkswagen and Siemens:

  1. Search – For a new part, we automatically find the most geometrically similar reference part in the existing data – regardless of file name or storage structure.
  2. Transfer – We detect which regions of the reference part and the new part correspond to each other and automatically transfer the points already defined there – clamping, weld, measurement, and inspection points. The engineer only needs to review and fine-tune our proposal.
  3. Validation – We validate the transferred concept numerically, for example through structural simulation, before it is released.

Application in manufacturing planning and quality assurance

We are currently developing this approach for several use cases along the process chain, from manufacturing planning through to quality assurance. Initial pilots on a range of body and part geometries already show that points from a reference part can be reliably transferred to a similar but not identical new part. Further use cases are currently being trialed.

The goal: less manual work on safety-critical routine tasks

Our aim is to relieve engineers of this repetitive but safety-critical task. The necessary hands-on knowledge almost always already exists – it simply sits in a different CAD model. For the first time, we make it systematically findable and reusable, instead of having to reinvent it for every new project. And our approach is not an isolated solution: it can be integrated as an additional capability into existing engineering and PLM environments, where CAD data, release processes, and quality checks already converge anyway – for example in the data management, simulation, and quality systems that engineers already work with today.

More Industrial Engineering innovations

AI-enhanced weldability testing

Testing the weldability of multi-sheet joints is an important technical prerequisite for production. We show how AI shortens this validation step and makes it reproducible.

+

AI-supported analysis of production data

In times of growing variant diversity, increasing quality requirements and limited expert resources, traditional analysis methods reach their limits. AI-supported analysis of production data helps companies spot patterns and clues to root causes.

+

Game changer: computer vision in shopfloor automation

Find out how to scale your AI solutions along the value chain with standardised AI hardware and software components.

+