Fraunhofer Institute for Industrial Mathematics ITWM
Fraunhofer Institute for Industrial Mathematics ITWM
Metallbauteile sind bereit für die folgende Verarbeitung
© freepik / Fraunhofer ITWM
Mit »eQuality« entwickeln wir eine digitale Fehlerbibliothek, die Unternehmen aus der Produktion bei der Inspektion von Defekten unterstützt – mit Künstlicher Intelligenz und standardisierten Erfassen der Fehler. Rechts im Bild: Beispiele für verschiedene detektierte Fehler und Defekte auf Metalloberflächen.

With »eQuality«, we are developing a digital defect library that supports production companies in the inspection of defects – with Artificial Intelligence and standardized recording of defects. On the right: Examples of various detected defects on metal surfaces.

© freepik / Fraunhofer ITWM

eQuality – A Digital Defect Library for Visual Surface Inspection

Automatically Generate Defect Models and Integrate Them Into Your Own Simulation Pipeline

In the project, we are developing »eQuality«, a digital defect library that supports companies from the production sector in the inspection with Artificial Intelligence (AI) and in the standardized recording of defects. At the same time, it is intended to enable virtual inspection planning processes that recognize application-relevant defects and generate realistic training data sets for the AI. Borderline cases are also taken into account.

From household items to components for vehicles or in aviation through to medical technology – when manufacturing products, even the smallest defects need to be detected reliably, quickly and automatically. The manufacturing market requires flexible inspection solutions that

  • can be integrated into their digital twin
  • are optimized and verified through simulations
  • can be used again and again

Such automated inspection systems are expensive and take a lot of time and expertise to develop. They usually cannot be used in the same way for all products and are often difficult to transfer to all processes.

Detect Faults and Defects on Components
© Fraunhofer ITWM
Each component has its own typical defects with specific parameters that can be integrated into »eQuality«, the digital defect library.

AI Needs Reliable Data

Use of artificial intelligence for inspection systems is a promising solution which, if implemented correctly, has the capacity to cut down development time and costs, while increasing precision and flexibility. At the same time, we must be very careful to make sure that the inspection system is reliable and has a predictable response in unexpected situations. This comes of particular importance in cases where the inspection is performed on products whose failure can endanger human life (e.g. airplane blisk).

In order to develop a reliable inspection system based on AI, it is necessary to have well specified quality standards and a large number of physical defect samples, used to train the inspection models.

These samples and data are often not available because they are to be avoided in production. If there are standards, they are usually defined with a view to manual inspections carried out by people. Defective samples are also often difficult to obtain, especially when it comes to specialized components. Furthermore, there is no guarantee that all defects are adequately represented in samples.
 

Artificially Generated Image Data Helps to Train AI

In our research focusing on virtual inspection planning, we want to tackle precisely these data-related challenges. Our solution: photorealistic image synthesis to generate training data. In order to generate the corresponding data, defect specifications must be available  what is a defect in the first place and what can occur but is not a defect? We provide such questions and suitable answers or information in the form of fault models.

The Initial Focus Is on Defects on Metal Surfaces

With the »eQuality« library, we are providing an online platform that people from industry and research can use to create and download defects corresponding to their specific products. As a first step, the library will cover a range of defects that are characteristic of metal surfaces such as dents, scratches, cracks, coolant residues, metal chips, dust particles or stains, etc. – and take into account the corresponding parameters from this focus.

Using these applications as an example, an initial production field can be developed for the library and this approach can then be transferred to other production areas in further research.
 

Milestones in the eQuality Project

The eQuality project team has set itself three milestones:

  1. Develop the stochastic geometry models capable of generating the defects with industry relevant parameters
  2. Research how using the developed defect models with synthetic dataset generation can impact the inspection reliability performance
  3. Make the library publicly available through an online defect generation platform
     

Project Funding and Duration

The project »eQuality« is funded by the Ministry of Science and Health of the State of Rhineland-Palatinate and runs until end of 2024.

 

The focus of »eQuality« is initially on defects on metal surfaces.
© Fraunhofer ITWM
The focus of »eQuality« is initially on defects on metal surfaces.