Automated dye penetrant inspection with AI.
An AI system automates dye penetrant inspection, reliably detects defects, reduces false indications and supports industrial quality.
At a glance
- Project:KI-FEP Automated dye penetrant inspection with AI
- Beneficiary:Vivaldi Digital Solutions GmbH, Harsum
- Project duration:1 September 2026 to 31 August 2029
- Funding programme:Lower Saxony innovation funding programme for research and development in companies
- Funding bodies:European Regional Development Fund (ERDF) and State of Lower Saxony
With the support of the European Union and the State of Lower Saxony.
The project is funded by the European Regional Development Fund (ERDF) and the State of Lower Saxony.
A critical inspection method that is still largely performed by hand.
Dye penetrant inspection (DPI) makes surface defects such as cracks or pores visible. A penetrant that fluoresces under UV light is applied to the component. It collects in defects that would otherwise be invisible to the eye, for example micro-cracks, and highlights them.
The challenge: For complex components and diverse damage patterns, inspection in industrial practice is usually still carried out entirely by hand. Distinguishing between false indications, for example caused by cracks in the tool, and genuine component cracks is particularly demanding. It requires a great deal of experience and knowledge of the component's three-dimensional geometry.
Manual inspection is time-consuming, ties up trained personnel and produces inconsistent results, because the assessment depends on the individual inspector. In addition, it generates hardly any structured data that could be used for process optimisation or predictive maintenance.
What this means in practice
- Visual inspection is a potential bottleneck in the value chain.
- False indications can lead to unnecessary scrap.
- Skilled inspection staff are hard to replace.
- Results are only partially reproducible and difficult to analyse.
Automating dye penetrant inspection.
In the project “KI-FEP Automated dye penetrant inspection with AI”, Vivaldi Digital Solutions GmbH and its partners are developing an AI-based system that performs DPI automatically. It is intended to reliably detect and classify defects, distinguishing genuine flaws from false indications.
Reliably detect defects
Detection and classification of near-surface defects, objective and standardised rather than subjective.
Reduce false indications
Genuine cracks are to be separated from false indications and non-critical surface features.
Support industrial quality
A prioritised assessment directs inspection staff specifically to the relevant findings.
Measuring defects and making data usable
In addition, the system is intended to measure defect geometry quantitatively, i.e. the length, width and extent of a defect, and compare it with conformity criteria specific to each area of the component. The resulting structured data can be used for condition-based maintenance and for optimising manufacturing and repair processes.
Combining image data, AI and 3D knowledge of the component.
The approach is based on combining images captured in daylight and UV light with data-driven image analysis and linking them to the component's 3D model. This allows the system to map image information spatially and better identify false indications based on the component geometry.
Modular and transferable: The system is built in a modular way so that it can be flexibly integrated into different industrial applications. A generic pipeline is intended to make it easier to adapt to other component types and fields of application.
Potential fields of application for the system: foundry, automotive, aerospace, mechanical engineering, energy technology.
Technologies in the project
- Automated image capture and analysis under normal and UV light
- Deep learning methods, including convolutional neural networks as well as patching and meta-learning approaches
- 3D image projection with pose estimation for the spatial mapping of defects
- Synthetically generated training data to reduce the effort required for real image data
- Consideration of process parameters such as ambient temperature and light intensity
From image capture to conformity assessment.
The project is divided into six technical work packages, which are shared among the consortium.
Modular data acquisition systems
Test rigs with camera and lighting units for daylight and UV-A capture image data as well as process parameters such as temperature and ambient light.
Annotation & dataset generation
Tools for efficient, semi-automated labelling of defects provide quality-assured datasets for training and validation.
Data augmentation & data preparation
Synthetic image data from physically based rendering complements real images and makes the models more robust.
Damage segmentation & classification
AI models find defects in the images, delineate them and classify them, in particular as a genuine crack or a false indication.
Pose estimation & 3D image data projection
The position of the component is determined from the images, so that pixels can be mapped to the 3D model and findings can be located spatially.
3D conformity assessment
The length, width and extent of defects are measured and compared with component-specific criteria. The results are documented.
A system demonstrator for industrial practice
Inspection cell with AI assessment: The central result is a system demonstrator: a manually loaded inspection cell in which components are illuminated with normal and UV light and captured by several cameras. The AI developed in the project assesses the images, and the results, including detailed detection information, are displayed on a screen. The demonstrator is being tested under real conditions together with associated industrial partners and serves to demonstrate the system's industrial benefit in practice.
For quality assurance
- Greater reproducibility of inspection results
- Fewer incorrect assessments and less scrap
- Shorter inspection times and better cost-effectiveness of DPI for complex components
- More transparency and end-to-end digital traceability
For employees and the environment
- Relief for skilled personnel, for example from work in darkened UV booths
- Less contact with chemical testing agents
- Fewer physically demanding tasks thanks to automation and assistance systems
- Conservation of resources through less scrap and fewer repeat inspections
This describes the objectives and expected effects of the project, not results already achieved.
A collaborative project between industry and research.
The project is being carried out as a collaborative project between two companies and one research institution. Vivaldi Digital Solutions GmbH is the consortium leader.
Vivaldi Digital Solutions GmbH
Integration of the hardware into existing inspection processes, development of platforms for traceability and structured data.
Sightwise GmbH
Generation and use of synthetic data, pose estimation and 3D image data projection.
Leibniz Universität Hannover, IMR
Development of AI algorithms and methods for 3D conformity assessment at the Institute of Measurement and Automatic Control.
The project is supported by associated industrial partners who contribute as subject-matter experts, pilot users and potential future customers.
Let's discuss your process.
Tell us about your current challenge. Together, we will work out which solution approach fits your quality goal.