Visual Inspection - Fraunhofer IGD

The oldest form of visual inspection is human observation. With the introduction of industrial image processing, this process has become increasingly automated. However, conventional systems reach their limits when operating under challenging environmental conditions, as they require complex rule-based configurations and the expertise of skilled personnel. Advances in Artificial Intelligence (AI) and Machine Learning (ML) now enable new approaches that make visual inspection more efficient, robust, and reliable—even in demanding environments.

In addition to surface inspection, 3D measurement of components is widely used in quality assurance (QA). It is employed to verify dimensional and geometric accuracy and can be carried out using tactile or non-contact measurement methods. Robot-assisted systems improve repeatability, but their extensive programming or teaching effort is often only economically viable for large quantities of identical parts.

At Fraunhofer IGD, we develop technologies that make optical inspection processes more efficient and robust, creating measurable added value for our customers.

Your Benefits for Quality Assurance

Efficient Quality Assurance

Automated optical quality inspection with high reliability
Achievement of required process speeds and cycle time targets
Reduced inference times
Lower costs and shorter training times through optimized AI/ML models, including model pruning
Reduced effort for collecting training data and images through synthetic training data generation

Flexibility and Autonomy in 3D Scanning and Processing

Robot-based 3D scanning without teaching, CAD models, scan plans, or programming
Complete capture of all visible surfaces
High accuracy (up to 30 micrometers) and excellent color fidelity
Ready-to-use 3D models for quality assurance, 3D printing, visualization, and more—without manual post-processing
Automated post-processing of components after scanning, such as coating removal, paint stripping, and similar operations

Optimization of the Training Process

Training AI models exclusively with "good data" (images of defect-free parts)
Synthetic generation of training data, including from 3D CAD data and 3D models
Generation of arbitrary viewpoints from CAD data, for example for determining a component’s position in space
Availability of training data before physical objects even exist
Simulation of defects such as wear, damage, or incorrect assembly
Minimization of effort required to collect data and images
Acceleration of algorithms and AI models for faster training cycles

To fully realize these benefits in practice, several technological challenges must first be overcome. This is precisely where our solutions come into play.

Products and Applications

Optical Quality Inspection in High-Volume Airbag Igniter Manufacturing

Training of Machine Learning Models Using Only Defect-Free Data

Comparison of scanned parts and nominal CAD model even under deformations

Autonomous 3D Scanning Robotic Arm for High-Resolution 3D Model Generation

Award-Winning Real-Time Wear Monitoring

Robot-Assisted Component Processing

Despite significant advances in AI technologies, challenges remain in both visual inspection and 3D scanning.

Typical Challenges in Automated Quality Assurance – and How We Address Them

Visual inspection has long been a proven method in quality assurance—historically through human inspection and increasingly through computer-based image processing. Traditional systems, however, depend on rigid rules and extensive expertise. Advances in AI and machine learning have opened up new possibilities for making visual inspection more robust, flexible, and automated—even under difficult operating conditions.

Challenges in Applying Machine Learning to Optical Quality Inspection

We address challenges such as:

High effort required to generate training data and train AI models
Strict cycle-time requirements for inference during inline inspection
Quality and robustness of automated visual inspection
The domain gap between the appearance of real manufactured objects and their CAD models
Data drift caused by changing environmental conditions

Challenges in Automated 3D Scanning

The 3D acquisition of components plays a key role in manufacturing and quality control, particularly for verifying dimensional and geometric accuracy. Non-contact measurement systems and robots are often used for this purpose. To date, these systems generally require extensive programming or manual teaching. The goal of fully automated 3D scanning—without manual teaching—introduces additional challenges.

We address issues such as:

Scanning objects for which no prior information is available (no CAD model, no scan plan)
Minimizing user interaction during preparation and execution of 3D scans
Ensuring complete component capture with the desired quality in terms of accuracy, resolution, and color fidelity
Process repeatability and collision avoidance
Handling challenging surfaces such as glossy or reflective materials
Generating watertight 3D models with little or no post-processing effort
Incorporating not only geometry but also all process-control and machining parameters required for robot-assisted component processing

Fraunhofer IGD: Your Reliable Partner

Fraunhofer IGD develops customized systems for optical quality assurance. Our solutions combine: AI-driven image processing, simulation-based data and process preparation, robot-assisted 3D technologies. We simulate, configure, and implement solutions tailored precisely to your requirements.

Would you like to automate your quality assurance processes or discuss a specific project? Get in touch with us—we look forward to the exchange.

Visual Inspection and 3D Scanning Systems for Quality Assurance