AI-supported Vision Inspection

Unlike programmed image evaluation based on fixed rules, images or image sections are classified in AI applications. With the help of neural networks, an AI system can also assess images that differ from those already known.

The use of AI algorithms is useful, for example, in surface inspection and defect detection, but also in presence checks or in difficult code or text recognition.

To train an AI system, training images that have been previously evaluated are necessary. Different methods are distinguished:

In supervised learning, images are assigned known labels provided by humans.

In unsupervised learning, however, the training images are not labeled. The system learns to recognize structures and anomalies independently and without human guidance.

Technological developments in AI open up many new possibilities. Solutions can be realized that would have been impossible or only feasible with enormous effort using conventional, rule-based programming.

However, at Robotec, we use both technologies depending on the task at hand. The great advantage of conventional, rule-based systems is that they follow fixed processes and clear decision-making foundations, which are always comprehensible. Additionally, conventional systems require less computing power, which can positively impact investment costs.