In complex automated processes such as object detection, sorting and quality control, classic, rule-based image processing systems quickly reach their limits – especially when components vary greatly or defects occur in a wide variety of forms. A typical example is quality inspection in the injection molding process: Here, surface defects, air inclusions or burn marks can appear in irregular shapes and forms.
This is precisely where artificial intelligence shows its strength. By using AI classification, even complex inspection tasks can be solved efficiently and precisely. Convolutional neural networks (CNNs) are used to analyse images, recognize features and automatically assign objects to specific classes. The system independently learns to differentiate between characteristic properties – and thus copes with a wide range of variants and types of defects that would be almost impossible to handle using conventional methods.
With the ONNX module of the wenglor uniVision 3 Machine Vision Software, existing AI models can be executed directly on powerful wenglor hardware – such as the B60 series smart cameras or the MVC series machine vision controllers. This allows rule-based and AI-based modules to be flexibly combined, while at the same time making use of established industry interfaces. The result: maximum flexibility with minimum integration effort.
As a comprehensive image processing platform, uniVision 3 provides all the tools for pre-processing – such as filter functions, layer tracking or the definition of regions. These standard modules can be seamlessly linked with AI classification to master demanding applications with ease. For example, once a component has been captured using layer tracking, it can be measured, calibrated or further processed directly – without leaving the usual workflow.
After successful analysis, uniVision 3 enables data exchange via common interfaces such as TCP/IP, PROFINET, EtherNet/IP or EtherCAT. Special interfaces for robotics vision and robot-guided welding are also available for use in robot-assisted applications.
The advantage is obvious: ONNX models created using third-party software can be integrated directly into the uniVision inference and run on wenglor devices without any additional hardware effort. There is no need for time-consuming data preparation – integration is fast, uncomplicated and scalable.
With the ONNX module, uniVision 3 thus offers a powerful and future-proof solution for seamlessly integrating AI into industrial image processing processes – for maximum precision, efficiency and flexibility in factory automation.
Job title