Our Smart View solutions can overcome the limitations of classic vision systems.

They are specific to each customer and application context, minimizing the cost of integration with existing lines and offering broad parameterization flexibility

They adopt advanced Machine Learning techniques, offering resilient performance to environmental changes and the ability to improve over time

They are based on commercial hardware products of recognized reliability, being cost-effective and flexible according to the operating environment

Machine Learning

The system is trained to distinguish between positive and negative situations through a preliminary learning phase, the duration of which depends on the availability of case histories and the complexity of the features to be recognized

Thanks to the learning phase, the system is also able to recognize situations never previously observed, thus adapting to changing operational contexts

The Machine Learning models adopted are directly derived from recognized mathematical models widely adopted by the scientific community

The learning phase can be repeated cyclically to refine system performance or to adapt automatically to changes in the production environment

Multiple recognition models can coexist simultaneously, so the functionality of Smart View stations can be expanded over time

Flexibility and modularity

The control software can be installed on both common Panel PCs and dedicated industrial PCs

The computing resources required are based on the adoption of NVidia Quadro GPU modules or cards.

Ordinary IP cameras with interchangeable optics are sufficient

The system can automatically coordinate complex lighting systems in order to best highlight the features to be recognized

Via industrial communication bus, the SmartView system is configured to interact directly with the line controller to receive trigger events and return inspection outcomes

The User Interface is customized according to the customer’s needs and allows changing working parameters and enabling different features to be recognized.

Conclusions

Introduced solutions based on “machine learning” have been shown to adapt well to a manufacturing environment where materials and defects are characterized by wide variability in shape and position.

The integration of the Smart View system with the existing station did not involve any particular disruption in production, as did the fine-tuning of the system

Recognition performance can be further improved with new training campaigns, without the need for any upgrade of system software or hardware components

The features to be recognized can be expanded with not particularly invasive software upgrades