FUTURE MANUFACTURING
Source : shutterstock
In glass bottle production , the AI system identifies defects directly at the “ hot end ” before the downstream processes .
AI only gets better with regular training
There is a widespread assumption that learning machines steadily become more and more intelligent . But this is only partially true . Properly trained , Visual Inspection will become increasingly precise . Without training , however , the algorithms will deteriorate over time . This means that the model has to be regularly trained with new data and findings . And there are various modules for this , which can be combined according to the particular client ’ s needs . For example , a human inspects the rejected parts in detail one more time and reports back to the AI whether its decisions were correct or whether it produced pseudo-rejects . These cyclical optimisations , also called machine learning operations ( ML Ops ), continuously improve the machine learning application .
These training operations require tremendous computing power , which is why it makes sense to run them in the cloud . Everything else , however , can be run locally . After all , in fast-paced manufacturing , milliseconds matter – latency times
for image matching via the internet can be fatal . It is therefore advisable to download the trained Visual Inspection model and run it locally . It only has to be regularly transferred to the cloud for further training . For security reasons , local and cloud applications are separated in a way that no bidirectional communication and no communication from the cloud to the local network is possible .
Visual Inspection reduces operating costs by more than 50 percent
AI-based applications are the next logical step for those who already use classic image recognition , but are dissatisfied with the reaction speed for adjustments as well as with the operating costs . The latter are more than halved on average with Visual Inspection . The application also makes sense if a company is planning to introduce automation . And the breadth of possible applications is also worth noting : an image does not necessarily have to be taken with a classic camera for Visual Inspection . Infrared , microscope or computer tomography images can also be pro- cessed . The tool can be easily integrated into existing processes and enables automatic in-line inspection that detects defects during the process rather than afterwards .
Visual Inspection is a comparatively simple application with huge potential . It can be the first step on the path to predictive production control . A machine learning algorithm can monitor and compare hundreds of thousands of variables simultaneously when it receives them . As Visual Inspection can be combined with data from the machine ’ s control system ( PLC ), it is possible to establish a correlation between a component identified as defective and its production parameters . Once this correlation has been established , the reasons for the defect can be found , predicted and quickly remedied – in the sense of predictive quality , the ultimate in quality assurance . l
Anke Roser Head of Marketing & Communications Germany GFT Technologies SE
22