Electricity + Control September 2018

PLANT MAINTENANCE, TEST + MEASUREMENT

Manufacturers know that their hard-earned data carries hidden value, but they are often at a loss when it comes to extracting the potential value of this resource.

tremely difficult to reproduce when relying solely on human operators. Capturing traceability data can be as straightforward as mounting sensor-trig- gered cameras at strategic points on your pro- duction line to record images of part ID and batch number information. At this point, we can safely assume that visual process data is a rich source of manufacturing in- sights and that its real value is revealed when pro- cessed by machine learning models with a prov- en track record in the manufacturing industry. As with all applications of machine learning, machine vision systems need to be trained to become truly useful to the businesses implementing them. Ad- mittedly, this can be a time-consuming process, but recent advances in data science have provided an effective workaround. Manufacturers that choose to invest in machine vision technology are often impatient to see what the technology can do for their businesses. Unfor- tunately, training these systems takes time. AI de- velopers get around this issue by training systems before implementation. The most apparent benefit of training machine vision systems in advance is that far less time is needed to get them off the ground and generating real ROI. A lack of relevant training data is a constant challenge when training effective machine learn- ing models – this is no different for machine vision systems. Often manufacturers do not have suffi- Training time and relevant data are essential

ciently large databases of images to train a system from the ground up. DataProphet is aware of this challenge and trains its machine vision models on large databases of relevant images before imple- menting them. Though sufficient process data is, and always will be, a requirement for effective machine vi- sion solutions, training these systems before de- ployment takes some pressure off manufacturers since systems trained in this way require fewer process images to train to a level where they be- come truly useful to manufacturers. It is also worth mentioning that solutions such as OMNI Vision continuously improve throughout their life cycles. As new process images are gener- ated, quality control operators feed them into the machine vision system; helping it become more effective at detecting defects and tracking compo- nents through the production line.

Electricity + Control

SEPTEMBER 2018

29