Electricity + Control September 2018

PLANT MAINTENANCE, TEST + MEASUREMENT

Generate real ROI with your visual data Visual data is a rich source of manufacturing in- sights that, if properly exploited, can significantly improve quality assurance and traceability in your production facility. Capturing this type of data need not be complicated, nor expensive, and is worth your while as a manufacturer. You can make the most of your visual data by feeding it into a proven industrial machine vision system such as OMNI Vision. Implementation time for these systems can be reduced significantly by training them on relevant images before implementation; however, a minimum amount of appropriate images will al- ways be necessary for effective training after the system has been installed.

Though an effective solution to reducing imple- mentation time, training machine vision systems before implementing them is only effective if two basic principles are followed. First, machine vision systems should be trained on images relevant to the task at hand. For exam- ple, a system trained to recognise animals in ama- teur photographs would be less valuable for visual quality control applications than one trained to de- tect paint defects in automotive assembly lines. Secondly, for every machine vision model, there exists a critical mass of images necessary to train the model to a point where it becomes useful. Training a model before deployment greatly reduces the amount of data necessary to effec-

tively train that model when implemented, but it does not, and never will eliminate the need for relevant training data in new domains.

AI-enabled machine vision systems do not rely on template matching to detect defects, but rather rely on powerful machine learning algorithms that allow for the consistent location and classification of defects – even on previously unseen compo- nents. This enhanced capability makes AI-enabled systems superior to other visual quality control solutions on the market. Your visual data can contribute significant value to your manufacturing processes; however, it can only achieve its full potential when combined with the appropriate data processing infrastructure and hardware.

Willem Beckmann is a Mar- keting Officer at DataProphet Machine Learning Specialists.

30 Electricity + Control

SEPTEMBER 2018