Electricity + Control September 2018

PLANT MAINTENANCE, TEST + MEASUREMENT

Using visual data to its full potential

Willem Beckmann, DataProphet

In a world where data analytics is increasingly woven into the manufacturing space, straightforward and practical discussions on the value of data are essential.

Take Note!

Machine vision systems see at a higher resolution than humans. AI-enabled visual quality control is consistent and always on duty. These systems can maintain traceability of manufactured compo- nents throughout the production process

1

M anufacturers know that their hard- earned data carries hidden value, but they are often at a loss when it comes to extracting all the potential value this resource offers. After serving its primary purpose of keep- ing the wheels of industry turning, process data often ends up as the digital equivalent of surplus fuel oil stored in process historians and databases. In the present article, we explore how manufac- turers can take practical steps towards making the most of their untapped visual data by combining it with proven machine learning technology. As with fuel oil, data can be ‘split’ to produce useful derivatives that are suited to specific analyt- ical applications. Three main types of data can be obtained from manufacturing processes: quality data, traceability data, and control data. Today we will focus on the value that can be obtained by us- ing visual process data for quality and traceability purposes. Do not be fooled; many valuable manufacturing insights are lying dormant in those monotonous images coming off your production line. Maximising quality and traceability in your factory For many manufacturers, visual quality control, also known as visual inspection, is the domain of human quality controllers tasked with determining whether products conform to predetermined qual- ity standards. Human quality control is often effective since controllers tend to use precise instrumentation to obtain quantitative data that inform good decision making. However, there are many factors which influence the observations of quality control per- sonnel. Human quality control is subject to the availability of quality control infrastructure, mental and physical health, work experience, work con- ditions, and the persistent quality of professional relationships.

When combined with external factors such as a weak economy and unfavorable labor legislation, human visual inspection becomes a less attractive solution for manufacturers.

2

3

Recent advances in machine learning techniques are making the automation of visual quality control possible. AI-enabled machine vision systems such as DataProphet’s OMNI Vision are superior to hu- man process control in at least three ways. First, machine vision systems see at a higher resolution than humans since they are equipped with cutting-edge cameras that consistently out- perform the human eye.

Second, these systems do not tire and are not sub- ject to the various factors that could potentially af- fect the actions of their human counterparts. AI-en- abled visual quality control is consistent and always on duty – there are no fatigue-induced errors. Thirdly, AI-enabled systems can maintain trace- ability of manufactured components throughout the production process – a capability that is ex-

28 Electricity + Control

SEPTEMBER 2018