Our View #3 - 2013 Q3

Vision systems may produce valuable 'Big Data'

Written by Jørgen Læssøe

How much data does an average vision system generate? Let's assume that an image is 1MB of data. On a production line it is common to record 2 images every second, generating 2MB/s. This will lead to approximately 170 GB/day, a rather large amount of data that has to be processed. Many vision systems are only used for presence check or merely provide a reject signal which reduces the amount of data significantly. In other cases vision systems are designed for measuring several different dimensions and will based on tolerances decide whether a part is good or bad. But what happens with the measurement data? Usually data are discarded when a new part arrives, but perhaps long term statistics could provide key information about the production.

What can you gain from the 'big data' your vision system produce?

How do you exploit your vision system as an information tool – and why should you? Perhaps certain statistical features can be extracted from the measurement data over time and through an analysis predict when a critical error will occur. Another potential benefit can be supplier control. Are some suppliers doing better than others, based on tracking raw material over time?

Larger companies with many vision systems located geographically at different places may go one step further and collect production statistics from several different sites at a central server and benchmark the sites based on their performance. This may help ensure "best practice" can be copied from one site to another.

Vision systems can also be used as adaptive process tools, where the primary goal is not to reject bad parts, but more to trim and optimize the yield of the production with the aim of becoming more efficient. Systems like the Hot End tableware Inspector from JLI, which helps operators of a glass production to adjust production on the spot. Or systems like the laser print inspection that is able to predict small inaccuracies in the print and compensate them actively during production and keep performance up.

Ask yourself whether you use the 'big data' that your vision systems produce over time?

Are you ready to give it a try - on a larger scale?

Jørgen Læssøe




A vision system automatically predicts and compensates small inaccuracies in the laser print from six lasers.