Machine Learning: It’s All about the Data 0
Posted on 16, July 2018
in Category Blog
Machine learning is one of the most exciting areas for innovation in industrial automation. Predictive maintenance and process optimization are the two areas with the strongest traction and the highest degree of usefulness, but a fully autonomous factory isn’t impossible.
Machine learning means that machines use the data they receive to improve their results and get better at what they do. But where do they get the data? This is an issue for industrial automation.
To get a handle on the data question, we need to step away from the industrial environment for a minute. Outside the factory, data is readily available in enormous torrents. For example, consider a new food-tracking app, Bitesnap, which describes itself as a photo food log. You snap a picture of your meal on your phone, and Bitesnap figures out what you’ve eaten and calculates nutritional value. The makers worked hard to get their app out for free to as many people as possible. The flood of data coming in from users, for free, allows the planned machine learning and improves the app.
Now think about your factory. Natural streams of data flow from sensors. But in order for predictive maintenance to work, data must include breaks and failures. If your machine has enough breaks and failures to supply sufficient data for machine learning about predictive maintenance, you’re going to replace it.
Nor can you expect many factories to share their data in the crowdsourcing model Bitesnap uses. Most factories aren’t currently collecting much data at all. Those that are will generally be unwilling to share it.
What happens if there isn’t enough relevant data? Google’s ability to recognize human faces was compromised by a lack of ethnic diversity in the photos used by their bots for machine learning facial recognition. MIT was able to create a robot that scored as a psychopath on the Rorschach inkblot test by feeding it “gruesome” images from Reddit.
The results of inadequate data from industrial machines will be less picturesque… but just as problematic.
If you need support for your Rexroth electric industrial motion control systems, you don’t have to rely on machine learning. That’s probably a good thing, because breakdowns of Rexroth components are very rare. They’ll probably never produce enough data for predictive maintenance machine learning. Not in your lifetime, in any case. So call us for human expertise on your Rexroth machinery. We offer phone support, field support, and factory repair and reman for all Rexorth electric industrial motion control devices, new or legacy.