Artificial Baby!

By Florian Güldner

Category:
Company and Product News

Already back in autumn I was searching the halls of the SPS/IPC/Drives show in Nuremberg for applications and offerings in automation for artificial intelligence (AI).  The results of my hunt were exciting but also a bit sobering, as the actual haul was really small.  This spring, on the Hanover Fair, it seems to be the next big thing.  In my opinion, we are just seeing the beginning of the real hype, so we will talk a lot about this topic over the next years.  To give you an idea of the speed at which innovations are entering the market in manufacturing: we expect to see the first major offerings with use cases in around 3-5 years and a wide-spread application in around 10 -15 years.  Sobering, I know, but let’s take a look at what was already on show today.  Please note that list is not complete, reflects my personal experience and is in alphabetical order.

Festo acquired AI company Resolto, my colleague covers this in a separate blog.

IBM has a long history in the AI space with its supercomputer Watson.  ARC has already interviewed IBM quite a while ago.  In Hanover, IBM demonstrated a number of applications, among them plant optimization as well as predictive maintenance, called cognitive field services.  The unique ability here is that IBM’s Watson goes beyond predictive maintenance and uses unstructured data from ERP and other maintenance reports to generate work orders for the maintenance staff.  This is especially important in countries and regions with lower educational standards, but may lead to conflicts in countries with highly educated workforce.

Mitsubishi has already presented its MAISART AI technology at the SPS show and was among the early adopters of AI.  We covered it already here.  Mitsubishi benefits from the fact that its AI technology is used across the complete company, from air conditions to mirrorless cars, creating synergies in R&D. 

Omron’s AI module does not only trigger alarms, but is also designed to take direct actions.  The module is designed as an IO and is sitting below the PLC.  In a simple control task, the AI module can basically take over control completely and the PLC only handles the task of communicating.  In this case it could be replaced with a gateway. The concept is called AI equipped machine controller the target is to integrate the AI learning in real time into machine control.  The real-time decision making is the difference to other AI applications. 

Rockwell has released its Project Sherlock.  One of the differences is that Rockwell’s Sherlock is designed as an IO module, where it can optimize smaller tasks. The AI is self-designed by Rockwell and – according to the company – it also understands physics.  That means that the learning should be quicker and the speed higher.  It also is an advantage, when it comes to process automation.  The use case is that once the model is built, it continuously watches the operation looking for anomalies against its derived, principled understanding.  If it spots a problem, it can trigger an alarm on an HMI screen or dashboard. Future iterations will go beyond diagnostics to guide users on how to remedy the issue or to automatically adjust system parameters to fix the problem without human intervention.  Sherlock can also learn from historical data saved in the historian, however there is not the possibility to teach via simulation.

Siemens demonstrated AI with two robot arms assembling a din rail, wherever the din-rail was located and wherever the modules were placed, the robots reacted and adapted.  This is a major application for flexible, yet automated production.  This quick problem solving was until now the typical human domain.  The AI is in this case is sitting in an IPC close to the controller. 

Toyota demonstrated the optimization of AGVs for logistics and warehousing.  The drones, in this case small automated forklifts, optimize autonomously the task to get the boxes as quickly as possible from A to B.  Important is that the environment was simulated first.  The drones learned in a simulated environment, which is far more efficient than letting them bump into each other in real life.  This means you can teach them millions of times, before the first physical box is actually handled.  Also, the drones share their learnings.  When one has learned that you should not hit a pillar, the others know this as well.  Real swarm intelligence. 

 

I know that there we more examples, I was not able to cover all at the Hanover show and we will continue to research this topic.  There are big hurdles ahead.  Most of them are not technical, but concern topics such as the cultural change needed at the OEM side to implement AI, or the challenge of certify a machine or even a COBOT using AI.  The technology is at a very early stage in manufacturing, we will continue to keep you updated.

Engage with ARC Advisory Group