Artificial intelligence (AI) remains a major buzz word across many industries, including manufacturing, with the media covering new AI capabilities and trends playing a key role in enabling the digitalization of production. But what’s the real story?
In many cases, AI exists only in theory, and there’s a long way to go before it’s ubiquitous. Third-party service providers touting their often-expensive AI-based technologies can make AI look even more mysterious than it already is. Giving up internal expertise and control to any external contractor can be equally expensive with unclear profits.
Does the potential to leverage AI justify the investment? Is the optimization of a few parameters with AI actually worth it? Isn’t AI just being used to chase a trend?
Becoming flexible with AI
AI should not just be used as a means of optimizing processes that have long been automated. The real potential is to do something completely new with the technology. Tasks that were previously done by humans or physical machines can now be done by AI-controlled software powering robots. This increases flexibility and traceability for the robots, and in many cases, reliability, and enables a more successful operation on the market.
But two hurdles remain:
- A limited number of specialists
- Lack of transparency of the technology itself
Few AI professionals are available; does the industry need them?
As the history of digitalization demonstrates, there has always been a limited number of specialists for new technologies at the start, but this has never stopped progress. Before the triumph of PCs in the 1980s, it would have been easy to believe that every company needed a data center with its own computer scientists to participate in the first wave of digitalization. That isn’t what happened. Instead, there were ready-made products with clearly defined interfaces that enabled every business, no matter how small, to capitalize on IT innovations. The key was PCs: easy-to-understand, flexible computing technology that is now used everywhere.
AI will follow the same path in manufacturing. Instead of paying for external resources to lead an AI project, manufacturers will be able to buy products that provide essential AI functions in such a way that they can be used without outside assistance. That’s one of the basic assumptions under which some component suppliers are developing AI products. You certainly need to concentrate to solve a complex control problem with the product, but you shouldn’t need to be a specialist with a computer science degree to do so.
Building trust in new technology
The second hurdle is the technology itself, which initially seems inscrutable to many. Here, it is important to dispel the widespread concern that AI-controlled robots will suddenly start gross mischief at night, of their own volition. Some claim that it is unpredictable and incomprehensible how AI systems arrive at their decisions. That isn’t true. Neural networks are sequences of multiplications and additions. They are deterministic and their workings can be followed with school mathematics, but they do have a lot of parameters. So you can’t tell at a glance how they make their decisions.
Q&A: Micropsi Industries CEO Ronnie Vuine
There are also calls for AIs to make their decision paths understandable, ideally in comprehensible rules following the if-then-else pattern. If this was possible, there would be no need for a complex model as conventional programming would be enough. AI, however, is the answer to problems for which no solution exists in easily explained if-then-else rules. What is needed instead for building trust in these systems are testable, reliable systems that can be explored simply by using the system and developing a feel for how the AI responds in a given use case. When this testing is quick and painless, the findings —and the AI-powered robots— will be trusted.
Automating manual workstations
Enabling rapid testing is currently a technological challenge for AI vendors. It can sometimes require some patience to train an AI system to be ready for use in production, but it is worth it. Once they get the hang of it, manufacturers can use the AI-based robot control solution to flexibly automate manual workstations. Picking parts, tracking contours, plugging cables, assembling products, it can all be implemented with a single small camera on the robot’s wrist. Since all components can be flexibly trained for new tasks, the robot arm together with the AI software can be used at different points in production.
At an automotive supplier, for example, a simple automation solution has been built for the sorting of metal parts from a semi-ordered grid. Lighting conditions at the facility were unpredictable, often receiving direct sunlight. In addition, metal parts are highly reflective, and the occurrence of flash rust had to be factored in. The supplier approached Micropsi Industries because its AI system could handle these variances – position, lighting conditions, color, and occlusions from leftover packaging. To do this, the technology had to learn to find the next part regardless of time of day, sunlight intensity, surface condition, and packaging coincidences.
More difficult to solve are the testing applications currently going through a validation phase at a manufacturer of white goods. Here, probes must be positioned with great accuracy. The AI has to find solder joints on copper lines that are being tested for leaks and that differ greatly in position, orientation, shape and material properties.
These two applications were implemented with almost identical hardware: a UR5e collaborative robot from Universal Robots, the AI system and a wrist camera, as well as a tool tailored by the customers for the applications. Factory employees trained the systems on site. Source code only needed to be written on the PLC side.
Build AI expertise internally
There are many AI products emerging at the moment for use in manufacturing. They trigger a change in mindset and enable software-controlled and flexible production processes. The resulting complexity is kept in check in easy-to-learn products.
As a result, you can achieve much more with AI than just a little optimization. The technology enables more flexibility, independence, resilience, and efficiency. The market must offer products that can be learned in an exploratory manner and thus allow the AI to be trusted. If this succeeds, a major automation wave comparable to the introduction of PC technology is possible.
About the Author
Ronnie Vuine is the founder and CEO of Microspi Industries. Founded in 2014 in Berlin, Micropsi Industries is a software company that provides ready-to-use AI systems for controlling industrial robots. Microspi has raised about $9 million over six funding rounds and has 35 employees. Its MIRAI controller, which generates robot movements directly and in real-time, is compatible with ABB and Universal Robots.
Prior to founding Micropsi Industries, Vuine was head of software development at txts and was a digital banking consultant for IND Group, spearheading retail banking transformation projects for BNP Paribas Fortis (Brussels) and UBS (Zürich).
Tell Us What You Think!