AI and collaborative robotics are projected to see impressive growth over the next few years. Manufacturers are deploying robots to meet evolving customer needs and fluctuating market demands. Yet, as the digital transformation continues to influence Industry 4.0, we’re beginning to see the convergence of automation and IoT in the form of AIoT: the artificial intelligence of things.
We caught up with Jens Beck, partner heading up Data Management & Innovation at Syntax, to get the details on how AIoT can improve manufacturing systems.
Design News: What type of AI is impacting manufacturing? Machine learning? Training? Quality inspection?
Jens Beck: In general, there are four types of AI being discussed currently. The main two that are impacting manufacturing are reactive machines and limited memory. With a reactive machine, AI reacts to input and produces output, e.g. if the temperature is above a threshold, raise an alert. In practical use, you would refer to this as condition monitoring. This is broadly used in the cases of quality inspection or simply within MES systems. Nevertheless, this information is not stored. With limited memory, the input and output are correlated and stored to allow for predictive use cases or visual inspection. When we talk about anomaly detection to identify outliers, or when we use optical systems for visual inspection, we use limited memory AI.
In the simplest cases, there are simple AI models trained based on the operators’ or quality inspectors’ knowledge and need retraining over time. In this approach, you provide the “machine” with good and bad pictures to teach it what a good outcome is and what a bad outcome is. This is especially important when you want continuous quality checks during high-cycle times in manufacturing, e.g. cathodes for batteries. In a more complex scenario, you would implement competing neural networks that do not only store data, but train themselves on execution.
So, where to use AI in manufacturing? Well, predictive maintenance (i.e. the prediction of when a machine needs maintenance). Predictive quality is another example, which allows machines to predict an outcome and adapt accordingly based on sensory and environmental data.
Visual inspection is a great use case as it can increase product quality, reduce manual efforts for quality inspection, reduce manufacturing time, and therefore increase throughput. But this is not the end of the multiple uses of AI in manufacturing. Augmented reality and natural language processing with chatbots can signal operators when to increase workplace safety.
DN: How is AI being used with collaborative robots? Are the robots communicating with each other? Handing off work to each other?
Jens Beck: Well, collaborative robots are robots that interact with humans and of course, safety is a major concern here. This is why robots on the shop floor are mostly kept behind solid fences and interacting with them always means a production stand-still.
Now imagine you put sensors in the environment of the robot that let it recognize what happens around them. In this scenario, instead of stopping the robot, it could simply slow down its arm or alter its movement to avoid damage to its human colleague. Robots could also adjust to their colleague’s work speed or behavioral pattern to achieve optimal operations.
All of this requires AI in the background. So, it’s fair to say that collaborative robots without AI do not exist. Again, this would not be the type of AI that becomes independent, it still stays human-controlled and gets retrained regularly to ensure maximum safety for co-workers. Of course, a coworker could also be another robot, and in this sense, the same applies, but with the objective of maintenance reduction and OEE optimization.
Last but not least, AI can also be used to train robots. Let’s assume the robot can mirror typical human movements. Then you could simply record a human doing this movement and project it on the robot. This technology is in place and uses AI in the background as well. Of course, this is more applicable in areas where such behavior is desirable, such as healthcare, where humanoid robots are becoming more and more used.
DN: Does AI in manufacturing requires that the manufacturing equipment be “smart” equipment?
Jens Beck: If you intend to optimize the maintenance of a machine, its outcome, or OEE, first you need to gather data from this machine. Then, correlate it with relevant data from other sources like MES, ERP, and historians, to get relevant insights and actions.
So the simple answer would be yes. Nevertheless, the shop floors of this world hold three different generations of machines. The youngsters – are talkative, polyglot and use the latest slang (or protocols in the sense of IoT). This generation comes smart from the shelf.
The middle generation – is talkative, not fully polyglot, and maybe uses some aged slang. For the middle generation, there are translator solutions in place, which make them also “fully” smart.
The last generation is the “granddad”-generation – quiet, not talking much or at all, and not using any slang. For these you can use retrofitting to make them smart, i.e. putting sensors on them to make them talk. In my experience, this works pretty well and apart from exceptions, provides the insights on the “granddad” you need.
So to answer the question, does the machine have to be “smart” – yes, but this does not mean you have to undertake major investments to achieve this objective.
DN: When AI is used with factory equipment, does it matter what vendors created the equipment?
Jens Beck: No, it does not and it should not. Of course, if you bought a brand-new machine that comes with an IoT portal as part of the prize, you would love to leverage all your shop floor machines on this portal. But if all your machines are from the same brand and generation, you inevitably run into obstacles on the shop floor.
However, when you look at agnostic IoT platforms, i.e. IoT platforms which are not created by a machine manufacturer, you will find that these are super open about their input capabilities. Their major differentiation is on the output and the cost side.
DN: Explain the difference between IoT and AIoT.
Jens Beck: IoT is when things communicate with each other, e.g. my alarm clock with my coffee machine; raising the alarm signals the preparation of the coffee machine. AIoT is the world where artificial intelligence helps to make more things talk to each other, i.e., where the output of one thing requires interpretation to form an insight, which then serves as the input into the next thing.
So, when in the coffee example: My alarm clock rings, then a camera in the bathroom mirror takes a picture of my face, it notes I look very tired, so my coffee is prepared with a double-shot espresso vs my usual lungo.