Ever clash with your family members over a slice of pizza because that’s the one that has the most toppings? You know the one where all of the salami and extra cheese have congregated?
For those evenings where dinner has to go fast, at my house we warm up frozen pizzas. Brilliantly, they’re fair down to the gram. So no salami squabbles there. I know that from my experience working with automation in the food and beverage sector. Exciting for me, it involves AI, which is one of my favorite topics.
To achieve a high level of uniformity, manufacturers can use AI to make sure that all pizzas have a similar distribution of toppings. It might sound simple enough, but it’s a bit more complicated. For a solution like this to work, you have an interplay between hardware, software and – something very essential – service. Here’s a breakdown:
1. Hardware: For pizzas, this means cameras, runtime environments in the form of edge computers, and automatized equipment like a pusher to sort out rejects.
2. Software: Algorithms that run on edge computers analyze every image from the camera within a second to decide if a pizza is good or bad. If a bad pizza is spotted, the edge computer triggers a reaction, such as pushing the pizza out of the line or alerting an operator. Some might wonder why the data isn’t sent to the cloud. In this case, at one pizza a second, it would take too much time to send the image to the cloud and back. Edge computing is clearly the better choice here.
3. Service: This is the really interesting point. So say a recipe has changed, or a new pizza is being introduced into the product line. Here we need services focusing on AI in order to adapt or retrain the AI models to the new recipe. Those services also ensure that the AI is working as intended, for example with monitoring predictions, finetuning and the occasional tweak. Like I’ve pointed out before, AI works best when it is actively managed. Best of all, most of this is automatic, much like the system inspecting the pizzas. But service isn’t just important for the big things. Suppose the camera lens gets dirty, or the connection between the camera and the edge computer is interrupted. Someone who knows their way around is invaluable in these situations too.
So you see: Automatically sorting out the imperfect pizza requires a holistic interplay of hardware, software and service. By combining these three elements, we are able to put a value hypothesis into production quickly. Of course, at Siemens we do more than pizza: When using AI to sort out operational issues, we consider the end-to-end requirements and focus on the big picture. Only then can AI contribute to productivity gains.