Nobody likes spare parts! After all, everyone wants smoothly functioning products.
Patrick Hey: Yes, exactly – but wear and tear is part of life, whether you like it or not. But how the issue is dealt with is the decisive factor – on both the customer and the manufacturer side. Many of our customers are in the industrial manufacturing sector. They produce machines that make consumer goods elsewhere. Service is a fundamental part of customer loyalty. They don’t just replace heavy machinery like you or I would replace a toaster at home.
How should I visualize this: machine defect, production line at a standstill. What now?
Patrick Hey: Of course, it would be better if the machine’s connection to the IT systems would detect that the machine is about to fail before it does. Keyword predictive maintenance.
Gerhard Heinzerling: Exactly, and that’s where smart algorithms can do an excellent job. The algorithm learns from the data fed to it by the machine sensors. Gradually, it discerns patterns that indicate an imminent failure. These can be specific vibrations, the smallest deviations from the norm, which humans would never notice – because, unlike the sensors, they don’t spend all their time in the machine.
Patrick Hey:Exactly. Specifically, this happens: the algorithm detects indications of an imminent failure and automatically creates a service ticket. Based on this, the service can be deployed in a coordinated manner so that, as far as possible, production is not negatively impacted. Another advantage: all necessary spare parts and the appropriate tools can be scheduled immediately. The gain in efficiency means that these systems quickly pay for themselves.
You guys are also working with AI in the spare parts ordering area?
Patrick Hey: Yes, that’s right. The part number must often be specified when ordering a spare part. But who has them always at hand? There are, however, programs that display a CAD drawing of the machine. You can zoom further and further into this drawing until you find the spare part you need. Gerhard and his team have developed an alternative approach.
Gerhard Heinzerling: For this, we use a specialized image recognition algorithm. We train it with spare parts photos until it reliably recognizes them with a very high probability. This massively simplifies the ordering of a spare part: take a picture of the broken part, upload it, and the spare part is added to the shopping cart.
Patrick Hey: If the algorithm is not quite sure, it asks the user to select the part from a list of possibilities found. By combining artificial intelligence with the user’s expertise, the correct spare part can be identified efficiently and at a reasonable cost. An incorrect spare part is of no use to anyone!
What are the biggest challenges in spare parts recognition with AI?
Gerhard Heinzerling: For one thing, defective parts can look rather battered and scruffy. It is not easy to get these kinds of images to train the algorithm. On the other hand, the recognition of very similar objects such as a 5mm screw and a 6mm screw can be quite tricky – for humans as well. We have to work with reference objects that have to be included in the image. We all know this from crime stories, where a ruler is placed next to an object.
Patrick Hey: And that brings us back to the topic of usability. Having to laboriously position the ruler next to the defective object on the machine doesn’t help me either. That’s why in such cases I like to recommend the combination of the parts list, the CAD drawing (if necessary), and image upload. In this way, the algorithm can check which machine part most closely resembles the photographed part and generate a corresponding suggestion list – possibly with reference to the installation location. The correct part can then be selected from this list. The bottom line is that ordering spare parts just has to be as easy as possible.
There is one thing I would like to mention. I often hear that having two systems in one spare parts store is a lot of work. So my response is always: what is the alternative? Scaring customers away by providing sub-optimal service? Service time is the most important time for customer loyalty. Satisfied customers come back and recommend your company. This is actually the most effective marketing there is. It is much cheaper to keep satisfied customers than to find new ones. And more pleasant for all involved, too. This is actually the most effective marketing there is. It is much cheaper to keep satisfied customers than to find new ones. And more pleasant for all involved, too.
Patrick and Gerhard, thank you for the interview.
