Until now, most companies have relied on rigid maintenance models. Accordingly, robotics applications are only checked when they are in need of repair, when a specified number of operating hours has been reached or preventively without taking the machine's condition into account (maintenance cycles).

The reason for such rigid maintenance cycles is the lack of usable and high-quality data. Many robots are monitored with just a few sensors, which only guarantee rough data collection. However, a simple rule of thumb applies to robotics: the more data that can be collected through monitoring, the more reliable statements can be made about the future condition of the machine. This applies to everything from grippers and bearings to gears and drives - from pressure sensors and temperature to vibrations, current curves and ultrasound. But even if the sensors collect enough data, how can the sometimes huge amounts of data (which can be collected by ultrasonic sensors, for example) be transmitted?

Where high-resolution sensors are already in use, this problem has so far been solved using edge AI solutions. Here, for example, an AI on the sensor cuts out the data that is deemed relevant and sends only this pre-processed data to the central processing unit or even to the cloud, where it is evaluated by another AI. With the new megatrend of "embedded AI" - which has only been made possible in recent years by the increasing performance of semiconductors - data evaluation can now be transferred directly to the device, enabling deep, comprehensive data evaluation without transferring the data to a server on site. This brings the idea of predictive maintenance and the avoidance of unplanned downtime within reach.

If you look beyond predictive or preventive maintenance for specific areas of application for the technology, there are various optimization options for all different applications. In the case of robots, for example, this could be an analysis of the gripper - if it no longer functions properly, a workpiece could slip and lead to a production stop. This is an old problem that has still not been optimally solved.

Cobots in the SME sector

The market is currently dominated by "large" cobots that can withstand loads of more than 10 kilograms - nothing compared to conventional robot arms. However, experts assume that the use of medium-sized and small cobots will also increase rapidly, especially in SMEs, as increasing automation will expand and improve the technology's application possibilities.

For the cobots, the use of embedded AI offers particular advantages in the area of user interaction, i.e. human-machine interaction. This includes controlling work steps by voice command or gesture. This not only simplifies the collaboration flow, but also offers collaboration in any language and even with disabilities. In addition, for many work steps, it would be possible to automatically recognize when the human work step is complete so that the cobot can carry out the next step. As a result, no communication is necessary. Another advantage is the emergency stop by voice, which can be implemented very robustly using embedded AI, even with high ambient noise levels. Depending on the safety directive, this is also possible with a synthetic feedback loop.

Creating effective and transparent production with automated guided vehicles (AGVs)

Between 2018 and 2020, the number of AGV installations increased by more than 100 percent, from 52,000 to 114,000 [1]. These "little" helpers can optimize the flow of materials and make conventional forklifts and warehouse ants obsolete.

AGVs are currently used primarily in logistics, e.g. for unloading trucks and transporting goods from A to B. This allows logistics, but also industrial production, to be controlled flexibly and transparently. AGV solutions are of interest to all companies that want to successively automate and streamline their processes, which is also linked to their high scalability. Their importance will continue to grow in the coming years, influenced by variant diversity, smaller batch sizes and increasing quality requirements.

In driverless transport systems, load and obstacle detection and maneuvering around people or objects are particularly important - imaging and other perception methods that are made even more robust by AI can be used here. Lidars (also known as time-of-flight sensors) or radars are now so inexpensive that they can either be fused together or work much more three-dimensionally and robustly individually without any lighting requirements. In addition, the new user interaction possibilities with people, as with the cobot, would open up, which would also benefit safety. It is crucial that this technology is adapted on a broad scale.

Increasing production efficiency with embedded AI and smart maintenance

Embedded AI helps to make robots, cobots and AGVs fail-safe. In contrast to the previous cloud and edge solutions, embedded AI collects data on site at the respective application and does not have to send it to the cloud for processing. It enables much deeper data evaluation directly on site in the device, does not have to transfer large amounts of data and thus makes the failure of a component predictable. Unplanned outages are therefore a thing of the past.

Embedded AI is not just limited to maintenance, but it is important to integrate the technology more into general use within production. Especially with regard to the increasing use of cobots and AGVs. If this integration and optimization is missed by innovations in Germany and Europe, it will be even more difficult to keep up with the highly automated production in Asian countries.

[1] See Intralogistics: Three trends for 2022 (source: Torwegge)

Author: Viacheslav Gromov, Managing Director of the German embedded AI provider AITAD