"Factory production is undergoing a transformation: Artificial intelligence is inspecting, sorting, documenting, and learning. Companies like Trumpf and Bosch want to revolutionize industrial manufacturing with smart technology.
The seat rail in a car, a not very spectacular metal part, but one that fulfills an eminently important function: It locks the car seat in place, holds it securely, and can save the driver's life in the event of an accident. For this reason, manufacturers of such rails must be certain that all welds and welding points are flawless – and they need machines that guarantee precisely that. Or at least machines that can detect that the third welding point on the back left is defective and that the workpiece needs to be rejected.
These are the kinds of questions that preoccupy Richard Bannmüller. He is the Chief Technology Officer of the laser division at the technology company Trumpf and wants to optimize processes, minimize scrap, and prevent production errors as much as possible, especially for industries like the automotive sector with its low margins. And the mechanical engineer is increasingly turning to artificial intelligence (AI) to achieve this.
Until now, such inspection processes have been complex and inefficient. Components regularly have to be removed from production, and welds cut open and inspected to ensure the strength of the parts. "But if I now use AI, I can not only document every component, but I can also have all the data in the process evaluated and analyzed simultaneously, so that I can identify not only defective parts but also the reason why a process is going wrong," explains Bannmüller. The engineer repeatedly uses the term "going wrong" to describe what shouldn't happen: that a work process is flawed and produces defective workpieces.
One example of such an inspection would be a welding process that is simultaneously monitored by a camera and a laser beam – not the welding beam itself, but one that scans the depth of the weld in the molten phase of the metal." "When I review all the data, I find not only the defective parts but also the reason for the defects, because the AI checks all the data and can thus determine which parameters changed at the time of the error," explains Bannmüller. "This allows me to achieve a level of quality and speed that is impossible with manual inspections."
Complex computing operations
However, such systems reach their limits when they have to process thousands upon thousands of data points in the form of images in a very short time – for example, in cases where workpieces need to be constantly and quickly compared with technical drawings. The reason: Normal AI applications perform their computing operations in the cloud, meaning the data has to be sent back and forth, which takes time and leads to latency, slowing down production. In addition, AI chips require a lot of energy and therefore need complex cooling, which is not always easily implemented in machine tools.
Two problems that a startup from America aims to solve: Sima is developing computer chips that bring artificial intelligence to small end devices, so that the necessary computing operations no longer take place in the cloud and in remote data centers, but directly on-site in sensors of cars, lasers of cutting machines, and cameras of traffic monitoring systems. "AI will migrate into the physical world – into the things that connect the network of many data centers with the physical world: cars, robots, machines," explains Harald Kröger, head of Sima Germany.
The chip specialist worked for many years at Bosch and Mercedes and dealt with precisely the problems that Trumpf is now facing in the development of IT systems for cars.
When Kröger talks about the advantages of the Sima concept, he mentions other reasons in addition to the reduced latency: Since data and images are no longer stored and processed in other locations, data protection is simplified. And finally, costs and energy consumption are reduced because data lines and data centers are no longer needed. "The chip for this needs to be super fast and super energy-efficient because I have to integrate it into small devices," says Kröger.
For Trumpf laser specialist Bannmüller, the Sima chips solve several problems at once: "With them, we can process several thousand images per second, and the chip can be used in the machine without a cloud connection and integrated into the control system," the engineer says. "On the other hand, Sima has an architecture that requires very little energy and processing power – and therefore no cooling."
Trumpf plans to use the Sima technology in its machine tools in the first quarter of 2026. The machines could then be used, for example, by automotive suppliers when the contours of sheet metal and punched holes need to be checked in a short time.
AI becomes a unique selling proposition
The more complex the process, the faster such AI solutions will prevail. For Trumpf, this is already a unique selling proposition and a competitive advantage. "There are enough applications that require more or less individual settings for each component. And that's where AI has an incredible advantage, because it learns in a short time what the optimal parameters are and how a machine should be set up," explains Bannmüller.
This alone helps with cutting sheet metal. Manual adjustment often requires several test cuts to find the best parameter configuration. "The AI-optimized machine knows the material, thickness, laser beam, and speed and can then independently optimize the parameters after seeing the first cut pattern," explains Bannmüller. "The goal is to find the best set of parameters extremely quickly and efficiently."
Sima Germany CEO Kröger goes a step further. From his perspective, AI applications on the laser, drill, press, or adhesive nozzle enable immediate corrections during the production process. "So I'm not only delivering perfect products, but I'm no longer producing any defective ones at all, because I'm making adjustments in real time," explains Kröger. "This enables incredible leapfrogging effects: A company that perhaps doesn't produce the best gears can overtake a competitor with decades of experience overnight."
In theory, Richard Bannmüller would agree here, but the regulations in many industries still limit this optimization. "Automatic process changes – processes are usually approved by the customers – are not yet accepted or permitted in many areas," he explains. "The permanent adjustment of processes in laser cutting, welding, etc., is currently a gray area and not the norm."
Controlling Complex Systems
Just ten kilometers from Trumpf's headquarters in Ditzingen, another Baden-Württemberg-based technology company is working on optimizing industrial manufacturing processes with the help of AI. Bosch's industrial technology division has developed an AI agent designed to enable even unskilled workers and temporary staff to restart complex manufacturing systems in the event of a problem, thus reducing costly downtime. In this case, AI agents are software-based systems that combine data acquisition and analysis with algorithm-based decision-making. They are intended to take over routine tasks, detect malfunctions, and suggest solutions.
Philipp Glaser, the Bosch project manager responsible for agent-based AI, explains it as follows: "When a machine breaks down, quick action is necessary. The employee then describes the situation and explains the error pattern via chat or verbally," says Glaser. "The AI agent then asks a few questions, such as whether the machine restarts or whether certain lights are illuminated, and then tries to identify the problem and suggest initial solutions." Bosch's goal is for "people and AI agents to work hand in hand in production."
The suggested solutions are typically initial, immediately implementable instructions that would otherwise require experienced experts to come to the machine, which, depending on availability, repeatedly leads to longer downtimes. "We are trying to reduce the need for expensive service technicians to complex problems by empowering the employee at the machine to find solutions themselves for simple malfunctions," explains Glaser.
The AI agent also handles the documentation of the case: It describes the problem and the solution found, so that this knowledge is available in the future if the problem recurs. This documentation also facilitates shift handover, because normally there isn't enough time to explain the errors that occurred shortly before the end of the shift and the procedures followed so far to the incoming workers. "The AI agent is operated via keyboard and screen – it's an interface similar to the ones we know from ChatGPT and other AI systems," says Glaser. "In the background, the agent then accesses the data from the manufacturing IT system."
Bosch is already using agents itself
According to AI expert Glaser, companies that produce individual components such as ball bearings, valves, sensors, or control units in their production lines will benefit enormously from such agents. Bosch itself is already using these shop floor AI agents in factories where the company produces components for the automotive industry, for example in Naganathapura, India, in Miskolc, Hungary, and in Bamberg. "Within Bosch, there is already a great demand for the agents because the pilot plants are already seeing the added value they offer," explains Glaser.
The Bosch engineer doesn't want to quantify the potential productivity gains for companies through AI agents. However, he says it's clear that "real progress is made when we form agent teams, where all the shop floor agents in a factory communicate with each other and then also include agents responsible for maintenance, purchasing, and shift planning." One example could be that a shop floor agent recognizes a problem, identifies the solution, and immediately reports the defective part to the maintenance agent, who then organizes the replacement. The maintenance agent, in turn, informs purchasing that the spare parts have run out, while the shift planning agent readjusts the shifts because certain lines will be down for several hours.
"Our goal is for the costs of the AI agent to pay for themselves for the customer within one year. This means that the purchasing decision is relatively easy for our customers," explains Glaser. "The systems are primarily a door opener for us – regardless of the size of the company. This allows customers to get to know the technology and verify the added value for their own production."
One Bosch customer that is currently using the AI agent is the Baden-Württemberg-based sensor manufacturer Sick. The company, which offers devices for measuring, detecting, monitoring, identifying, and positioning for factory, logistics, and automation systems, uses the Bosch solution at its plant at its headquarters in Waldkirch in the Black Forest, among other locations. "On the very second day, we achieved results at the level of experienced experts – with reliable performance directly on our factory floor," says Sick's Chief Technology Officer, Niels Syassen. "The shop floor AI agent helps us to resolve errors in production systems more efficiently, accelerate troubleshooting, and provide targeted recommendations for action." Sick sees a particular advantage in the system's multilingual capabilities. The AI agent supports both workers and plant operators by generating fault reports in multiple languages and transferring the information directly into the system. This not only increases efficiency but also improves the quality of the documentation, explains Christiane Becherer, who is responsible for developing AI applications at Sick. Her conclusion: "The linguistic flexibility promotes the integration of international teams, overcomes barriers, and enables inclusive collaboration – directly at the interface of humans, machines, and information." [1]
1. Wenn der Laser mitdenkt und sich selbst korrigiert. Frankfurter Allgemeine Zeitung; Frankfurt. 21 Oct 2025: 21. Von Benjamin Wagener, Ravensburg
Komentarų nėra:
Rašyti komentarą