The AI revolution is here – now it’s time to get in on the action
Artificial intelligence (AI) is playing to its strengths, especially in Industry 4.0. The systems make predictive maintenance come true, turn robots into autonomous helpers in the smart factory and uncover errors in production. However, the technology should no longer be the focus only of research and development departments of large corporations, because it is now affordable for everyone. The guidebook uses examples to show where and how the advantages of AI can be used profitably and which standard components are already available for this today.
Artificial intelligence is currently one of the driving forces in the digital transformation. The promises are great, even if not all of them can be delivered right now. This is because so-called strong AI is still a thing of the future – in other words, systems that are equipped with intellectual, human-like abilities.
Self-learning software, on the other hand, which is geared to precisely one task, is increasingly taking over our everyday lives. Such machine-learning systems are not programmed like classic software, but are trained with a large amount of data. They are then constantly learning new things as they are used. Experts group such methods under the term weak AI. But the word “weak” gives a false impression of the systems’ capabilities. Because they are already impressive – and we are just at the beginning of development.
In everyday private life, software based on machine learning controls heating or lighting systems in the smarthome, for example. Many of the apps on our smartphones also use AI. In companies, intelligent chatbots take over communication with customers. And in cars, AI is taking over more and more tasks from humans – right up to autonomous control of the vehicle.
IoT data makes maintenance more efficient
Artificial intelligence shows its added value particularly clearly in industry. This applies primarily to applications in which it is used to process data from the Internet of Things (IoT). Sensors from production machines provide a wide range of different information – for example, on speed, temperature or pressure. When this is combined with other data – for example from the production control systems, for example – and analyzed with an AI system, amazing insights can be gained. Industrial companies can use it, for example, to recognize at an early stage when a machine is no longer functional. Maintenance work is carried out when it is really necessary and before the machine actually breaks down. This makes maintenance more efficient and prevents downtime. We are talking about predictive maintenance.
20,000 to 30,000 sensor messages per machine
At its Munich plant, for example, carmaker BMW collects sensor data from 600 welding guns used by robots in car body construction. Sensors measure friction three times per shift and report when deviations occur. Software continuously evaluates the data with the help of AI and can thus make predictions as to when a failure is imminent.
Printing press manufacturer Koenig & Bauer also analyzes the data from its products. The company receives 20,000 to 30,000 sensor messages per machine every day. The goal here is also to have maintenance and the associated downtimes take place in a planned manner.
Predictive maintenance with the help of AI can become an important cost factor. German production companies alone spend a total of 14 billion euros too much on maintenance costs each year (according to the guide from the Machine Tool Laboratory (WZL) at RWTH Aachen University, which deals with predictive maintenance).
Machine learning detects cracks in car bodies
Artificial intelligence also brings concrete benefits in quality assurance. This can be seen, for example, at Audi’s press shop in Ingolstadt. Cameras there check all components directly after production. Machine learning software developed in-house analyzes these images. If it detects defects in the sheet metal parts – such as fine cracks – it marks them with pixel precision. In this way, the software speeds up the inspection processes because the cameras previously had to be configured for each new component. Thanks to AI’s learning ability, this is no longer necessary. The system also ensures more reliable results.
The example illustrates the potential that AI offers, especially in quality control via image processing. Many tasks there are difficult to formalize – partly because of the variety of possible defects. If you want to perform these tasks with a classic image processing system, this requires an expert. This is not necessary with a system that works with machine learning. It learns by being shown a certain number of images. After that, it can then independently detect the defects on the components.
The next step would be for a manufacturing machine to correct itself when it produces a defective part. Thanks to AI, this possibility is within reach. One example is a project at the Augsburg Institute of Textile Technology. The experts are using AI to recognize how various machine parameters – for example, speed – are related to the quality parameters of the products during the production of nonwovens. The machine is then supposed to independently diagnose faults and make appropriate adjustments. Rejects are expected to be reduced by 30 to 50 percent in this way.
Robot becomes quality inspector
But these are by no means all the possible applications for artificial intelligence in industry. Today, the technology is already enabling robots to act largely autonomously. Cameras become the eyes of the mechanical helpers. AI processes the optical information and thus helps with orientation.
One example comes from quality assurance. In the Aumero research project, Zeiss, BMW and scientists at the University of Ulm have developed an autonomous measuring robot. Among other things, it can independently inspect the gap dimensions of body parts such as vehicle doors. With the help of cameras, Aumero recognizes its target independently and aligns itself with the car body to measure the gap dimension. The user uses software to select the object to be inspected, its rough location and then the measurement plan. The robot does the rest on its own.
AI is also being used to help mobile robots or so-called automated guided vehicles (AGVs) navigate through factory floors. For example, there are pilot projects with robots equipped with 3D cameras. Here, the image data is sent to a computer via a 5G wireless connection, where it is analyzed with AI. In this way, an AGV can not only identify obstacles, but also recognize whether it is a worker or a pallet, for example. If it is a human, the robot moves around it with the necessary minimum distance.
In this way, the mobile robots can also search for things on the factory floors. If a certain object is missing in the factory halls, a command goes out to the robot via the ERP system. The robot then starts searching and is able to locate the object using image processing and AI. If it is successful, it sends a corresponding message to the system.
Simple programming in robotics
AI doesn’t just expand the capabilities of robots, however. Thanks to the technology, they can also be programmed more easily. A number of companies have developed corresponding procedures based on machine learning. To teach the robot its task, for example, the user guides its arm with his hand. These movements are recorded by a camera and a force-moment sensor and then processed with the help of AI. In another method, the user uses a sensor-equipped pen to indicate the movement that the robot should perform. If the robot is used for welding, for example, the user simply traces the welding path on the corresponding workpiece with the pen. The software then generates the programming code for the robot.
Small and medium-sized businesses can also use AI
Small and medium-sized companies can now also benefit from these advantages. This cannot be taken for granted. The lack of know-how and skilled workers do not make it easy for SMEs to introduce AI systems. In addition, self-learning systems require data with which they can be trained. However, these are not always available in sufficient quantities in small and medium-sized companies.But while research and increasing computer capacity have caused the number of AI applications in general to grow strongly, the barriers to entry for using the technologies have also been lowered. In the meantime, even medium-sized companies can equip themselves with easy-to-use technology. This not only allows them to gain initial experience, but also to implement even complex AI projects.
From LiDAR to Raspberry Pi – the technology is there
These include low-cost LiDAR sensors, powerful LiDAR laser scanners or multi-layer LiDAR systems. After all, sensors provide the data that AI systems work with. Without data, there is virtually nothing for AI to do. And a LiDAR system can provide information from quite a few areas. For example, it can be used as a distance, level or range sensor, as a safety sensor for AGVs or for obstacle avoidance for robots. Thus, the technology creates the basic prerequisite for the use of AI in very many applications.Raspberry Pi or Arduino, for example, can serve as an entry-level technology. What began as a single-board computer for private hobbyists has now reached industry. The Raspberry Pi can be used to easily implement image processing applications, among other things. The computer can also be used to monitor automation and control systems. Popular products include the Portenta H7 Board, the Google Coral Dev Board, especially in combination with the Google Coral USB Accelerator, and the MAX78000 Feather Board.
With the Robot Operating System (ROS), an open source framework is also available for developing autonomous systems, for example. Its advantage is also that it is easy to use. Users can program applications without expert knowledge. The spectrum of possibilities is wide: ROS supports 145 robot systems. The TurtleBot3 serves as the organization’s official development platform for companies, research institutions and universities. The TurtleBot3 is equipped with a Raspberry Pi development board, a 360° LiDAR system, an OpenCR control unit, a Raspberry Pi camera and two servo motors for locomotion.Furthermore, there are standard robots such as the Dingo, which is suitable for research and education projects, among others, and can be extended by a variety of additional modules, e.g. with a 3D camera. A robot like the QUADRUPED Go2 Air, which is equipped with high-precision sensors and powerful artificial intelligence, and which can operate in a wide range of terrains, is particularly suitable for the development of autonomous systems.
Conclusion
The topic of AI holds a lot of potential and will impact all areas of manufacturing. At the same time, AI is about to radically change the competitive landscape, providing early adopters with tremendous benefits. While productivity improvements, quality control and process optimization are usually the main focus, the optimization potentials reach further and include, for example, product development, logistics and customer experience.
The technologies and affordable standard components are available to everyone today to get in on the big AI revolution. It’s also worthwhile for small and medium-sized companies. Because the advantages of AI can bring the decisive meters to be or remain ahead in the competitive race.
Images: Adobe Stock
