Artificial intelligence (AI) is still a hot topic in many industries, including manufacturing industry. Media covering new AI functions and trends plays a key role in realizing digital production.
In many cases, artificial intelligence only exists in theory, and there is still a long way to go before it can be popularized. Third party service providers tout their AI based technology to make it look more mysterious and expensive than before.
Can the potential of AI be used to justify its investment? Is it really worth optimizing several parameters with AI? Isn’t artificial intelligence just used to catch up with the trend?
Become flexible through AI
AI should not be used only as a means to optimize long-term automated processes. The real potential is to do something completely new with this technology. Tasks previously performed by humans or physical machines can now be performed by AI control software that powers robots. This increases the flexibility and traceability of the robot, improves reliability in many cases, and enables more successful operation in the market.
As the history of digitization has proved, the number of experts in new technology is always limited at the beginning, but it has never stopped making progress. Before PC won in the 1980s, it was easy to believe that every company needed a data center with its own computer scientists to participate in the first wave of digitization. That’s not what happened. Instead, off the shelf products have well-defined interfaces that allow every enterprise, large or small, to leverage it innovation. The key is PC: easy to understand flexible computing technology, which is now widely used.
AI will go the same way in manufacturing. Instead of paying external resources to lead AI projects, manufacturers can buy products with basic AI functions and use them without external help. This is one of the basic assumptions for some component suppliers to develop AI products. Of course, you need to focus on solving complex control problems of products, but you don’t have to be an expert with a degree in computer science.
Building trust in new technologies
The second obstacle is the technology itself, which was initially difficult for many people to understand. Here, it’s important to eliminate the common worry that robots controlled by AI will suddenly burst their will at night. It has been claimed that how AI systems make decisions is unpredictable and incomprehensible. It’s not a real neural network. It’s a sequence of multiplication and addition. They are deterministic, and their working principle can be related to school mathematics, but they do have many parameters. As a result, you can’t tell them at a glance how they made their decisions.
Some people also call on AI to make its decision path easy to understand, and it is better to follow the understandable rules of if then else pattern. If possible, there will be no need for complex models, because regular programming is enough. However, artificial intelligence is the answer to problems, in which there is no solution if other rules are easy to explain. What is needed to build trust in these systems is a testable and reliable system, which can be explored by using the system and understanding how AI responds in a given use case. When this test is done quickly and easily, the findings and the AI driven robots will be trusted.
Automatic manual workstation
For AI vendors, enabling rapid testing is currently a technical challenge. Sometimes it may take some patience to train the AI system for use in production, but it’s worth it. Once these skills are mastered, manufacturers can use AI based robot control solutions to flexibly automate manual workstations. Picking parts, tracking contours, inserting cables, and assembling products can all be achieved through a single small camera on the wrist of the robot. Because all components can be flexibly trained for new tasks, the manipulator and AI software can be used in different positions in production.
For example, at an automotive supplier, a simple automation solution has been established to classify metal parts from a semi ordered grid. The lighting conditions of the facilities are difficult to predict and are often exposed to direct sunlight. In addition, metal parts have high reflectance, so the occurrence of flash rust must be considered. The supplier contacted micropsi industries because its AI system can handle these differences – location, lighting conditions, color and remaining packaging blockages. To do this, the technology must learn to find the next part, regardless of the time of day, sunlight intensity, surface condition and packaging coincidence.
What is more difficult to solve is that the white appliance manufacturers are passing the test application in the verification stage. Here, the positioning of the probe must be very accurate. AI must find solder joints on the copper wire being tested for leakage, and the position, direction, shape and material properties of these solder joints vary greatly.
The two applications are implemented with almost the same hardware: universal robots’ ur5e cooperative robot, AI system and wrist camera, as well as customized tools for the application. The factory staff trained the system on site.
Build Ai expertise internally
At present, many AI products are emerging for manufacturing. They trigger the change of thinking mode and realize the flexible production process controlled by software. Easy to learn products can control the resulting complexity.
Therefore, you can make some optimizations through AI, not just some optimizations. The technology can achieve greater flexibility, independence, flexibility and efficiency. The market must provide products for exploratory learning so that AI can be trusted. If it is successful, the wave of automation, which is comparable to the introduction of PC technology, will become possible.