When Machines Learn to Become More Precise: cross-ING at Swiss Mechatronics
How can motion control become even more precise when conventional control technologies reach their limits? This was the central question addressed by Max Rech, Chief Competence Center Motion at cross-ING, during his presentation "From FeedForward to Learning Control" at the Swiss Mechatronics Cluster. The focus was on how machine learning can effectively complement traditional motion control systems.
Precision Starts Before Errors Occur
Modern machines and robotic systems are expected to operate faster while achieving increasingly higher levels of precision. However, high speeds and acceleration introduce significant physical challenges.
Traditional control systems often rely on feedback: They detect deviations and correct them. The challenge? An error must first occur before the system can respond.
This is where feedforward control comes into play. By using known motion profiles and physical models, it predicts the required control inputs before errors occur. However, this approach also has its limitations. Factors such as friction, temperature variations, gearbox effects or changing loads cannot always be fully captured by mathematical models.
When Machine Learning Bridges the Gap
This is where learning-based systems open up new possibilities. Rather than completely replacing existing physical models, machine learning can capture effects that are difficult to describe mathematically.
The idea: If a controller repeatedly has to correct the same deviations, a learning system can anticipate and compensate for them in the future.
Max Rech presented different approaches to combining conventional control technology with data-driven methods. One key insight became clear: Not every challenge requires an entirely new model. Often, the greatest potential lies in intelligently enhancing existing solutions.
The Best of Both Worlds
The central takeaway from the presentation: Maximum precision does not come from machine learning alone, but from the intelligent interaction of different technologies.
While feedback ensures robustness and physical models represent known system dynamics, machine learning can compensate for recurring effects that existing models cannot adequately describe.
The real question is therefore not whether machine learning can replace traditional models, but where it can deliver the greatest added value.
With his presentation at the Swiss Mechatronics Cluster, Max Rech provided insights into the potential of learning-based control systems and demonstrated how existing approaches in drive and motion technology can be further enhanced through intelligent, data-driven methods.
Would you like to explore this topic in more depth?
Book an expert consultation now and receive tailored advice from our specialists.
Want to learn more about our Competence Center MOTION at cross-ING?
Discover how our interdisciplinary team tackles complex technical challenges.
Comments