RL-based Parameter Auto-tuning

  • Control and optimization
  • Manufacturing


In modern manufacturing, software-based motion control is essential for regulating equipment and facilities. However, a persistent challenge arises due to the mismatch between commanded control values and the actual values received through the device driver when equipment parameters are not finely tuned. On-site engineers are responsible for fine-tuning these parameters, but variations in operator proficiency lead to inconsistent outcomes and extended tuning times. This lack of standardization hampers progress and optimization efforts for the enterprise.


To address this challenge, we introduced a machine learning-based simulator, known as the dynamics model, to predict the device driver's actual behavior based on control and parameter tuning values. Alongside this, we leveraged reinforcement learning techniques to conduct virtual trial-and-error experiments on a large scale. This powerful combination enabled us to identify optimal tuning values, streamlining the fine-tuning process and optimizing the motion control system's peak performance.


Compared to expert-driven tuning, we achieved an impressive reduction in tuning time by up to 52%, while simultaneously increasing accuracy by up to 20%. This enhancement ensures a much closer alignment between the intended commands and the actual behavior of the machines.

Want to learn more about this use case?
Get in touch with our industrial AI experts.

Talk to AI Experts