Abstract
Cascaded controller tuning is a multi-step iterative procedure that needs to be performed routinely upon maintenance and modification of mechanical systems. An automated data-driven method for cascaded controller tuning based on Bayesian optimization is proposed. The method is tested on a linear axis drive, modeled using a combination of first principles model and system identification. A custom cost function based on performance indicators derived from system data at different candidate configurations of controller parameters is modeled by a Gaussian process. It is further optimized by minimization of an acquisition function which serves as a sampling criterion to determine the subsequent candidate configuration for experimental trial and improvement of the cost model iteratively, until a minimum according to a termination criterion is found. This results in a data-efficient procedure that can be easily adapted to varying loads or mechanical modifications of the system. The method is further compared to several classical methods for auto-tuning, and demonstrates higher performance according to the defined data-driven performance indicators. The influence of the training data on a cost prior on the number of iterations required to reach optimum is studied, demonstrating the efficiency of the Bayesian optimization tuning method.
Introduction
In industrial manufacturing, linear and rotational drives are the workhorses of the factory floor. These systems typically rely on cascaded PID control loops—nested architectures where an outer position loop provides the reference for an inner speed loop. However, tuning these controllers remains a significant challenge. Mechanical properties often drift over time due to friction or wear, and changes in system load frequently necessitate a complete re-tune.
Engineers often resort to conservative settings that sacrifice performance for stability, or time-consuming manual "trial and error" methods. As nested loops increase the number of parameters, the complexity of finding optimal performance gains grows exponentially. This paper presents an automated, data-driven framework for tuning these cascaded controllers using Bayesian Optimization (BO), identifying near-optimal gains with minimal experimental trials.
Methodology
The proposed approach utilizes Sequential Bayesian Optimization. Rather than a blind search, controller tuning is treated as an optimization problem where a "cost" function—a weighted sum of performance metrics like overshoot, settling time, and tracking error—is minimized.
The core components of the methodology include:
- Gaussian Process (GP) Regression: A GP creates a probabilistic model of the cost function, allowing the algorithm to predict performance at untested configurations and quantify uncertainty.
- Acquisition Function: A Lower Confidence Bound (LCB) strategy balances exploration of uncertain regions with exploitation of predicted high-performance areas.
- Sequential Tuning: To manage complexity, the inner speed loop is tuned first. Once optimized, its gains are fixed, and the BO algorithm proceeds to tune the outer position loop.
The method was evaluated using a high-fidelity numerical model of a linear axis drive, incorporating DC motor dynamics and ball-screw mechanics. It was benchmarked against classical methods such as Ziegler-Nichols (ZN), ITAE optimization, and Relay tuning.
Results and Discussion
The BO method demonstrated remarkable efficiency, reaching near-optimal gains in just 20–30 iterations. In contrast, a traditional grid search would require thousands of evaluations to achieve similar precision. The BO-tuned gains ($K_p=225, K_v=0.37, K_i=130$) closely matched the exhaustive grid search results ($K_p=225, K_v=0.36, K_i=130$).
As shown in the performance comparison, classical heuristics like Ziegler-Nichols often produce aggressive gains leading to high overshoot and oscillations. The BO approach, however, identifies a balanced configuration that respects physical limits while providing superior tracking. The researchers also noted that the cost surface for the speed controller is relatively flat near the minimum, a region BO successfully identified, suggesting inherent robustness in the found parameters.
Conclusion
Bayesian Optimization provides a data-efficient and high-performance alternative to traditional controller tuning methods. By modeling the cost landscape probabilistically, it avoids the pitfalls of manual tuning and the conservatism of standard heuristics. Future work will focus on incorporating "safe exploration" algorithms to automatically detect instabilities, ensuring the system remains within safe physical bounds throughout the automated tuning process.
Summary updated