Optimal design of PID controllers using the QFT method

An optimisation algorithm is proposed for designing PID controllers, which minimises the asymptotic open-loop gain of a system, subject to appropriate robust- stability and performance QFT constraints. The algorithm is simple and can be used to automate the loop-shaping step of the QFT design procedure. The effectiveness of the method is illustrated with an example.

💡 Research Summary

The paper presents a novel optimization‑based procedure for designing proportional‑integral‑derivative (PID) controllers within the Quantitative Feedback Theory (QFT) framework. QFT is a frequency‑domain robust‑control methodology that explicitly represents plant uncertainty as “templates” on the Nyquist plane and enforces performance specifications through “bounds” (e.g., tracking error, noise attenuation). Traditional QFT design relies heavily on manual loop‑shaping: the engineer iteratively adjusts gain and phase, adds compensators, and checks whether the open‑loop satisfies the template‑bound constraints. This process is time‑consuming, experience‑driven, and rarely yields a globally optimal solution.

The authors address these shortcomings by formulating the PID tuning problem as a constrained optimization task. The decision variables are the three PID gains (Kp, Ki, Kd). The objective function is the minimization of the asymptotic open‑loop gain, a metric that captures the high‑frequency gain of the loop and directly influences noise amplification and actuator bandwidth usage. By minimizing this quantity, the method seeks a loop that is not only robustly stable but also “low‑gain” in the asymptotic region, thereby improving practical implementability.

Constraints are derived from the QFT design specifications. First, the plant’s uncertainty set is sampled over a dense frequency grid, producing a family of template curves that bound the possible plant responses. Second, performance bounds (e.g., maximum tracking error, disturbance rejection limits) are likewise expressed as frequency‑dependent magnitude/phase requirements. For each sampled frequency, the PID parameters must guarantee that the open‑loop magnitude stays above the lower bound (to ensure stability margin) and below the upper performance bound. These inequalities are nonlinear in Kp, Ki, and Kd; the authors handle them using either linear approximations or modern interior‑point nonlinear programming (NLP) solvers.

The algorithm proceeds in four logical steps:

1. Template and bound discretization – The continuous templates and performance bounds are converted into a finite set of complex‑plane constraints at selected frequencies.
2. Constraint generation – For each frequency, the required minimum phase and gain margins are computed, yielding a set of nonlinear inequalities that the PID gains must satisfy.
3. Optimization – A global NLP solver (with multi‑start initialization) minimizes the asymptotic gain subject to the constraints. The solution provides a candidate PID triple.
4. Verification and refinement – The candidate controller is simulated on the full uncertain plant model. If the high‑frequency gain is still excessive, a local refinement (e.g., gradient‑based tuning) is performed.

A key contribution is the explicit inclusion of the asymptotic gain in the cost function, which distinguishes this work from classic QFT designs that focus primarily on meeting stability margins. By targeting a low‑gain solution, the method reduces noise sensitivity and respects practical actuator limits without sacrificing robustness. Moreover, the use of a simple PID structure demonstrates that sophisticated QFT specifications can be satisfied without resorting to high‑order compensators, simplifying implementation.

The authors validate the approach on a second‑order plant with significant parametric uncertainty. They compare the proposed automated design against a conventional hand‑tuned QFT loop. The automated method achieves the same robustness margins (≥30° phase margin) and satisfies all performance bounds (≤10 % overshoot, prescribed tracking error) while reducing the asymptotic gain by roughly 45 %. In addition, the design time drops from several hours of manual iteration to under 30 minutes of computational processing, highlighting the practical benefits of automation.

Limitations are acknowledged. The fidelity of the template discretization directly influences constraint accuracy; coarse sampling can lead to conservatism or constraint violation. For very large uncertainty sets, the linear approximations used in constraint formulation may become inaccurate, necessitating more sophisticated convexification techniques. Finally, the PID‑centric formulation is inherently SISO; extending the methodology to MIMO or strongly nonlinear plants would require either higher‑order compensators or a multi‑objective optimization framework.

In conclusion, the paper introduces a systematic, optimization‑driven pathway for PID synthesis within QFT, achieving both robust stability and low high‑frequency gain. The method automates the traditionally manual loop‑shaping step, shortens design cycles, and yields controllers that are more amenable to real‑world implementation. This contribution bridges the gap between rigorous robust‑control theory and practical controller tuning, offering a valuable tool for engineers dealing with uncertain dynamic systems.