PID Tuning via Desired Step Response Curve Fitting

Technical Summary: PID Tuning via Desired Step Response Curve Fitting (PID-SRCF)

Problem Statement
The tuning of Proportional-Integral-Derivative (PID) controllers remains a central challenge in control engineering, requiring the calculation of coefficients ($K_p, K_i, K_d$) to meet specific time-domain performance criteria such as settling time, overshoot, and robustness. While numerous analytical and empirical methods exist—including Ziegler-Nichols, Lambda tuning, Pole Placement, and MATLAB's proprietary tools—these approaches often rely on specific plant models (e.g., First-Order Plus Time Delay), heuristic rules, or implicit target responses that may not align precisely with user-defined transient specifications. Furthermore, analytical methods like pole placement often neglect the impact of closed-loop zeros, potentially leading to unexpected overshoot, while proprietary tools may lack transparency regarding their optimization objectives.

Methodology
The paper proposes the PID Step Response Curve Fitting (PID-SRCF) method, a numerical optimization approach that directly minimizes the $L_2$-norm error between an actual closed-loop step response and a user-defined desired step response.

• Objective Function: The core of the method is a constrained nonlinear optimization problem that minimizes the root-mean-square error (RMSE) between the desired trajectory $y_{desired}(t)$ and the actual system response $y_{PID}(t)$ over a simulation time $T_{sim}$:
  $$\min_{K_p, K_i, K_d} \sqrt{\int_{0}^{T_{sim}} |y_{desired}(t) - y_{PID}(t)|^2 dt}$$
• Desired Response Definition: Users define the target response either as a transfer function (e.g., First-Order Plus Time Delay for non-oscillatory behavior) or by specifying performance parameters for a second-order system (settling time $T_s$ and percent overshoot $PO$). The algorithm automatically calculates the corresponding damping ratio ($\zeta$) and natural frequency ($\omega_n$).
• Optimization Algorithm: The problem is solved using MATLAB's fmincon solver with the Sequential Quadratic Programming (SQP) algorithm. The optimization is constrained to ensure non-negative PID gains ($K_p, K_i, K_d \geq 0$) and is initialized from zero to avoid bias.
• Validation: Post-optimization, the method verifies closed-loop stability by inspecting pole locations and evaluates robustness via the maximum sensitivity ($M_s$) metric.

Key Contributions

1. Unified Framework: The paper posits that PID-SRCF acts as a "meta-method" that subsumes classical tuning techniques. By selecting specific desired transfer functions, the method can replicate the results of Ziegler-Nichols (targeting a quarter-decay response), Lambda tuning (targeting a FOTD response), and Pole Placement (targeting a critically damped response).
2. Explicit Specification: Unlike traditional methods where the target response is implicit in the tuning rules, PID-SRCF allows the engineer to explicitly define the desired closed-loop dynamics, decoupling the design specification from the solution method.
3. Open-Source Implementation: The authors provide a transparent, open-source MATLAB implementation, contrasting with proprietary black-box tools. This allows for inspection and modification of the optimization logic.
4. Robustness Analysis: The study analytically demonstrates that specifying a non-overshooting response (First-Order or Critically Damped Second-Order) inherently bounds the maximum sensitivity ($M_s$) within robust industrial thresholds ($1 < M_s < 2$). Specifically, a critically damped second-order target yields an $M_s$ of $2/\sqrt{3} \approx 1.155$.

Results
Comparative evaluations were conducted across various system orders and tuning scenarios:

• Ziegler-Nichls Case: For a third-order process, PID-SRCF achieved PID coefficients and performance metrics (settling time, overshoot, IAE) nearly identical to the Ziegler-Nichols reaction curve method, confirming that Ziegler-Nichols implicitly targets a specific second-order response shape.
• Lambda Tuning Case: For a First-Order Plus Time Delay (FOTD) process, the PI coefficients derived by PID-SRCF closely matched those of the Lambda tuning method (with pole-zero cancellation), validating its ability to replicate standard industrial practices.
• Pole Placement Case: When targeting a non-overshooting response for a first-order process, PID-SRCF yielded lower gain values than analytical pole placement. The analytical method, which ignores the closed-loop zero, resulted in significant overshoot (9.2%), whereas PID-SRCF achieved the target with only 2.5% overshoot.
• High-Order Systems: For a third-order process, PID-SRCF produced PI coefficients almost identical to MATLAB's proprietary "Control System Tuner" (differing only in the fourth significant figure). This similarity is attributed to both tools minimizing the deviation between the closed-loop response and a reference model, though PID-SRCF achieves this via an open-source solver (fmincon) rather than a patented one (systune).

Significance and Claims
The paper claims that PID-SRCF offers a flexible, computationally efficient alternative to both heuristic analytical methods and proprietary software. Its primary significance lies in its ability to:

• Replace multiple distinct tuning methods with a single, unified algorithm by simply changing the desired response model.
• Provide precise control over transient response characteristics (overshoot and settling time) without relying on approximations that may fail for higher-order systems.
• Ensure robustness by naturally bounding sensitivity when non-overshooting targets are selected.
• Offer transparency and accessibility through an open-source implementation, allowing practitioners to understand and modify the tuning logic without licensing constraints.

The authors conclude that while the underlying optimization principle (minimizing error against a reference model) is not novel, the explicit formulation and open-source availability of PID-SRCF provide a practical, unified tool for industrial PID tuning that can effectively replace established methods.