Model-free practical PI-Lead control design by ultimate sensitivity principle

This paper proposes a model-free, three-step procedure for designing robust PI-Lead controllers based on the ultimate sensitivity principle and loop shaping characteristics, which is validated through experiments on a noise-perturbed electro-mechanical actuator.

The Gist

Imagine you have a very complex machine, like a robotic arm or a motor, and you need to make it move precisely to a specific spot. Usually, to control such a machine, engineers need a detailed "blueprint" (a mathematical model) of how it works. But what if you don't have that blueprint? What if the machine is old, mysterious, or just too complicated to map out perfectly?

This paper presents a clever, "model-free" way to tune a controller for such machines. Think of it as tuning a radio or adjusting the suspension on a car by listening and feeling rather than reading a manual. The author, Michael Ruderman, proposes a three-step recipe to get the machine moving smoothly without ever needing to know its internal math.

Here is the breakdown of the method using everyday analogies:

The Goal: The "Goldilocks" Control

The paper focuses on a specific type of machine (called a "Type-One" system) that naturally tends to drift or integrate motion, like a car coasting or a motor turning a wheel. The goal is to add a "PI-Lead" controller.

• PI (Proportional-Integral): Think of this as the main driver. The "Proportional" part pushes harder if you are far from the target. The "Integral" part is like a patient memory that keeps pushing gently until the error is gone, even if the push is small.
• Lead: This is a "turbo boost" or a "shock absorber" that adds a little extra stability and speed to the reaction, preventing the machine from wobbling.

The Three-Step Tuning Recipe

The author suggests a simple, experimental process to find the perfect settings:

Step 1: Finding the "Sweet Spot" for Patience (The Integrator)

Imagine you are trying to balance a broom on your hand. If you are too slow to react, it falls. If you are too jittery, you shake it off.

• The Experiment: You start with a very "patient" setting (a slow reaction time). Then, you gradually make the controller "impatient" (faster reaction).
• The Signal: You watch the machine's output. At first, it's calm. As you speed it up, it starts to wobble. You keep speeding it up until it starts wobbling back and forth forever (permanent oscillation).
• The Result: The moment it starts that endless wobble is the "danger zone." The author says, "Okay, we found the edge. Let's back off just a little bit to be safe." This gives you the perfect "patience" setting for the controller.

Step 2: Adjusting the "Push" (The Gain)

Now that the machine is stable but maybe a bit sluggish, you need to decide how hard it should push.

• The Experiment: You gradually turn up the "volume" (the gain) of the controller.
• The Signal: You watch how much the machine "overshoots" (goes past the target and then comes back).
• The Goal: You want the machine to overshoot just enough to be snappy, but not so much that it crashes. The author suggests aiming for an overshoot of about 30% to 40%. It's like jumping off a diving board: you want to go high enough to clear the water, but not so high you hit the ceiling. Once you hit this "just right" overshoot, you lock in that setting.

Step 3: Adding the "Turbo Boost" (The Lead Compensator)

Even with the right patience and push, the machine might still be a bit sluggish when things get tricky (like when there is noise or friction).

• The Solution: The author adds a "Lead" element. Think of this as adding a shock absorber to a bumpy ride. It doesn't change how the car drives on a straight road, but it smooths out the bumps and helps the car recover faster from a sudden jolt.
• The Magic: This step is calculated automatically based on the settings you found in Step 1. It adds a little extra "phase advance" (a fancy way of saying it helps the machine react before the problem gets worse), making the whole system more robust.

The Real-World Test

The author tested this on a noisy, real-world electric motor system.

• The Challenge: The motor had friction, noise, and non-linear quirks (like a sticky brake).
• The Result: The new method worked beautifully. When they pushed the motor (disturbed it), the new controller snapped back to the target position much faster and smoother than a standard controller tuned by old, famous rules (Ziegler-Nichols).
• Comparison: The old method made the motor jump around aggressively (like a car with no suspension), while the new method was firm but smooth.

Why This Matters

The biggest takeaway is simplicity. You don't need to be a mathematician or have a perfect blueprint of the machine. You just need to:

1. Make it wobble until it oscillates, then back off.
2. Turn up the volume until it overshoots just right.
3. Add a pre-calculated "shock absorber."

This makes it possible to tune complex industrial machines quickly and reliably, even when you don't know exactly how they work inside. It turns a complex engineering puzzle into a practical, step-by-step experiment.

Technical Summary

Technical Summary: Model-free Practical PI-Lead Control Design by Ultimate Sensitivity Principle

Problem Statement
The paper addresses the challenge of designing and tuning feedback controllers for dynamic processes where an explicit mathematical model is unavailable. This is a common scenario in industrial applications, particularly for electro-mechanical and hydro-mechanical systems. The specific focus is on Type-one systems (systems with integrating behavior, such as motion control where relative displacement is the output), which present unique tuning difficulties due to their inherent $-90^\circ$ phase lag at low frequencies and lack of open-loop steady-state values. While PID controllers are industry standards, their reliable tuning without a model remains a persistent need, especially given the limitations of integral action in handling saturation and Coulomb friction.

Methodology
The proposed method is a model-free, three-step experimental tuning procedure for a PI-Lead controller. It is derived from the ultimate sensitivity principle (associated with Ziegler–Nichols tuning) but diverges from standard heuristic approaches by focusing on phase margin guarantees rather than just ultimate gain and period. The procedure relies solely on monitoring the system output during closed-loop operation:

1. Integrator Time Constant ($T_i$) Determination:

   • Starting with a nominal proportional gain ($K_p$), the integrator time constant $T_i$ is gradually reduced.
   • The reduction continues until the closed-loop system exhibits permanent oscillations (indicating zero phase margin).
   • The ultimate integrator time constant ($\bar{T}_i$) and the corresponding ultimate corner frequency ($\bar{\omega}_{c,pi} = 1/\bar{T}_i$) are recorded.
   • The final $T_i$ is set to $10 / \max(\bar{\omega}_{gc}, \bar{\omega}_{c,pi})$, where $\bar{\omega}_{gc}$ is the observed oscillation frequency. This factor of 10 shifts the controller's phase advance back by one decade to ensure a sufficient phase margin.

2. Control Gain ($K_p$) Tuning:

   • With $T_i$ fixed, the control gain $K_p$ is varied (incremented and decremented) starting from a nominal value.
   • The tuning target is to achieve a transient overshoot ($M$) in the step response of 30% to 40%.
   • This specific overshoot range corresponds to a damping ratio ($\zeta$) that yields a phase margin ($\phi_m$) between 30° and 40°, ensuring stability without excessive sluggishness.

3. Lead-Compensator Design ($L(s)$):

   • A Lead-compensator is added to enhance robustness and transient performance without significantly altering low-frequency characteristics.
   • The compensator is designed with a fixed ratio $\alpha = 0.1$ (providing $\approx 55^\circ$ of phase lead).
   • The frequency of maximum phase lead ($\omega_{max}$) is set to $10^{1.5}/T_i$ to extend the effective phase advance range.
   • The compensator gain is set to $K_L = 1$ to preserve the loop gain at lower frequencies.

Experimental Evaluation
The method was validated on a noise-perturbed electro-mechanical actuator system (a voice-coil motor with translational degrees of freedom). The system included significant nonlinearities, such as input-gain nonlinearity, coil force ripples, and Coulomb friction, and was subject to sensor noise. No system model or parameter values were used during the tuning process.

• PI vs. PI-Lead: The tuned PI controller achieved the target overshoot but exhibited slower settling and struggled with disturbances. The addition of the Lead-compensator significantly improved the transient response and disturbance rejection.
• Comparison with Ziegler-Nichols (ZN): The proposed PI-Lead controller was compared against a standard PID controller tuned via the Ziegler-Nichols ultimate sensitivity method.
  • The ZN-tuned PID exhibited aggressive behavior with large overshoots and undershoots, though it settled slightly faster due to high-frequency dither helping to overcome Coulomb friction.
  • The proposed PI-Lead controller provided a more balanced response with acceptable overshoot and smoother control effort, avoiding the extreme aggressiveness of the ZN-PID.

Key Contributions and Claims

• Practical Model-Free Design: The paper proposes a straightforward, three-step procedure that requires only experimental monitoring of the output, eliminating the need for system identification or modeling.
• Robustness to Noise: The authors claim the method is robust against feedback noise. The detection of ultimate oscillations relies on low-frequency phenomena dominated by the system's poles at the origin, making them distinguishable from high-frequency sensor noise.
• Handling Type-One Systems: The method specifically addresses the challenges of Type-one (integrating) processes, which are often difficult to tune experimentally due to their phase characteristics.
• Superiority over Heuristics: The paper claims that while the Ziegler-Nichols method is widely used, the proposed approach offers a more controlled phase margin and less aggressive transient behavior, which is often preferable in practical applications where excessive overshoot is unacceptable.

Significance
The paper positions this work as a practical alternative to existing heuristic tuning rules. By grounding the design in the ultimate sensitivity principle but refining the parameter assignment through specific phase-margin and overshoot targets, the method aims to provide a "reliable, straightforward, and practically accessible" tuning solution for engineers dealing with complex, unmodeled industrial systems. The authors note that while the current work focuses on Type-one systems, the methodology could potentially be adapted for Type-zero systems or systems with oscillatory dynamics in future research.