⚡ electrical engineering

From Stabilizing Regions to Certified Controllers: Closing the Selection Gap in Unified PID/PI Analysis for Time-Delay Plants

This paper enhances a unified PID/PI analysis framework for time-delay plants by analytically determining unstable-pole counts to automate region partitioning, introducing a time-domain selection rule to identify certified monotone controllers with minimum settling time, and addressing neutral-type stability pitfalls to deliver a fully automated pipeline for controller design.

Original authors: Senol Gulgonul

Published 2026-06-24

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Senol Gulgonul

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to tune a radio to get a clear signal, but the station is moving, and the signal is delayed by the time it takes to travel through space. In engineering, this is like controlling a machine (a "plant") that has a time delay. The goal is to find the perfect settings (gains) for a controller (like a PID or PI controller) so the machine runs smoothly without crashing or wobbling.

This paper addresses a recent method that helps engineers find the "safe zone" of settings, but the authors argue that method leaves two big jobs unfinished. Here is the paper explained in simple terms:

The Problem: A Map Without a Compass

A recent study (An et al., 2026) created a sophisticated map called a "D-partition." Think of this map as a territory divided into different colored cells.

What the map does: It draws lines that separate areas where the machine might be stable from areas where it might be unstable. It uses a tool called a "Boundary Gradient Vector" (BGV) to show which side of a line is "better."
What the map misses:
1. The "Where am I?" problem: The map tells you which way to turn to get more stable, but it doesn't tell you if you are currently in a safe zone or a dangerous one. You still have to pick a random spot inside a cell and run a slow, manual test to see if it works.
2. The "Which one?" problem: Even if you find a safe zone, the map gives you a huge area of possibilities. It doesn't tell you which specific setting is the best one to actually use. Two settings in the same safe zone might make the machine react very differently (one might be slow, another might overshoot and crash).

The Paper's Solution: Three New Tools

The authors of this paper say, "We can fix this." They offer three specific improvements:

1. The Automatic Counter (No More Guessing)

The Analogy: Imagine the map has a hidden counter that tells you exactly how many "bad" roots (instabilities) are in each cell. The old method said, "Go check a random spot to count them."
The Fix: The authors show you don't need to guess.

If there is no delay, you can use a simple algebraic formula (like a Routh test) to count the bad roots instantly.
If there is a delay, they use a mathematical "argument principle" (a way of counting how a function twists around a circle) to get the exact count without searching.
Result: Every cell on the map is now labeled with its exact stability score. You know immediately which cell is the "Goldilocks" zone (zero bad roots) without doing any manual testing.

2. The "Perfect Speed" Selector (Picking the Winner)

The Analogy: Finding the safe zone is like finding a parking lot. But you still need to pick the one parking spot that is closest to the exit and won't scratch your car. The old method just said, "Park anywhere in this lot."
The Fix: The authors add a rule to pick the single best controller from the safe zone.

They look for a controller that makes the machine move as fast as possible without overshooting (going past the target and coming back).
They use a "tangency condition," which is like finding the exact moment a ball touches the ground without bouncing.
They prove this choice is safe by checking "external positivity," which ensures the machine's reaction is always smooth and never jerks backward.
Result: Instead of a whole region of options, you get one certified controller that is guaranteed to be fast, smooth, and stable.

3. The "Trap" Warning (Avoiding the Neutral Pitfall)

The Analogy: Imagine a bridge that looks solid but has a hidden structural flaw: if you put too much weight on one specific part, the whole thing starts vibrating forever.
The Fix: The authors point out a specific trap in the old method. If you use a specific type of controller (Ideal PID) on a specific type of machine (First-Order-Plus-Dead-Time), the math becomes "neutral type."

This means if the derivative gain is too high, the system develops an infinite chain of unstable roots right on the edge of safety.
The old method didn't warn about this. The new paper says, "Don't cross this line," effectively adding a safety fence that the original map missed.

The Proof: Testing the Theory

The authors didn't just talk about this; they tested it:

Recreating the Past: They took a famous "delay-free" example from the original paper and reproduced the results exactly, proving their new counting method works perfectly.
The Full Pipeline: They applied their new method to a machine with a time delay. They showed the map, identified the safe zone automatically, and then selected the single best controller.
The Result: They produced a controller that settled the machine in about 2.4 seconds with zero overshoot. They showed that if you picked a slightly different setting (even one in the same safe zone), the machine would either overshoot or be too slow.

The Bottom Line

The original method was a great map that showed you where the safe territory was, but it left you to guess your exact location and didn't tell you which path to take.

This paper says: "Here is a way to label every spot on the map automatically, and here is a rule to pick the single best path that guarantees a smooth, fast ride." They moved the conversation from "Where can we go?" to "Here is exactly where we should go, and here is proof it will work."

What is NOT solved: The paper admits that while this works great for simple machines or 3-parameter setups, the "holy grail" of solving for all four parameters (Kp, Ki, Kd, and delay) simultaneously in a clean, simple way is still an open problem. They solved the immediate gaps, but the biggest mountain is still there.

Technical Summary: From Stabilizing Regions to Certified Controllers

Problem Statement
The tuning of PID and PI controllers for time-delay plants remains challenging due to the transcendental nature of the closed-loop characteristic equation, which results in complex, awkward-to-describe stabilizing sets. A recent unified treatment by An et al. (2026) utilizes a Boundary Gradient Vector (BGV) combined with D-partitioning to map the boundaries of stabilizing, relative-stability, and stability-margin regions. While this method effectively answers the feasibility question of where admissible gains lie, it leaves two critical gaps:

Absolute Pole Count: The BGV method only provides relative changes in the number of right-half-plane (RHP) poles across boundaries. Determining the absolute count of unstable poles within a specific cell still requires a manual interior-point test, which is computationally expensive for transcendental equations.
Controller Selection: The method yields a region of feasible parameters but offers no mechanism to select a single, optimal controller. It provides no guarantees regarding time-domain performance (e.g., overshoot or settling time), as gain/phase margins do not strictly constrain these metrics.

Additionally, the unified analysis fails to delimit the plant class for ideal PID control on First-Order-Plus-Dead-Time (FOPTD) plants, potentially leading to neutral-type systems with marginal stability.

Methodology
This work proposes a three-stage pipeline to close the selection gap and correct structural omissions:

Analytic Root Counting (Stage 1):
- Delay-Free ( $\tau=0$ ): The unstable pole count is determined exactly via the eigenvalues of the companion matrix or a Routh array, eliminating the need for interior-point searches.
- Retarded Delay ( $\tau>0$ ): For retarded-type loops, the absolute count is derived using an argument-principle (Mikhailov) evaluation along a contour enclosing the RHP. This frequency-domain evaluation provides the absolute count at a single "anchor" point.
- Integration: By combining the analytic anchor count with the BGV crossing signs (which indicate relative changes across boundaries), every cell in the parameter space is automatically labeled with its exact RHP pole count, rendering the partition fully decided without manual testing.
Time-Domain Certified Selection (Stage 2):
- Within the feasible stabilizing set, a specific controller is selected based on time-domain criteria.
- Monotonicity: The method enforces external positivity (a nonnegative closed-loop impulse response), which guarantees a monotone step response with no overshoot or undershoot.
- Optimization: Among monotone responses, the controller with the minimum 2% settling time is chosen. This optimum is characterized by a tangency condition where the impulse response touches zero ( $\min h(t) = 0$ ) without becoming negative.
Structural Correction (Neutral-Type Pitfall):
- The paper identifies that an ideal PID controller applied to a FOPTD plant creates a neutral-type system.
- It establishes the necessary constraint $|kK_d| < T$ to ensure stability. At the boundary $kK_d = T$ , a chain of roots lies on the imaginary axis, leading to numerical instability and ill-posed designs. The analysis restricts the scope to PI control or filtered derivatives to avoid this.

Key Results

Reproduction of Benchmarks: The paper exactly reproduces the delay-free benchmark (Example 2 from An et al.) for a fifth-order system. Using the analytic Routh count, it recovers the admissible $K_p$ intervals with high precision, confirming the stabilizing cells without any interior-point tests.
FOPTD Case Study: On a FOPTD plant ( $G(s) = e^{-0.5s}/(s+1)$ $G (s) = e^{- 0.5 s} / (s + 1)$ ), the pipeline successfully identifies the stabilizing region and selects a certified PI controller ( $K_p \approx 0.90, K_i \approx 0.85$ $K_{p} \approx 0.90, K_{i} \approx 0.85$ ).
- The selected controller achieves a 2% settling time of $\approx 2.4$ s (approx. $4.8L$ ) with zero overshoot.
- The results align with the closed-form tangency rules for FOPTD plants found in prior literature (An et al., 2026; [15]), validating the numerical approach.
- The impulse response of the selected controller is nonnegative and touches zero, satisfying the external positivity certificate, whereas higher-gain designs violate this condition and exhibit overshoot.
Volume Visualization: The method extends to three-parameter slices (including adjustable delay $\tau$ ), visualizing the stabilizing volume which contracts as delay increases, all without per-slice interior testing.

Significance and Claims
The authors position this work as a necessary extension of the BGV framework rather than a replacement of its core mathematics.

Removal of Manual Steps: The primary technical contribution is demonstrating that the absolute unstable-pole count is analytically available (via Routh or Mikhailov methods), thereby removing the manual interior-point test conceded as necessary by the original authors.
From Region to Controller: The substantive improvement is the addition of a selection stage. While the BGV method identifies where stable gains exist, this pipeline determines which controller to use and provides a rigorous guarantee of its time-domain behavior (monotonicity and settling time).
Scope and Limitations: The authors are modest about the novelty of Stage 1, noting that the underlying counting methods are classical. The genuine increment is Stage 2. They also clarify that the "four-parameter" problem (simultaneous tuning of $K_p, K_i, K_d, \tau$ ) remains an open problem, as the current frameworks (including this one) effectively handle at most three parameters at a time. The paper concludes that moving from identifying regions to delivering a certified, single controller is the critical step for practitioners that region-only methods cannot provide.