Root Locus for Time Delay Systems

Time delays are ubiquitous in control systems, appearing in chemical processes, networked control, and robotic teleoperation. These delays fundamentally alter system dynamics, making stability analysis and controller design more challenging. Understanding how to apply the Root Locus method to systems with delays is crucial for designing robust controllers that can maintain performance despite the destabilizing effects of lag.

Modeling Time Delays in the s-Domain

In control theory, a pure time delay (or transport lag) of $T$ seconds is represented in the Laplace domain by the term $e^{-sT}$. If a system with a forward transfer function $G(s)$ has an input delay, its overall open-loop transfer function becomes $G(s)e^{-sT}$. This exponential term is not a rational function (a ratio of polynomials), which complicates analysis using classical tools like the Routh-Hurwitz criterion or the standard root locus procedure, which are designed for rational transfer functions.

The primary challenge is that $e^{-sT}$ introduces an infinite number of poles and zeros into the system. The characteristic equation of a unity feedback system with delay becomes $1+G(s)e^{-sT}=0$, or equivalently $e^{sT}=-G(s)$. Solving this equation yields an infinite number of roots (system poles), as dictated by the periodic nature of the exponential function in the complex plane. To make this tractable with the root locus technique, we use an approximation to convert the non-rational delay term into a rational form.

The Pade Approximation

The most common method for approximating a time delay for root locus analysis is the Pade approximation. This technique replaces the transcendental function $e^{-sT}$ with a rational transfer function, a ratio of two polynomials. The first-order Pade approximation is given by:

$$e^{-sT}\approx \frac{1-\frac{T}{2}s}{1+\frac{T}{2}s}$$

This is a all-pass filter with a magnitude of 1 for all frequencies, but it provides a useful approximation of the delay's phase lag. Higher-order Pade approximations are more accurate over a wider frequency range but result in more complex rational functions. The critical feature of any Pade approximation is that it introduces right-half-plane (RHP) zeros into the open-loop transfer function. For the first-order approximation, the zero is located at $s=+2/T$.

When you substitute the Pade approximation into the open-loop transfer function, you are now working with a rational function: $G_{ol}(s)=G(s)\cdot \frac{1-(T/2)s}{1+(T/2)s}$. You can then plot the root locus of this approximated system using the standard rules, treating the denominator $(1+(T/2)s)$ as part of the modified $G(s)$.

How Delays and Pade Approximations Affect the Root Locus

The introduction of the Pade approximation dramatically changes the root locus plot and the resulting closed-loop pole locations in two key ways.

First, the right-half-plane zeros added by the approximation are zeros of the open-loop transfer function. Recall that root locus branches originate at open-loop poles and terminate at open-loop zeros. Therefore, one or more branches of the root locus will be attracted to these RHP zeros. As the gain increases, these branches must cross into the right-half plane to reach their terminating zeros, indicating that the closed-loop system will become unstable at a lower gain than the delay-free system. This visually explains why delays reduce the stability margin.

Second, the phase contribution of the delay itself, whether exact or approximated, reduces the phase margin. In the frequency domain, a delay adds phase lag ($-\omega T$ radians) without changing the magnitude. On a root locus, this is equivalent to adding a phase shift that rotates the branches toward the right-half plane. The net effect is that for a given controller gain, the closed-loop poles will be further to the right (less damped, or even unstable) compared to the system without delay. Consequently, to maintain stability, you must use a lower, or "detuned," controller gain, which typically results in slower, more conservative system response.

Stability Analysis and Design Implications

Analyzing stability using the root locus for a system with Pade-approximated delay follows the standard procedure, but with a critical interpretation of the results. You plot the locus for the rational approximation, find the gain at which loci cross the imaginary axis (the stability boundary), and select a gain that provides adequate damping.

For example, consider a simple plant $G(s)=\frac{1}{s(s+1)}$ with a time delay of $T=1$ second. The open-loop transfer function with a first-order Pade approximation is:

$$G_{ol}(s)=\frac{1}{s(s+1)}\cdot \frac{1-0.5s}{1+0.5s}$$

The open-loop zeros are at $s=+2$ (RHP) and the poles are at $s=0,s=-1,s=-2$. Drawing the root locus for this system will show branches emanating from the three poles. One branch from the pole at $s=-2$ will terminate immediately at the zero at $s=+2$. This branch lies entirely on the real axis and crosses into the RHP. The other branches will show that the system becomes unstable at a finite gain, which is lower than the stability limit for the delay-free system ($G(s)=\frac{1}{s(s+1)}$).

The key design takeaway is that a controller designed for the delay-free system will almost certainly destabilize the real system with delay. You must either redesign the controller (using the approximated model for your root locus sketches) to be less aggressive or employ more advanced control strategies like the Smith Predictor, which is specifically designed to compensate for known time delays.

Common Pitfalls

1. Ignoring the RHP Zero's Effect: The most common mistake is treating the Pade approximation as just "more dynamics" without recognizing that the RHP zero is a non-minimum phase element. This zero imposes fundamental limitations on achievable performance and guarantees that high gain will lead to instability. Always check the location of these approximated zeros.
2. Over-Reliance on a Low-Order Approximation: A first-order Pade approximation is only accurate for relatively low frequencies ($\omega <2/T$). Using it to predict high-frequency dynamics or stability for very large delays can be misleading. For critical applications or larger delays, validate your design with a higher-order approximation or through time-domain simulation of the exact delay model.
3. Forgetting the True Nature of Delay: The root locus of the Pade-approximated system shows a finite number of branches. Remember, the actual delayed system has an infinite number of poles. The approximation helps us find the dominant closed-loop poles, but for large gains, higher-frequency poles from the true system may also become unstable. Your gain margin from the approximated analysis should include a significant safety factor.
4. Tuning Based on the Wrong Model: Avoid the temptation to tune your controller on the delay-free model and then add a delay. The destabilizing effect is nonlinear. Always perform your root locus sketching and gain selection on the combined model that includes the Pade approximation of the delay from the very beginning of the design process.

Summary

• Pure time delays are modeled as $e^{-sT}$ in the transfer function, creating a non-rational characteristic equation that is challenging for classical analysis.
• The Pade approximation (e.g., $e^{-sT}\approx (1-(T/2)s)/(1+(T/2)s)$) replaces the delay with a rational all-pass filter, enabling the use of the root locus method.
• This approximation introduces right-half-plane zeros that attract root locus branches, forcing them into the RHP at lower gains and graphically explaining why delays drastically reduce stability margins.
• The phase lag from any delay reduces phase margin, requiring the use of lower, "detuned" controller gains to maintain closed-loop stability compared to a system without delay. Controller design must be performed on the model that includes the delay approximation.