Questions: Model Uncertainty and Robust Stability Analysis

Question 1

A controller is designed for a simplified plant model that ignores a resonance at 500 Hz. The closed-loop system has high gain at 500 Hz. What is the primary robustness concern?

• A. The controller will reject disturbances less effectively because its bandwidth is too narrow
• B. The unmodeled resonance adds phase lag and gain at exactly the frequency where the controller maintains high gain — the actual plant could be pushed to instability
• C. The simplified model underestimates plant gain at low frequencies, reducing steady-state performance
• D. High gain at 500 Hz improves noise rejection, so this is beneficial rather than dangerous

Answer: (B)

Unmodeled high-frequency dynamics — resonances, delays, flexible modes — represent multiplicative uncertainty that grows at high frequencies. If the controller maintains high loop gain at frequencies where the true plant differs significantly from the model (large ΔG), the closed-loop system can be driven unstable even though the nominal model appears well-controlled. The robust stability condition |T(jω)| ≤ 1/W(ω) is violated when the complementary sensitivity T is large at frequencies where the uncertainty bound W is large. This is why controllers should roll off gain well before unmodeled dynamics become significant.

Question 2

What does the multiplicative uncertainty model G_true = G_nominal × (1 + ΔG) capture that gain margin alone cannot?

• A. It captures pure gain changes at the crossover frequency, which is exactly what gain margin measures
• B. It captures frequency-dependent uncertainty — the model error can vary in magnitude across frequencies, not just at the gain crossover point
• C. It captures time-domain variations like ramp disturbances that frequency-domain margins miss
• D. It shows how the plant model changes when the controller is tuned more aggressively

Answer: (B)

Gain margin measures how much the loop gain can change at one specific frequency (the phase crossover frequency) before instability. Multiplicative uncertainty W(ω) characterizes how large the model error could be at every frequency simultaneously. This is far richer: a plant might be well-modeled at low frequencies (small W) but have large uncertainty at high frequencies (large W due to unmodeled resonances). Gain margin only gives a scalar guarantee at one frequency and cannot capture this frequency-dependent structure. The robust stability condition |T(jω)| ≤ 1/W(ω) must hold across all frequencies, not just at crossover.

Question 3

Robust stability requires the closed-loop complementary sensitivity function T(jω) to roll off at high frequencies where model uncertainty is large.

• T. True
• F. False

Answer: True

The robust stability condition for multiplicative uncertainty is |T(jω)| ≤ 1/W(ω) at all frequencies. At high frequencies, unmodeled dynamics, parasitic resonances, and pure time delays make W(ω) large — the uncertainty bound grows. For the inequality to hold where W is large, T must be small. This means the closed-loop transfer function must roll off at high frequencies, limiting bandwidth. This is not a restriction imposed by robust stability analysis on top of good design — it is a quantitative expression of the intuition that controllers should not be tuned too aggressively into frequency ranges where the model is unreliable.

Question 4

A system with large gain and phase margins is very likely to remain stable for any bounded uncertainty in the plant model.

• T. True
• F. False

Answer: False

Gain and phase margins provide conservative stability guarantees only for specific types of perturbation — simultaneous gain and phase changes at specific frequencies (crossover frequencies). They do not protect against all bounded uncertainties. For example, a plant with excellent gain and phase margins can still be destabilized by an unmodeled high-frequency resonance if the controller has high gain at that resonance frequency. Complete robustness guarantees require frequency-domain conditions like the complementary sensitivity bound, which must hold across all frequencies, not just at crossover points.

Question 5

Explain why a controller designed for excellent nominal performance can still fail to be robustly stable, and what the multiplicative uncertainty framework reveals about this failure mode.

Model answer: Nominal performance is optimized against the design model, not the true plant. A controller can achieve high bandwidth, fast disturbance rejection, and low steady-state error — all measured on the model — while maintaining large loop gain at frequencies where the model is unreliable. The multiplicative uncertainty framework reveals this: if |ΔG(jω)| is large at some frequency (e.g., an unmodeled resonance), but |T(jω)| is also large there (the controller fights hard at that frequency), then the closed-loop system can be destabilized by the real plant's actual behavior at that frequency. The robust stability condition |T(jω)| ≤ 1/W(ω) makes the tradeoff explicit: good nominal performance at frequency ω requires high T, but robustness to uncertainty at ω requires low T. These demands are in direct tension, and the framework quantifies exactly where they conflict.

The key insight is that nominal performance and robustness trade off in the frequency domain. A controller that is good at rejecting disturbances (high sensitivity function suppression) at some frequency necessarily has high complementary sensitivity T at nearby frequencies. Where uncertainty W is large, this is dangerous. Robust control design is the discipline of managing this tradeoff explicitly.