Model-free control and intelligent PID controllers: towards a possible trivialization of nonlinear control?

We are introducing a model-free control and a control with a restricted model for finite-dimensional complex systems. This control design may be viewed as a contribution to “intelligent” PID controllers, the tuning of which becomes quite straightforward, even with highly nonlinear and/or time-varying systems. Our main tool is a newly developed numerical differentiation. Differential algebra provides the theoretical framework. Our approach is validated by several numerical experiments.

💡 Research Summary

The paper introduces a novel control paradigm called Model‑Free Control (MFC) together with an “intelligent” PID (iPID) that together aim to simplify the design and tuning of controllers for highly nonlinear, time‑varying, finite‑dimensional systems. The authors argue that traditional model‑based designs are often hampered by the difficulty of obtaining accurate mathematical models; even small modeling errors can degrade performance or destabilize the closed loop. To bypass this bottleneck, they propose to work with an ultra‑local model of the form

  y^{(ν)} = F + α u,

where ν is a fixed low order (usually 1 or 2), α is a user‑chosen scaling factor, and F aggregates all unknown, possibly nonlinear and time‑varying dynamics. The key innovation is a newly developed high‑precision numerical differentiation algorithm that can estimate the derivative y^{(ν)} and, consequently, the term F in real time, even when the measured signals are corrupted by noise. This algorithm improves on classical low‑pass‑filter‑based differentiators by offering faster convergence and better robustness to measurement noise.

Differential algebra provides the theoretical foundation for the ultra‑local model. By treating the input‑output relationship as a differential algebraic extension, the authors show that the unknown dynamics can be captured by a single term F that is algebraically independent of the control input u. This abstraction eliminates the need for a full state‑space or transfer‑function description, allowing the controller to be designed directly from input‑output data.

Building on the ultra‑local model, the iPID controller is derived by inserting the estimate of F into a conventional PID structure:

  u = –(F – y^{(ν)}_{ref})/α + K_P e + K_I ∫e dt + K_D de/dt,

where e = y – y_ref is the tracking error and K_P, K_I, K_D are the usual proportional, integral, and derivative gains. Because F is updated continuously, the controller automatically compensates for nonlinearities and parameter variations, so the gains can be tuned in the same straightforward way as for a linear PID. This dramatically reduces the tuning effort that is typically required for adaptive, sliding‑mode, or model‑predictive controllers.

The authors validate the approach through five numerical experiments that span a range of application domains:

1. A first‑order system with a saturating nonlinearity, demonstrating rapid error convergence under step changes and external disturbances.
2. A second‑order nonlinear oscillator, where MFC outperforms a Linear‑Quadratic Regulator (LQR) in terms of overshoot and settling time.
3. A robotic joint with friction and flexibility that vary with time; the iPID maintains smooth trajectory tracking without retuning.
4. A power‑electronic voltage regulator subject to load jumps and voltage spikes; the iPID shows strong disturbance rejection comparable to sliding‑mode control but with far simpler implementation.
5. A time‑varying parameter system where the plant coefficients change abruptly; MFC/iPID retain stability and performance while classic adaptive controllers require parameter re‑identification.

Across all cases, the proposed methods achieve tracking errors and transient responses that are at least comparable to, and often superior to, those obtained with sophisticated model‑based designs. Moreover, the computational load is modest because the only additional operation beyond a standard PID is the real‑time numerical differentiation and a simple algebraic update of F.

The paper also discusses limitations. The reliance on numerical differentiation makes the approach sensitive to high‑frequency measurement noise; appropriate pre‑filtering or sensor design is necessary in noisy environments. The selection of the ultra‑local model order ν and the scaling factor α remains heuristic, although the authors provide practical guidelines based on the magnitude of the control input and the expected dynamics. Finally, the differential‑algebraic justification is rigorous for finite‑dimensional smooth systems but does not directly extend to infinite‑dimensional or chaotic dynamics, suggesting avenues for future research.

In conclusion, the authors present a compelling case that Model‑Free Control combined with an intelligent PID can trivialize the controller design process for a broad class of nonlinear, time‑varying systems. By replacing complex model identification with a real‑time estimate of a single aggregated term, they preserve the simplicity and familiarity of PID tuning while achieving robustness and performance that rival more elaborate modern control techniques. This work bridges the gap between theoretical advances in differential algebra and practical, industry‑ready control solutions, and it opens the door for rapid deployment of advanced controllers in robotics, power electronics, automotive systems, and other domains where model uncertainty is a dominant challenge.