European Control Conference (ECC) --- July 7-10, 2026, Reykjavík, Iceland Air Supply Control for Proton Exchange Membrane Fuel Cells without Explicit Modeling

Meziane Ait Ziane, Michel Zasadzinski, Cédric Join, Michel Fliess · 2026

Read European Control Conference (ECC) --- July 7-10, 2026, Reykjavík, Iceland Air Supply Control for Proton Exchange Membrane Fuel Cells without Explicit Modeling by Meziane Ait Ziane, Michel Zasadzinski, Cédric Join, Michel Fliess (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-11433.txt.

Introduction

Ait Ziane and three colleagues control the air supply of proton exchange membrane fuel cells without explicit modeling. Their model-free strategy regulates oxygen ratio, adapting live to operating points with low computing load, an attractive mix for plant use. No detailed plant model is needed, and the controller still tracks its targets. Simulations across constant and variable oxygen ratio scenarios with two current profiles show good performance. Robustness is tested against big parameter changes in the air supply system.

Catalog record

CREATORMeziane Ait Ziane, Michel Zasadzinski, Cédric Join, Michel Fliess
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE19707
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.11433

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

European Control Conference (ECC) --- July 7-10, 2026, Reykjavík, Iceland Air Supply Control for Proton Exchange Membrane Fuel Cells without Explicit Modeling Meziane Ait Ziane, Michel Zasadzinski, Cédric Join, Michel Fliess Abstract Our objective is to study the performance and robustness of the model-free strategy for controlling the oxygen stoichiometry of a fuel cell air supply system with a proton exchange membrane. After reviewing the literature on modeling and control of this process, the model-free approach appears to be a good candidate because, on the one hand, it allows straightforward real-time adaptation to track operating points and, on the other hand, it requires a low computational burden, which is attractive for industrial applications. Numerical simulations for two scenarios (constant and variable oxygen stoichiometry) with two current profiles reveal satisfactory performance of the model-free control law. The robustness is addressed by considering significant variations in the parameters of the proton exchange membrane air supply system. I INTRODUCTION One potential technological solution for decarbonizing the transportation sector is to replace combustion engines with electrical motors supplied by fuel cells. Among the various existing technologies, the Proton Exchange Membrane Fuel Cell ( PEMFC ) is particularly well suited to automotive applications, as it operates at low temperatures (60 to 80°C) and is resilient to load variations [ 35 , 5 ] . A PEMFC is viewed as a complex system due to the interactions of several physical phenomena (electrical, thermal, chemical, …). It is mandatory to feed the PEMFC with a sufficient amount of hydrogen and oxygen to ensure the desired electrical power and to avoid performance degradation. Thus, air supply control is one of the key points in the proper operation of a PEMFC. Since air is supplied by a compressor, both the fuel cell and the compressor (referred to as the PEMFC air-feed system in the literature) must be taken into account [ 9 ] . The behavior of the PEMFC system is highly nonlinear. A PEMFC model including the compressor and oriented control has been initially proposed in [ 26 ] . From this model, an air-feed one with 5 states is presented in [ 21 ] , with 4 states in [ 29 ] , and with 3 states in [ 38 ] . These models have been used widely in the literature for control purposes. The design of all control laws cited in this paragraph is explicitly based on the aforementioned PEMFC air-feed models. A sliding mode control approach is designed in [ 23 ] with state feedback and in [ 38 , 15 ] by using an observer. A controller based on state feedback linearization is proposed in [ 8 ] and in [ 20 ] by adding an extended state observer. A LQR tracking controller based on an observer is synthesized in [ 19 ] . A control law strategy constituted by three parts, called a triple-step controller, is developed in [ 22 ] and [ 36 ] : {1}^{\textrm{st}} part is a PID controller, {2}^{\textrm{nd}} part is a steady-state feedforward controller, and {3}^{\textrm{rd}} is a steady-state feedback. It is challenging to obtain a model that accurately describes the actual behavior of the PEMFC air-feed system, as the PEMFC stack is difficult to model and to obtain the exact values of the physical parameters. The control laws mentioned in this paragraph are promising strategies, but they have a limitation on their efficiency due to the quality of the model used. PID are widely proposed in the literature to control PEMFC air-feed system. PID controllers without gain adaptation are proposed: feedforward PID in [ 7 ] , robust PID in [ 30 ] and PID based on {1}^{\textrm{st}} order approximation in [ 11 ] . The gains of PID controllers are adapted by using a fuzzy logic approach in [ 4 ] and [ 25 ] , while this adaptation is made using neural techniques in [ 17 ] , [ 33 ] , and [ 10 ] . Both fuzzy logic and neuronal approaches are used to adapt PID gains in [ 14 ] , [ 37 ] and [ 34 ] . Considering the high nonlinearity of the model, PID parameter tuning seems required to control the PEMFC air-feed system over a wide operating range. An artificial intelligence approach is also applied to design a controller for PEMFC air-feed systems without including PID in [ 32 ] . An adaptive B-spline neural network is applied in [ 27 ] , an adaptive observer is added to the neural network in [ 31 ] , and distributed deep reinforcement learning is employed in [ 18 ] . For real-time implementation considerations, the use of fuzzy logic and/or neural network techniques to perform either gains adaptation of PID controllers or the design of artificial intelligence controllers is computationally expensive. The Model-Free Control ( MFC ) developed in [ 12 , 13 ] can be considered an interesting alternative to the approaches mentioned above: the mathematical model is not necessary to design the control law, which is adapted using an ultra-local model in a straightforward manner. Furthermore, its implementation is simpler than those mentioned above and requires very few parameters. The aim of this paper is to investigate the performance and robustness of the MFC strategy on a PEMFC air-feed system model under several scenarios. The outline of this paper is as follows. MFC design is briefly recalled in Sect. II . The PEMFC air-feed system model used to test the designed controller in this paper and the control objective are given in Section III . The simulation results are presented in Sect. IV : the two scenarios considered in this paper are described in Sect. IV-A ; the simulations for both constant and variable desired oxygen stoichiometry, along with robustness analysis, are conducted in Sect. IV-B and IV-C , respectively. A conclusion with discussion is given in Sect. V . II SHORT RECALL ON MODEL-FREE CONTROL We only deal here with SISO systems. Elementary functional analysis and differential algebra show [ 12 ] , as well as concrete experiments, that most, or at least many, systems encountered in practice can be approximated by the ultra-local model \dot{y}=\mathcal{F}+\alpha u (1) where the control and output variables are respectively u and y , \mathcal{F} is a quantity that subsumes the poorly known system structure and the disturbances. The constant \alpha\in\mathbb{R} is chosen by the practitioner such that the three terms in ( 1 ) are of the same magnitude: \alpha does not need to be precisely estimated. The following data-driven real-time estimate \mathcal{F}_{\text{est}}(t) of \mathcal{F}(t) is given by {\small\mathcal{F}_{\text{est}}(t)=-\frac{6}{\tau^{3}}\int_{t-\tau}^{t}\left[(\tau-2\sigma)y(\sigma)+\alpha\sigma(\tau-\sigma)u(\sigma)\right]d\sigma} Close the loop with the following intelligent proportional controller , or iP , u=-\frac{\mathcal{F}_{\text{est}}-\dot{y}^{\ast}+K_{p}e}{\alpha} (2) where y^{\ast} is the reference trajectory, e=y-y^{\ast} is the tracking error, K_{p} is a tuning gain. Equations ( 1 ) and ( 2 ) yield \dot{e}+K_{p}e=\mathcal{F}-\mathcal{F}_{\text{est}}. If the estimate \mathcal{F}_{\text{est}} is “good”, i.e. , \mathcal{F}-\mathcal{F}_{\text{est}}\approx 0 , then \lim_{t\to+\infty}e(t)\approx 0 if and only if, K_{p}>0 . This local stability property proves that the tuning of K_{p} is straightforward. This is a major difference with the classic gain tuning for PIs and PIDs (see, e.g. , [ 13 ] ). III PEMFC AIR FEED SYSTEM MODELING

Read the complete text