betaFeedback

WorksEngineering

Air supply control for proton exchange membrane fuel cells without explicit modeling

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.

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 [1, 2].

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 [3].

The behavior of the PEMFC system is highly nonlinear. A PEMFC model including the compressor and oriented control has been initially proposed in [4]. From this model, an air-feed one with 5 states is presented in [5], with 4 states in [6], and with 3 states in [7]. 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 [8] with state feedback and in [7, 9] by using an observer. A controller based on state feedback linearization is proposed in [10] and in [11] by adding an extended state observer. A LQR tracking controller based on an observer is synthesized in [12]. A control law strategy constituted by three parts, called a triple-step controller, is developed in [13] and [14]: 1st part is a PID controller, 2nd part is a steady-state feedforward controller, and 3rd 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 [15], robust PID in [16] and PID based on 1st order approximation in [17]. The gains of PID controllers are adapted by using a fuzzy logic approach in [18] and [19], while this adaptation is made using neural techniques in [20], [21], and [22]. Both fuzzy logic and neuronal approaches are used to adapt PID gains in [23, 24] and [25]. 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 [26]. An adaptive B-spline neural network is applied in [27], an adaptive observer is added to the neural network in [28], and distributed deep reinforcement learning is employed in [29]. 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 [30, 31] 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.

Short Recall on Model-Free Control

We only deal here with SISO systems. Elementary functional analysis and differential algebra show [30], as well as concrete experiments, that most, or at least many, systems encountered in practice can be approximated by the ultra-local model

y˙=F+αu(1)\dot{y} = \mathcal{F} + \alpha u \tag*{(1)}

where the control and output variables are respectively uu and yy, F\mathcal{F} is a quantity that subsumes the poorly known system structure and the disturbances. The constant α∈R\alpha\in\mathbb{R} is chosen by the practitioner such that the three terms in (1) are of the same magnitude: α\alpha does not need to be precisely estimated. The following data-driven real-time estimate Fest(t)\mathcal{F}_{\mathrm{est}}(t) of F(t)\mathcal{F}(t) is given by

Fest(t)=−6τ3∫t−τt[(τ−2σ)y(σ)+ασ(τ−σ)u(σ)]dσ\mathcal{F}_{\mathrm{est}}(t)=-\frac{6}{\tau^{3}}\int_{t-\tau}^{t}\left[(\tau-2\sigma)y(\sigma)+\alpha\sigma(\tau-\sigma)u(\sigma)\right]\mathrm{d}\sigma

Close the loop with the following intelligent proportional controller, or iP,

u=−Fest−y˙∗+Kpeα(2)u=-\frac{\mathcal{F}_{\mathrm{est}}-\dot{y}^{*}+K_{p}e}{\alpha} \tag*{(2)}

where y∗y^{*} is the reference trajectory, e=y−y∗e=y-y^{*} is the tracking error, KpK_{p} is a tuning gain. Equations (1) and (2) yield

e˙+Kpe=F−Fest.\dot{e}+K_{p}e=\mathcal{F}-\mathcal{F}_{\mathrm{est}}.

If the estimate Fest\mathcal{F}_{\mathrm{est}} is “good”, i.e., F−Fest≈0\mathcal{F}-\mathcal{F}_{\mathrm{est}}\approx0, then lim⁡t→+∞e(t)≈0\lim_{t\to+\infty}e(t)\approx0 if and only if, Kp>0K_{p}>0. This local stability property proves that the tuning of KpK_{p} is straightforward. This is a major difference with the classic gain tuning for PIs and PIDs (see, e.g., [31]).

PEMFC Air Feed System Modeling and Control Objective

PEMFC air-feed system model

A PEMFC model oriented control has been initially proposed in [4]. From this model, Suh [6] has proposed an air-feed model, with 4 states, has been used widely in the literature: in [8] and [9] for sliding mode control, in [32] for fault detection with observer and in [28] for both control and state estimation using neural networks.

This model has been validated on test bench in [33] and [34] and is described by

x˙1=c1(x4−x1−x2−c2)−c7ξ−c3x1c4x1+c5x2+c6c17x1+x2+c2−c11,x˙2=c8(x4−x1−x2−c2)−c3x2c4x1+c5x2+c6c17x1+x2+c2−c11,x˙3=−c9x3−c10[(x4c11)c12−1]+c13u,x˙4=c14[1+c15[(x4c11)c12−1]]×[c21x3−c16(x4−x1−x2−c2)].(3)\begin{aligned} \dot{x}_{1}={}&c_{1}(x_{4}-x_{1}-x_{2}-c_{2})-c_{7}\xi-\frac{c_{3}x_{1}}{c_{4}x_{1}+c_{5}x_{2}+c_{6}}c_{17}\sqrt{x_{1}+x_{2}+c_{2}}-c_{11},\\ \dot{x}_{2}={}&c_{8}(x_{4}-x_{1}-x_{2}-c_{2})-\frac{c_{3}x_{2}}{c_{4}x_{1}+c_{5}x_{2}+c_{6}}c_{17}\sqrt{x_{1}+x_{2}+c_{2}}-c_{11},\\ \dot{x}_{3}={}&-c_{9}x_{3}-c_{10}\left[\left(\frac{x_{4}}{c_{11}}\right)^{c_{12}}-1\right]+c_{13}u,\\ \dot{x}_{4}={}&c_{14}\left[1+c_{15}\left[\left(\frac{x_{4}}{c_{11}}\right)^{c_{12}}-1\right]\right]\\ &\times[c_{21}x_{3}-c_{16}(x_{4}-x_{1}-x_{2}-c_{2})]. \tag*{(3)} \end{aligned}

where

  • x1=PO2x_{1}=P_{O_{2}} is the oxygen partial pressure, x2=PN2x_{2}=P_{N_{2}} is the nitrogen partial pressure, x3=ωcpx_{3}=\omega_{cp} is the compressor angular speed and x4=Psmx_{4}=P_{sm} is the supplied manifold pressure,

  • ξ\xi is the fuel stack current which is considered as measured disturbance in the sequel,

  • uu is the motor current and is the control input,

  • cic_{i} with i=1,…,21i=1,\ldots,21 are PEMFC and air-feed compressor parameters given in Table 1, where physical parameters values are given in Table II in [8].

c1=RTfkca,inxO2VcaMO2(1+ωatm)c_{1}=\dfrac{RT_{f}k_{ca,\mathrm{in}}x_{O_{2}}}{V_{ca}M_{O_{2}}(1+\omega_{atm})}c2=psatc_{2}=p_{sat}
c3=RTffcVcac_{3}=\dfrac{RT_{f}f_{c}}{V_{ca}}c4=MO2c_{4}=M_{O_{2}}
c5=MN2c_{5}=M_{N_{2}}c6=Mvpsatc_{6}=M_{v}p_{sat}
c7=RTffckca,inVca4Fc_{7}=\dfrac{RT_{f}f_{c}k_{ca,\mathrm{in}}}{V_{ca}4F}c8=RTffckca,in(1−xO2)VcaMN2(1+ωatm)c_{8}=\dfrac{RT_{f}f_{c}k_{ca,\mathrm{in}}(1-x_{O_{2}})}{V_{ca}M_{N_{2}}(1+\omega_{atm})}
c9=fJcpc_{9}=\dfrac{f}{J_{cp}}c10=cv−cvcprρaCppatm2πJcpηcpc_{10}=\dfrac{c_{v-c}v_{cpr}\rho_{a}C_{p}p_{atm}}{2\pi J_{cp}\eta_{cp}}
c11=patmc_{11}=p_{atm}c12=γ−1γc_{12}=\dfrac{\gamma-1}{\gamma}
c13=ηcmktJcpc_{13}=\dfrac{\eta_{cm}k_{t}}{J_{cp}}c14=RTatmMaVsmc_{14}=\dfrac{RT_{atm}}{M_{a}V_{sm}}
c15=1ηcpc_{15}=\dfrac{1}{\eta_{cp}}c16=kca,inc_{16}=k_{ca,\mathrm{in}}
c17=kca,outc_{17}=k_{ca,\mathrm{out}}c18=ηcmktc_{18}=\eta_{cm}k_{t}
c19=kca,inxO2(1+ωatm)c_{19}=\dfrac{k_{ca,\mathrm{in}}x_{O_{2}}}{(1+\omega_{atm})}c20=nMO24Fc_{20}=\dfrac{nM_{O_{2}}}{4F}
c21=ηv−cvcprrpa2πc_{21}=\dfrac{\eta_{v-c}v_{cpr}}{r_{pa}2\pi}×\times

Table 1. Constants c1c_{1} to c21c_{21}

SymbolParameterValue
ffMotor friction+20%
ktk_tMotor constant−5%
ηcp\eta_{cp}Compressor efficiency−10%
ηcm\eta_{cm}Motor mechanical efficiency−20%
kca,outk_{ca,out}Cathode outlet constant+10%
TatmT_{atm}Atmospheric temperature+10%
VcaV_{ca}Cathode volume+10%
VsmV_{sm}Supply manifold volume−10%
TfcT_{fc}Stack temperature, °C+12%

Table II. Parameters uncertainties

Remark 1: In this paper, the model described by (3) is not used to design the control law, but to emulate the behavior of the PEMFC air-feed system.

Control objective

Using measurements y1y_{1} and y2y_{2} given by

y1=x1+x2+c2+w1,(4)y_{1}=x_{1}+x_{2}+c_{2}+w_{1}, \tag*{(4)}
y2=x4+w2,(5)y_{2}=x_{4}+w_{2}, \tag*{(5)}

where w1w_{1} and w2w_{2} are independent zero mean Gaussian white noises.

The objective is to regulate the excess ratio between the oxygen flow rate entering the PEMFC and the oxygen flow rate consumed by the fuel cell due/from to the electrochemical reaction at a desired value. This excess ratio between flow rates is known as “oxygen stoichiometry (λO2\lambda_{\mathrm{O}_2})” and is defined by:

λO2=c19(y2−y1)c20ξ\lambda_{\mathrm{O}_2} = \frac{c_{19}(y_2-y_1)}{c_{20}\xi}

where c19c_{19} and c20c_{20} are known constants.

The desired oxygen stoichiometry (λO2∗\lambda_{\mathrm{O}_2}^{\ast}), i.e. the setpoint, can be set as follows 2≤λO2∗≤2.52 \le\lambda_{\mathrm{O}_2}^{\ast} \le2.5 to ensure sufficient oxygen flow for the electrochemical reaction. It is important to note that a value λO2≤1\lambda_{\mathrm{O}_2} \le1 can lead to a starvation fault which can generate some irreversible damages to the PEMFC stack. Furthermore, a value of λO2>2.5\lambda_{\mathrm{O}_2} > 2.5 may lead to an evacuation of water generated by the PEMFC, which causes the membrane drying and increases the compressor’s energy consumption. In the literature, there exist two main ways to define the desired oxygen stoichiometry λO2∗\lambda_{\mathrm{O}_2}^{\ast}

  1. λO2∗\lambda_{\mathrm{O}_2}^{\ast} is chosen as a constant 2≤λO2∗≤2.52 \le\lambda_{\mathrm{O}_2}^{\ast} \le2.5 despite variations of ξ\xi (see [28] and [35]),

  1. λO2∗\lambda_{\mathrm{O}_2}^{\ast} is defined as a polynomial function of the stack current and given by (see [8] and [13])

λO2∗=5×10−8ξ3−2.87×10−5ξ2+2.23×10−3ξ+2.5.(6)\lambda_{\mathrm{O}_2}^{\ast} = 5 \times10^{-8}\xi^3 - 2.87 \times10^{-5}\xi^2 + 2.23 \times10^{-3}\xi+ 2.5. \tag*{(6)}

Simulations Results

Scenarios

Two scenarios are considered for the simulations:

  1. the 1{sup}st` is devoted to regulating the oxygen stoichiometry λO2\lambda_{\mathrm{O}_2} at a fixed value λO2∗=2.2\lambda_{\mathrm{O}_2}^{\ast}=2.2,

  2. the 2{sup}nd` scenario is dedicated to regulating the oxygen stoichiometry λO2\lambda_{\mathrm{O}_2} at a variable value of λO2∗\lambda_{\mathrm{O}_2}^{\ast} which is given by (6).

For these two scenarios, two different current profiles of ξ\xi are considered. In the 1{sup}st current profile, the variations of $\xi$ can be regarded as small (see {xref}Fig. 1), while the 2{sup}nd current profile, which has significant variations of ξ\xi, is taken into account (see Figure 2).

2nd current profile

Figure 2. 2{sup}nd` current profile

In order to analyze the robustness of the iPiP controller, two cases are considered:

  • the physical parameters of PEMFC system in (3) given in table 2 in [8] are considered for nominal case,

  • some physical parameters of PEMFC system in (3) given in table 2 in [8] are modified and given in Table II for uncertain case.

The following notations are retained: superscript nn for the nominal case and superscript uu for the uncertain one.

Constant desired oxygen stoichiometry

  1. 1{sup}st current profile: {xref}Fig. 3 shows the control of oxygen stoichiometry when 1{sup}st profile shown in Figure 1 is applied. λO2\lambda_{\mathrm{O}_2} is restored to λO2∗\lambda_{\mathrm{O}_2}^{\ast} in 5 s for all six current changes for the nominal case, see Figure 3a. For the uncertain case, the iPiP controller performs very well with regard to these uncertainties in the PEMFC system parameters, ensuring a restoration time of λO2\lambda_{\mathrm{O}_2} to λO2∗\lambda_{\mathrm{O}_2}^{\ast} in 6.5 s.

1st current profile

Figure 1. 1{sup}st` current profile

The differences between the control inputs unu^n and uuu^u shown in Figure 3b are due to all the uncertainties in the PEMFC parameters and confirms that the controller is capable of tolerating these uncertainties. However, the differences between y1ny_1^n, y1uy_1^u and y2ny_2^n, y2uy_2^u, see Figure 3c, are only due to uncertainties associated with the stack parameters (kca,ink_{ca,in}, kca,outk_{ca,out}, TatmT_{atm}, VcaV_{ca}, VsmV_{sm} and TfcT_{fc}), but not to uncertainties related to the compressor parameters (ff, ktk_t, ηcp\eta_{cp} and ηcm\eta_{cm}).

  1. 2{sup}nd current profile: {xref}Fig. 4 shows the oxygen stoichiometry control when the 2{sup}nd profile is applied. This profile presents higher current variations than the 1{sup}st one. For the nominal case, the $iP$ controller resets $\lambda_{\mathrm{O}_2}$ to $\lambda_{\mathrm{O}_2}^{\ast}$ between 2 s and 10 s depending on the magnitude of the variation in current $\xi$. With regard to the uncertain case, it should be noted that there is no significant difference; the restoration time is 1.5 s longer than in the nominal case as shown in {xref}Fig. 4a. The same comments can be made for {xref}Fig. 4b and {xref}Fig. 4c as for {xref}Fig. 3b and {xref}Fig. 3c`.

The simulation results show that the iPiP controller is robust with respect to parameter uncertainties and ensures satisfactory closed-loop behaviors for the two scenarii with constant desired oxygen stoichiometry.

Variable desired oxygen stoichiometry

In this Section, the desired oxygen stoichiometry λO2⋆\lambda_{O_2}^{\star} is variable, so the control objective is to regulate λO2\lambda_{O_2} to λO2⋆\lambda_{O_2}^{\star} even with PEMFC parameter uncertainties under the two above-mentioned current profiles.

  1. 1st current profile: For the 1st profile 2≤λO2⋆≤2.22 \leq\lambda_{O_2}^{\star} \leq2.2. The iP controller ensures both disturbance rejection and λO2⋆\lambda_{O_2}^{\star} tracking in 5 s for the nominal case and in 6 s for the uncertain case, as shown in Figure 5a. The robustness of iP controller is achieved since the responses are close in nominal and uncertain cases in Figure 5a. The control input and the measurements for both nominal and uncertain cases are shows in Fig. 5b and Fig. 5c, respectively. For both constant and variable desired oxygen stoichiometry, the transient behavior is closed in the two cases (see the settling times and the peaks in Fig. 3a and Fig. 5a). However, the difference between the control inputs and the measurements in both cases is mainly explained by the difference in the setpoints (see Fig. 3b, Fig. 3c, Fig. 5b, and Fig. 5c).

  1. 2nd current profile: For the 2nd profile 1.9≤λO2⋆≤2.51.9 \leq\lambda_{O_2}^{\star} \leq2.5. Fig. 6 shows the oxygen stoichiometry control when the 2nd profile is applied. Like in the 1st scenario, the iP controller resets λO2\lambda_{O_2} to λO2⋆\lambda_{O_2}^{\star} between 2 s and 10 s depending on the magnitude of the variation in current ξ\xi for the nominal case, and there is no significant difference for the uncertain case (the tracking time is 1.5 s longer than in the nominal case as shown in Fig. 6a). The same comments can be made for Fig. 6b and Fig. 6c as for Fig. 5b and Fig. 5c.

The above simulations illustrate the fact that the iP controller ensures good reference tracking and robustness against parameter uncertainties for both scenarios with a variable desired oxygen stoichiometry.

Conclusions

The results obtained with the iP controller are satisfactory in simulation with a model widely used in the literature and allow us to proceed to the next step: implementing the iP controller in a real test bench. This real-time implementation allows us to compare the proposed approach with those mentioned in the Introduction. This comparison will be carried out in terms of both the performance and robustness of the control laws and the computational burden in order to propose effective control strategies for this complex system.

Another perspective of this work is to address PEM fuel cell energy management [36] and fault-tolerant control and diagnosis using model-free approaches [37, 38], particularly to deal with sensor faults that affect λO2\lambda_{O_2} in (5).

Constant $\lambda_{O_2}$ for 1st current profile

(a) Constant λO2\lambda_{O_2} for 1st current profile

$u$ for 1st current profile

(b) uu for 1st current profile

$y_1$ and $y_2$ for 1st current profile

(c) y1y_1 and y2y_2 for 1st current profile

Figure 3. Constant desired oxygen stoichiometry λO2⋆\lambda_{O_2}^{\star} with and without uncertainties: 1st current profile

Constant $\lambda_{O_2}$ for 2nd current profile

(a) Constant λO2\lambda_{O_2} for 2nd current profile

$u$ for 2nd current profile

(b) uu for 2nd current profile

$y_1$ and $y_2$ for 2nd current profile

(c) y1y_1 and y2y_2 for 2nd current profile

Figure 4. Constant desired oxygen stoichiometry λO2⋆\lambda_{O_2}^{\star} with and without uncertainties: 2nd current profile

Variable $\lambda_{O_2}$ for 1st current profile

(a)

$u$ for 1st current profile

(b)

$y_1$ and $y_2$ for 1st current profile

(c)

Figure 5. Variable desired oxygen stoichiometry λO2∗\lambda_{O_2}^{\ast} with and without uncertainties: 1st current profile

Variable $\lambda_{O_2}$ for 2nd current profile

(a)

$u$ for 2nd current profile

(b)

$y_1$ and $y_2$ for 2nd current profile

(c)

Figure 6. Variable desired oxygen stoichiometry λO2∗\lambda_{O_2}^{\ast} with and without uncertainties: 2nd current profile

References

  1. [1]M. Yue, H. Lambert, E. Pahon, R. Roche, S. Jemei, and D. Hissel, “Hydrogen energy systems: a critical review of technologies, applications, trends and challenges,” Renewable and Sustainable Energy Reviews, vol. 146, p. ID 111180, 2021.
  2. [2]J. Aubry, N. Yousfi Steiner, S. Morando, N. Zerhouni, and D. Hissel, “Fuel cell diagnosis methods for embedded automotive applications,” Energy Reports, vol. 8, pp. 6687–6706, 2022.DOI
  3. [3]X. Chen, B. Gu, W. Feng, J. Tan, X. Kong, S. Li, Y. Chen, and Z. Wan, “Research on control strategy of PEMFC air supply system for power and efficiency improvement,” Energy, vol. 304, p. ID 132100, 2024.
  4. [4]J. Pukrushpan, A. Stefanopoulou, and H. Peng, Control of Fuel Cell Power Systems: Principles, Modeling, Analysis and Feedback Design. Principles, Modeling, Analysis and Feedback Design. London: Springer Science & Business Media, 2004.
  5. [5]J. Liu, Y. Gao, X. Su, M. Wack, and L. Wu, “Disturbance-observer-based control for air management of PEM fuel cell systems via sliding mode technique,” IEEE Trans. Control Syst. Techn., vol. 27, pp. 1129–1238, 2019.
  6. [6]K. Suh, “Modeling, analysis and control of fuel cell hybrid power systems,” Ph.D. dissertation, University of Michigan, Ann Arbor, USA, 2005.
  7. [7]Y. Zhu, J. Zou, S. Li, and C. Peng, “An adaptive sliding mode observer based near-optimal OER tracking control approach for PEMFC under dynamic operation condition,” Int. J. of Hydrogen Energy, vol. 47, pp. 1157–1171, 2022.
  8. [8]I. Matraji, S. Lagrouche, S. Jemei, and M. Wack, “Robust control of the PEM fuel cell air-feed system via sub-optimal second order sliding mode,” Applied Energy, vol. 104, pp. 945–957, 2013.DOI
  9. [9]S. Lagrouche, M. Narouche, F. Ahmed, and Y. Chitour, “Control of PEMFC air-feed system using Lyapunov-based robust and adaptive higher order sliding mode control,” IEEE Trans. Control Syst. Techn., vol. 23, pp. 1594–1601, 2015.DOI
  10. [10]J. Chen, Z. Liu, P. Wang, Q. Ouyang, and H. Su, “Optimal oxygen excess ratio control for PEM fuel cells,” IEEE Trans. Control Syst. Techn., vol. 26, pp. 1711–1721, 2018.
  11. [11]M. Li, H. Yin, T. Ding, J. Zhao, J. Gao, and H. Chen, “Air flow rate and pressure control approach for the air supply subsystems in PEMFCs,” ISA Transactions, vol. 128, pp. 624–634, 2022.DOI
  12. [12]M. Li, J. Lu, Y. Hu, and J. Gao, “Exogenous reactive controller design of PEM fuel cell,” in Proc. IFAC Conf. on Engine and Powertrain Control, Simulation and Modeling, Changchun, P.R. China, 2018.
  13. [13]Y. Ma, F. Zhang, J. Gao, H. Chen, and T. Shen, “Oxygen excess ratio control of PEM fuel cells using observer-based nonlinear triple-step controller,” Int. J. of Hydrogen Energy, vol. 45, pp. 29705–29717, 2020.DOI
  14. [14]C. Zhang, Z. Guan, X. Kong, J. Lin, and Y. Hu, “Triple-step control for a proton exchange membrane fuel cell air supply system,” in Proc. IFAC Conf. on Engine and Powertrain Control, Simulation and Modeling, Dalian, P.R. China, 2024.
  15. [15]X. Bian, G. Chen, N. Li, L. Tong, R. Yang, and B. Shen, “Feed-forward PID-based cooperative control strategy for air mass flow and pressure in PEMFC systems,” Energy Science & Engineering, 2025, https://doi.org/10.1002/ese3.70229.
  16. [16]F. Wang and C. Ko, “Multivariable robust PID control for a PEMFC system,” Int. J. of Hydrogen Energy, vol. 35, pp. 10 437–10 445, 2010.DOI
  17. [17]S. Divi, S. Sonawane, and S. Das, “Uncertainty analysis of transfer function of proton exchange membrane fuel cell and design of PI/PID controller for supply manifold pressure control,” Indian Chemical Engineer, vol. 61, pp. 138–152, 2019.
  18. [18]M. Aliashgary, “Control of PEM fuel cell systems using interval type-2 fuzzy PID approach,” Fuel Cells, vol. 18, pp. 449–456, 2018.
  19. [19]K. Ou, Y. Wang, Z. Li, Y. Shen, and D. Xuan, “Feedforward fuzzy-PID control for air flow regulation of PEM fuel cell system,” Int. J. of Hydrogen Energy, vol. 40, pp. 11 686–11 695, 2015.DOI
  20. [22]C. Damour, M. Benne, C. Lebreton, J. Deseure, and B. Grondin-Perez, “Real-time implementation of a neural model-based self-tuning PID strategy for oxygen stoichiometry control in PEM fuel cell,” Int. J. of Hydrogen Energy, vol. 39, pp. 12 819–12 825, 2014.DOI
  21. [23]M. Gheisanejad, J. Boudjadar, and M. Khooban, “A new adaptive type-II fuzzy-based deep reinforcement learning control: fuel cell air-feed sensors control,” IEEE Sensors J., vol. 19, pp. 9081–9089, 2019.DOI
  22. [24]J. Zhou, Y. Liao, J. Liu, Y. Xue, and S. Xue, “Deep reinforcement learning guided cascade control for air supply of polymer exchange membrane fuel cell,” Energy Technology, vol. 9, p. ID 2100149, 2021.
  23. [25]P. Yin, J. Chen, and H. He, “A control of oxygen excess ratio for a PEMFC air supply system by intelligent PID methods,” Sustainability, vol. 15, p. ID 8500, 2023.
  24. [26]Y. Wang and Y. Wang, “Pressure and oxygen excess ratio control of PEMFC air management system based on neural network and prescribed performance,” Engineering Applications of Artificial Intel- ligence, vol. 121, p. ID 105850, 20243.
  25. [27]V. Sanchez, R. Barbosa, L. Arriaga, and J. Ramirez, “Real time control of air feed subsystem in a PEM fuel cell by means of an adaptive neural- network,” Int. J. of Hydrogen Energy, vol. 39, pp. 16 750–16 762, 2014.
  26. [28]Y. Wang, Y. Wang, J. Zhao, and J. Xu, “Observer-based adaptive neural network control for PEMFC air-feed subsystem,” Applied Soft Computing, vol. 113, p. ID 108003, 2021.
  27. [29]J. Li and T. Yu, “Intelligent controller based on distributed deep reinforcement learning for PEMFC air supply system,” IEEE Access, vol. 9, pp. 7496–7507, 2021.
  28. [30]M. Fliess and J. Join, “Model-free control,” Int. J. Control, vol. 86, pp. 2228–2252, 2013.
  29. [31]——, “An alternative to proportional-integral and proportional- integral-derivative regulators: Intelligent proportional-derivative reg- ulators,” Int. J. Robust & Nonlinear Contr., vol. 32, pp. 9512–9524, 2022.
  30. [32]M. Niane, M. Zerrougui, and R. Outbib, “H−/H∞ filter-based fault detection application to PEM fuel cell air-feed system,” Asian J. of Control, vol. 27, pp. 169–178, 2025.
  31. [33]K. Suh and A. Stefanopoulou, “Performance limitations of air flow control in power-autonomous fuel cell systems,” IEEE Trans. Control Syst. Techn., vol. 15, pp. 465–473, 2007.
  32. [34]S. Lagrouche, J. Liu, F. Ahmed, M. Harmouche, and M. Wack, “Adaptive second-order sliding mode observer-based fault reconstruction for PEM fuel cell air-feed system,” IEEE Trans. Control Syst. Techn., vol. 23, pp. 1098–1109, 2015.
  33. [35]M. Ait Ziane, M. Péra, C. Join, M. Benne, J. Chabriat, N. Yousfi Steiner, and C. Damour, “On-line implementation of model free controller for oxygen stoichiometry and pressure difference control of polymer electrolyte fuel cell,” Int. J. of Hydrogen Energy, vol. 47, pp. 38 311–38 326, 2022.
  34. [36]Y. Belkhier, S. Tamalouzt, E. Delaleau, G. Yao, A. Mamoune, and M. Benbouzid, “Hybrid offshore wind-hydrogen microgrid for marine dockingsstations: design and energy management for economic operation,” in Proc. Annual Conf. of the IEEE Industrial Electronics Society, Madrid, Spain, 2025.
  35. [37]M. Ait Ziane, M. Zadasdinski, C. Join, and M. Péra, “Model-free active sensor fault tolerant control for systems with linear static characteristic,” Int. J. of Dynamics and Control, vol. 13, p. ID 267, 2025.
  36. [38]M. Ait Ziane, C. Join, and M. Zadasdinski, “Sensor fault accommodation: a model-free framework,” Int. J. Control, vol. 98, pp. 2811–2822, 2025.

Paper details

vlnc:W_uv653sx8
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Engineering

Not recorded for this paper

View statement
Looking for related papers…
Contents