Neural network model for a commercial PEM fuel cell system

Performance prediction of a commercial proton exchange membrane (PEM) fuel cell system by using artificial neural networks (ANNs) is investigated. Two artificial neural networks including the back-propagation (BP) and radial basis function (RBF) networks are constructed, tested and compared. Experimental data as well as preprocess data are utilized to determine the accuracy and speed of several prediction algorithms. The performance of the BP network is investigated by varying error goals, number of neurons, number of layers and training algorithms. The prediction performance of RBF network is also presented. The simulation results have shown that both the BP and RBF networks can successfully predict the stack voltage and current of a commercial PEM fuel cell system. Speed and accuracy of the prediction algorithms are quite satisfactory for the real-time control of this particular application.     

Real-time implementation of a neural model-based self-tuning PID strategy for oxygen stoichiometry control in PEM fuel cell

This paper proposes a real-time implementable self-tuning PID control strategy to tackle oxygen excess ratio regulation challenge of a proton exchange membrane fuel cell. Controller parameters are updated on-line, at each sampling time, using a not iterative procedure based on an artificial neural network model. The proposed controller takes account of nonlinear behaviors of the process, while avoiding heavy computations.     

Development of composite-metal hybrid bipolar plates for PEM fuel cells

In this study, in order to increase the electrical conductivity, a carbon composite-metal hybrid bipolar plate has been developed using pre-forming method followed by a plasma surface treatment. A pre-formed metal foil between the carbon fiber/polymer composite plates promotes the metal foil to follow the shape of the channels of the bipolar plates without tearing and permits a continuous flow of electrons. The pre-formed metal foil also reduces the residual stress between the composite and metal foils, which helps prevent delamination between the composite and metal foils. The composite surface has been treated with plasma to increase the contact area between the carbon fiber and the gas diffusion layer (GDL). The composite-metal hybrid bipolar plates have only 1.4% of the total electrical resistance of that of the conventional composite bipolar plates. Unit cell test results have proved that the developed composite-metal hybrid bipolar plates with reduced total electrical resistance increase the cell performance.     

Study of cell voltage uniformity of proton exchange membrane fuel cell stack with an optimized artificial neural network model

The cell voltage uniformity of the proton exchange membrane fuel cell stack, which may consist of tens or hundreds of cells in series, plays a significant role in the stack's lifetime and performance. But it is challenging to predict the multi-cell voltages and the uniformity with a physics-based model due to complex stack geometry and huge computation efforts. In this work, we develop an artificial neural network model to estimate the steady-state cell voltage distributions of a 60 kW 140-cell stack. The optimized and well-trained model can efficiently reproduce the 140-cell voltages at different operating conditions with the error of less than 2 mV. The increased cathode gas pressure improves the cell voltage consistency and stack performance, while the voltage uniformity worsens with ascending load current. The efficient model prediction of cell voltages is beneficial for accurate evaluation of fuel cell performance, health state, and reliability.     

Experimental analysis of dynamic characteristics on the PEM fuel cell stack by using Taguchi approach with neural networks

This study determines the optimum operating parameters for a proton exchange membrane fuel cell (PEMFC) stack to obtain small variation and maximum electric power output using a robust parameter design (RPD). The operating parameters examined experimentally are operating temperatures, operating pressures, anode/cathode humidification temperatures, and reactant flow rates. First, the dynamic Taguchi method is used to obtain the maximum and stable power density against the different current densities, which are regarded as the systemic inputs considered a signal factor. The relationship between control factors and responses in the PEMFC stack is determined using a neural network. The discrete parameter levels in the dynamic Taguchi method can be divided into desired levels to acquire real optimum operating parameters. Based on these investigations, the PEMFC stack is operated at the current densities of 0.4–0.8 A/cm². Since the voltage shift is quite small (roughly 0.73–0.83 V for each single cell), the efficiency would be higher. In the range of operation, the operating pressure, the cathode humidification temperature and the interactions between operating temperature and operating pressure significantly impact PEMFC stack performance. As the operating pressure increasing, the increments of the electric power decrease, and power stability is enhanced because the variation in responses is reduced.     

Prediction and analysis of the cathode catalyst layer performance of proton exchange membrane fuel cells using artificial neural network and statistical methods

A mathematical model was developed to investigate the cathode catalyst layer (CL) performance of a proton exchange membrane fuel cell (PEMFC). A numerous parameters influencing the cathode CL performance are implemented into the CL agglomerate model, namely, saturation and eight structural parameters, i.e., ionomer film thickness covering the agglomerate, agglomerate radius, platinum and carbon loading, membrane content, gas diffusion layer penetration content and CL thickness. For the first time, an artificial neural network (ANN) approach along with statistical methods were employed for modeling, prediction, and analysis of the CL performance, which is denoted by activation overpotential. The ANN was constructed to build the relationship between the named parameters and activation overpotential. Statistical analysis, namely, analysis of means (ANOM) and analysis of variance (ANOVA) were done on the data obtained by the trained neural network and resulted in the sensitivity factors of structural parameters and their mutual combinations as well as the best performance.     

Dynamic neural network controller model of PEM fuel cell system

This paper presents the artificial intelligence techniques to control a proton exchange membrane fuel cell system process, using particularly a methodology of dynamic neural network. In this work a dynamic neural network control model is obtained by introducing a delay line in the input of the neural network. A static production system including a PEMFC is subjected to variations of active and reactive power. Therefore the goal is to make the system follow these imposed variations. The simulation requires the modelling of the principal element (PEMFC) in dynamic mode. The simulation results demonstrate that the model-based dynamic neural network control scheme is appropriate for controlling, the stability of the identification and the tracking error were analyzed, and some reasons for the usefulness of this methodology are given.     

Automatic features extraction of faults in PEM fuel cells by a siamese artificial neural network

In this paper, a procedure aimed at the automatic extraction of the features from polymer electrolyte membrane fuel cell impedance spectra is proposed. An artificial neural network that is trained by exploiting the similarity learning concept has been used. The network learns the features of the impedance spectra and maps each of them into the embedding space by clustering them accurately and by emphasising differences among spectra corresponding to different faults. The siamese network structure is optimised and the quality of the learnt representation is evaluated by analysing the clusters obtained in the features space. The dataset of experimental spectra has been augmented in two different ways and the results are compared. The clustering quality of the proposed siamese network is compared with the one of other state of the art approaches.     

Development of optimisation model for direct methanol fuel cells via cell integrated network

A cell network consists of a combination of fuel cells to achieve the targeted power consumption for a specific application. The main objective of this study is to design and optimise direct methanol fuel cell (DMFC) via cell integrated network model targeted for small portable application, such as cell phones and tablets. The target current and voltage was 1400 mA and 3.7 V, respectively, for a 5.18 W of cell network power. The optimisation was performed using 16 cells that were arranged in series with a voltage output of 3.781 V and a current of 1400 mA. The overall active area for the cell network was 128 cm², and the cost of 1 set of cell networks is USD 1400.     

Data driven models for a PEM fuel cell stack performance prediction

The operating principles of polymer electrolyte membrane (PEM) fuel cells system involve electrochemistry, thermodynamics and hydrodynamics theory for which it is not always easy to establish a mathematical model. In this paper two different methods to model a commercial PEM fuel cell stack are discussed and compared. The models presented are nonlinear, derived from a black-box approach based on a set of measurable exogenous inputs and are able to predict the output voltage and cathode temperature of a 5 kW module working at the CNR-ITAE. A PEM fuel cell stack fed with H₂ rich gas is employed to experimentally investigate the dynamic behaviour and to reveal the most influential factors. The performance obtained using a classical Neural Networks (NNs) model are compared with a number of stacking strategies. The results show that both strategies are capable of simulating the effects of different stoichiometric ratio in the output variables under different working conditions.     

Performance prediction of PEM fuel cell with wavy serpentine flow channel by using artificial neural network

Effects of serpentine flow channel having sinusoidal wave at the rib surface on performance of PEMFC having 25 cm² active area are investigated at different flow rates, three different amplitudes changing from 0.25 mm to 0.75 mm and three different cell operation temperatures. A proton exchange membrane fuel cell (PEMFC) is modeled for the prediction of the output current by using artificial neural network (ANN) that is utilized the aforementioned experimental parameters. Effect of hydrogen and air flow rate, the fuel cell temperature, amplitude of channel is tested. The results indicated that model C1 having lowest amplitude is enhanced maximum power output up to 20.15% as compared to indicated conventional serpentine channel (model C4) for 0.7 SLPM H₂ and 1.5 SLPM air and also model C1 has better performance than C2, C3 and C4 models. The maximum power output is augmented with increasing the cell temperature due to raising the fuel and oxidant diffusion ratio. Cell temperature, amplitude, H₂ and air flow rate and input voltage is used as input variables in train and test of the developing ANN model. MAPE of training and testing is determined as 2.89 and 2.059, respectively. Prediction results of developed ANN model including two hidden layer shows similar trend with experimental results. Developed ANN model can be used to both decrease the number of required experiments and find the optimum operation condition within the range of input parameters.     

A hierarchical model for automotive PEM fuel cell system comprehensive evaluation and comparison

This paper proposes a hierarchical model for automotive fuel cell system evaluation, comparison and applicability assessment through Analytic Hierarchy Process. A dual-layer criterion based hierarchical model is proposed, which features 9 sub-performances and 18 indices. Criterion weights and scoring function are both established. Experimental application is carried out to illustrate the evaluation process and how the proposed model operates. Results show that the sub-performance derived from automotive powertrain requirements represented the overall characteristics of the PEMFCs. The selected 18 most significant indices also made each sub-performance quantitatively assessable.     

A fast two-phase non-isothermal reduced-order model for accelerating PEM fuel cell design development

A reduced-order model (ROM) is developed for proton exchange membrane fuel cells (PEMFCs) considering the non-isothermal two-phase effects, with the goal of enhancing computational efficiency and thus accelerating fuel cell design development. Using analytical order reduction and approximation methods, the fluxes and source terms in conventional 1D conservation equations are reduced to six computing nodes at the interfaces between each cell component. The errors associated with order reduction are minimized by introducing new approximation methods for the potential distribution, the transport properties, and the membrane hydration status. The trade-off between model accuracy and computational efficiency is studied by comparing the simulation results and computational times of the new model with a full 1D model. The new model is nearly two orders of magnitude faster without sacrificing too much accuracy (<4% difference) compared to the 1D model. The new model is then used to analyze the influence of the membrane electrode assembly (MEA) design on cell performance and internal state distributions, offering insights into MEA structural optimization. The model can be readily extended to account for more detailed physico-chemical processes, such as Knudsen diffusion or the influence of micro-porous layers, and it can be an effective tool for understanding and designing PEMFCs.     

Performance prediction of PEM fuel cell cathode catalyst layer using agglomerate model

Computation of Proton Exchange Membrane (PEM) fuel cell's cathode Catalyst Layer (CL) is performed using agglomerate models in this paper, and the results are compared with homogenous one. Following our earlier homogenous model for cathode CL (see Khajeh-Hosseini et al., 2010), the focus of the present study is on agglomerate model. In this study, the derivation of agglomerate model is performed in such a way that in the simplified case when agglomerate sizes shrink to zero, the homogeneous model condition is retrieved. Validations versus two sets of experimental data are performed. For example, in one of the validation cases, Case (II), it is observed that in I_tot = 3000 [A m⁻²] the homogeneous model overestimates the performance by 80%. But the agglomerate model agrees well with the validating test cases. A set of parametric studies are performed using the agglomerate model, in which the influences of some CL structural- and cell operating-parameters are studied. A sensitivity study on the cell performance is performed to rank the influence of the parameters, with rank 1 for the most influential parameter. It is observed the agglomerate sizes possess rank 1. These results give useful guidelines for manufactures of PEMFC catalyst layers.     

Development of a fast empirical design model for PEM stacks

A simple and fast empirical design model for a 5 kW proton exchange membrane (PEM) stack is presented in this paper. The performance analysis of the PEM stack operating on a membrane humidifying method is made through a series of experiments, including current–voltage–power characteristics, uniformity of cell unit voltages, gas pressure impact and air flux impact. Based on the above analysis, an empirical predicted model for the PEM stack has been developed by the combination of mechanistic and empirical modeling approaches to characterize and predict the voltage–current characteristics without examining in depth all physical/chemical phenomena. The good agreement between the predicted and experimental results covering a range of optimal operating conditions shows that the proposed model provides an accurate representation of the behavior for the PEM stack.     

Characterization of electrode structures and the related performance of direct methanol fuel cells

The performance of membrane electrode assemblies (MEAs) in fuel cells is substantially affected by the structures of the electrodes. An increase of about 25% in power performance was achieved merely by controlling the pressure of hot press in the MEA fabrication process for a given Pt loading, instead of by employing pore formers and heat treatment-a widely accepted method-to modify the structures of the electrode. The microstructures of the different hot-pressed electrodes were examined by transmission electron microscopy, scanning electron microscopy, and small angle X-ray scattering to assess the effect of the pressure on the structures of the electrodes. Based on experimental observations, the improved performance of the MEA is attributed to the porosity of the cathode electrode, in which a network of macrofissures and sub-microfissures allows air to penetrate the electrode. Emphasis is also placed on the relationship between the total porosity of the electrodes and the MEA performance. Results of this study demonstrate that the specific power density nearly doubles when the total porosity increased from 57% to 76%. Also, the MEAs mounted in an air-breathing DMFC small pack were fabricated in-house to supply power for a mobile phone.     

A novel surrogate model for channel geometry optimization of PEM fuel cell based on Bagging-SVM Ensemble Regression

Channel structure plays an important role on the performance of proton exchange membrane fuel cell (PEMFC). In this study, the channel geometry of a PEMFC is optimized through genetic algorithm to obtain better performance. For the first time, a machine learning method called Bagging Ensemble Regression is employed as the surrogate model to calculate the fitness value of the algorithm, which accelerates the optimization process. First, a three-dimensional PEMFC simulation model is developed as the optimization prototype through CFD technology. Second, the Bagging ensemble model is trained through training data obtained from the CFD model. Then the Bagging ensemble model is integrated into the genetic algorithm to conduct the optimization process. Finally, the optimal model obtained is compared with the optimization prototype in terms of polarization curves, pressure drop, and reactant distribution, and the advantages of using Bagging ensemble model are discussed. Results show that the optimal model has a smaller pressure drop and a more uniform reactant distribution than the basic model at the expense of just a little power density. The presented surrogate model shows high prediction accuracy with only a small amount of training data, which is superior to the commonly used surrogate models.     

Experimental investigation on the dynamic performance of a hybrid PEM fuel cell/battery system for lightweight electric vehicle application

A hybrid system combining a 2 kW air-blowing proton exchange membrane fuel cell (PEMFC) stack and a lead–acid battery pack is developed for a lightweight cruising vehicle. The dynamic performances of this PEMFC system with and without the assistance of the batteries are systematically investigated in a series of laboratory and road tests. The stack current and voltage have timely dynamic responses to the load variations. Particularly, the current overshoot and voltage undershoot both happen during the step-up load tests. These phenomena are closely related to the charge double-layer effect and the mass transfer mechanisms such as the water and gas transport and distribution in the fuel cell. When the external load is beyond the range of the fuel cell system, the battery immediately participates in power output with a higher transient discharging current especially in the accelerating and climbing processes. The DC–DC converter exhibits a satisfying performance in adaptive modulation. It helps rectify the voltage output in a rigid manner and prevent the fuel cell system from being overloaded. The dynamic responses of other operating parameters such as the anodic operating pressure and the inlet and outlet temperatures are also investigated. The results show that such a hybrid system is able to dynamically satisfy the vehicular power demand.