A neural network estimator of Solid Oxide Fuel Cell performance for on-field diagnostics and prognostics applications

Marra, Dario; Sorrentino, Marco; Pianese, Cesare; Iwanschitz, Boris

doi:10.1016/j.jpowsour.2013.04.114

Cited by 84 publications

(23 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Therefore, the prediction ability of the final discriminant method also depends on the selection and construction of the ANN. Marra et al 28 used multilayer‐perceptron‐feed‐forward neural networks to predict the impact of various input parameters on the output voltage, in which the output layer was a neuron representing the voltage. Through this study, the status of catalyst utilization can be obtained without knowing the specific value of utilization rate.…”

Section: Introductionmentioning

confidence: 99%

A novel predicting method on degree of catalytic reaction in fuel cells

Pan

et al. 2020

Int J Energy Res

View full text Add to dashboard Cite

Summary An accurate assessment of the utilization of a catalyst for fuel cells is beneficial for guiding the optimization of working conditions and adjustment of parameters. In this study, a composite model containing the numerical model of catalyst layers and improved support vector machine (SVM) is proposed to predict the catalyst utilization. The effectiveness factor of the field of catalytic reaction engineering is introduced into the numerical model. The adaptive learning factor and the differential evolution strategy are introduced to improve the ability of generalization and robustness of the distinguishing method in the SVM. The results indicate that the radial basis function is well‐suited as the kernel function of the SVM, and the accuracy of the prediction in the anode and cathode can be increased up to 99.31% and 99.65%, respectively. The increasing transfer coefficient leads to an enhancement of the catalytic reaction, and the particle size of the catalyst impacts the catalytic reaction mainly by changing the pattern of internal diffusion. Pressure has little impact on the Knudsen diffusion.

show abstract

Section: Introductionmentioning

confidence: 99%

A novel predicting method on degree of catalytic reaction in fuel cells

Pan

et al. 2020

Int J Energy Res

View full text Add to dashboard Cite

show abstract

“…31 The result suggests that the TPB length goes through a familiar curve that features a maximum at a low infiltration loading (about 4-7%), agreeing with our studies. 26,28 The complexity of investigation of parametric study leads researchers to couple theoretical models with artificial intelligence approach, such as artificial neural network 32,33 and genetic algorithm. 34 For example, after generating 3D microstructures of infiltrated electrodes, Tafazoli et al build a search engine with artificial intelligence approach to find the optimal geometric properties.…”

Section: Introductionmentioning

confidence: 99%

A numerical study of infiltrated solid oxide fuel cell electrode with dual‐phase backbone

2018

View full text Add to dashboard Cite

Summary Three‐dimensional microstructure of infiltrated solid oxide fuel cell electrodes with dual‐phase backbone is simulated numerically. This work employs LSM (lanthanum strontium manganite) infiltrated LSM yttria‐stabilized zirconia composite electrodes as an example. Important geometric properties, including percolation probability of LSM, total and percolated three‐phase boundary (TPB) lengths, and total and percolated surface areas of LSM, are calculated under various LSM nanoparticle loadings. One important finding is that compared with pure yttria‐stabilized zirconia backbone, dual‐phase backbone results in lower TPB length and has little impact on the surface area of LSM particles. Therefore, the addition of LSM backbone may be ineffective, even negative in promoting the electrode performance, however, can significantly reduce the LSM infiltration threshold loading to form a percolating network. The trade‐off between decreasing infiltration cycle and increasing TPB length demands an optimized content of LSM in backbone.

show abstract

“…These networks like the human brain consist of a number of neurons with different transfer functions to correlate the multi input/output parameters in engineering problems . Marra et al have used a neural network to estimate the solid oxide fuel cell performance; also, Bozorgmenhri et al have used ANN and a genetic algorithm (GA) to model the effect of constructive parameters (anode porosity, electrolyte thickness, and electrode support/functional layer thickness) of a single SOFC power density. Among the different types of ANNs, some researcher like Saengrung et al investigated the performance of a commercial proton exchange membrane fuel cell system with the back‐propagation neural network as a useful tool to quite satisfactory control of the output parameters based on the variations of input data.…”

Section: Introductionmentioning

confidence: 99%

Investigation of the geometric property hull for infiltrated solid oxide fuel cell electrodes

et al. 2017

View full text Add to dashboard Cite

Summary Design of optimal microstructures for infiltrated solid oxide fuel cell (SOFC) electrodes is a complicated task because of the multitude of electro‐chemo‐physical phenomena taking place simultaneously that directly affect working conditions (cell temperature, current density, and flow rates) of the SOFC electrode and therefore its performance. In this study, an innovative design paradigm is presented to obtain a part of geometry‐related electrochemical and physical properties of an infiltrated SOFC electrode. A range of digitally realized microstructures with different backbone porosity and electrocatalyst particle loading under various deposition conditions are generated. Triple phase boundary (TPB), active surface density of particles, and gas transport factor are evaluated in realized models on the basis of selected infiltration strategy. On the basis of this database, a neural network is trained to relate desired range of input geometric parameters to a property hull. The effect of backbone porosity, loading, distribution, and aggregation behavior of particles is systematically investigated on the performance indicators. It is shown that from the microstructures with very high amount of TPB and particle contact surface density, a relatively low gas diffusion factor should be expected; meanwhile, increasing those parameters does not have sensible contradiction with each other. Excessive agglomerating of particles has a negative effect on TPB density, but the distribution of seeds always has a positive effect. Direct search and genetic algorithm optimization techniques are used finally to achieve optimal microstructures on the basis of assumed target functions for effective geometric properties.

show abstract

A neural network estimator of Solid Oxide Fuel Cell performance for on-field diagnostics and prognostics applications

Cited by 84 publications

References 20 publications

A novel predicting method on degree of catalytic reaction in fuel cells

A novel predicting method on degree of catalytic reaction in fuel cells

A numerical study of infiltrated solid oxide fuel cell electrode with dual‐phase backbone

Investigation of the geometric property hull for infiltrated solid oxide fuel cell electrodes

Contact Info

Product

Resources

About