Development of solid oxide fuel cell stack models for monitoring, diagnosis and control applications

Abstract

In the present thesis different SOFC stack models have been presented. The results shown were obtained in the general framework of the GENIUS project (GEneric diagNosis Instrument for SOFC systems), funded by the European Union (grant agreement n° 245128). The objective of the project is to develop “generic” diagnostic tools and methodologies for SOFC systems. The “generic” term refers to the flexibility of diagnosis tools to be adapted to different SOFC systems. In order to achieve the target of the project and to develop stack models suitable for monitoring, control and diagnosis applications for SOFC systems, different modeling approaches have been proposed. Particular attention was given to their implementability into computational tools for on-board use. In this thesis one-dimensional (1-D), grey-box and blackbox stack models, both stationary and dynamic were developed. The models were validated with experimental data provided by European partners in the frame of the GENIUS project. A 1-D stationary model of a planar SOFC in co-flow and counter-flow configurations was presented. The model was developed starting from a 1- D model proposed by the University of Salerno for co-flow configuration (Sorrentino, 2006). The model was cross-validated with similar models developed by the University of Genoa and by the institute VTT. The crossvalidation results underlined the suitability of the 1-D model developed. A possible application of the 1-D model for the estimation of stack degradation was presented. The results confirmed the possibility to implement such a model for fault detection. A lumped gray-box model for the simulation of TOPSOE stack thermal dynamics was developed for the SOFC stack of TOPSOE, whose experimental data were made available in the frame of the GENIUS project. Particular attention was given to the problem of heat flows between stack and surrounding and a dedicated model was proposed. The black-box approach followed for the implementation of the heat flows and its reliability and accuracy was shown to be satisfactory for the purpose of its applications. The procedure adopted turned out to be fast and applicable to other SOFC stacks with different geometries and materials. The good results obtained and the limited calculation time make this model suitable for implementation in diagnostic tools. Another field of application is that of virtual sensors for stack temperature control. Black-box models for SOFC stack were also developed. In particular, a stationary Neural Network for the simulation of the HEXIS stack voltage was developed. The analyzed system was a 5-cells stack operated up to 10 thousand hours at constant load. The neural network exhibited very good prediction accuracy, even for systems with different technology from the one used for training the model. Beyond showing excellent prediction capabilities, the NN ensured high accuracy in well reproducing evolution of degradation in SOFC stacks, especially thanks to the inclusion of time among model inputs. Moreover, a Recurrent Neural Network for dynamic simulation of TOPSOE stack voltage and a similar one for a short stack built by HTc and tested by VTT were developed. The stacks analyzed were: a planar co-flow SOFC stack (TOPSOE) and a planar counter-flow SOFC stack (VTT-HTc). All models developed in this thesis have shown high accuracy and computation times that allow them to be implemented into diagnostic and control tool both for off-line (1-D model and grey-box) and for on-line (NN and RNNs) applications. It is important noting that the models were developed with reference to stacks produced by different companies. This allowed the evaluation of different SOFC technologies, thus obtaining useful information in the models development. The information underlined the critical aspects of these systems with regard to the measurements and control of some system variables, giving indications for the stack models development. The proposed modeling approaches are good candidates to address emerging needs in fuel cell development and on-field deployment, such as the opportunity of developing versatile model-based tools capable to be generic enough for real-time control and diagnosis of different fuel cell systems typologies, technologies and power scales. [edited by author]