The Functional Mock-up Interface (FMI) standard is a flagship in the co-simulation and model exchange domain. However, the integration of graph-based computational models—particularly neural networks—into Functional Mock-up Units (FMUs) has remained a technical challenge due to interoperability and platform-specific limitations. To address this, we propose ONNX2FMU, a command-line Python tool that facilitates the deployment of Open Neural Network Exchange (ONNX) models into FMUs. According to FMI's good practices, ONNX2FMU generates C source code to wrap ONNX models in Functional Mockup Units, supports FMI versions 2.0 and 3.0, and provides multi-platform compilation capabilities. The tool simplifies the mapping process between model description and ONNX model inputs and outputs via JSON files, ensuring accessibility and flexibility. This paper presents the tool architecture and methodology and showcases its applicability through illustrative examples, including a reduced-order model powered by a recurrent neural network.

Ammonia is a promising zero carbon and sustainable hydrogen carrier that can be used as a fuel in solid oxide fuel cells (SOFC) by offering advantages related to the ease of storage and the possibility of being used directly without an external reformer. In this study, a Modelica-based dynamic model of an 'Ammonia to Power' (A2P) system was developed by integrating ammonia decomposition kinetics, electrochemical reactions, all the system-level components and the main control loops. A novel Balance of Plant (BoP) configuration is proposed, featuring a five-way heat exchanger that recovers waste heat primarily using the fuel stream as the thermal energy vector instead of air. The model evaluates transient responses to operational perturbations, the behavior of the different control loops, and recirculation percentage rates to optimize system performance. Efficiency is calculated as the ratio of the power output from the SOFC to the power derived from the fresh ammonia line.

Interest in ammonia as an energy carrier is growing due to its superior storage and transport properties compared to hydrogen. The objective of this work is to construct a useful tool for predicting the behavior of a solid oxide fuel cell (SOFC) stack fed directly with ammonia. This configuration is particularly interesting because the internal cracking of ammonia eliminates the need for an external cracker, thus reducing the overall cost of the system. The ammonia decomposition reaction was implemented in the anode channel of the stack and calibrated against literature results. The model was then validated in the ohmic region only by calculating the area specific resistance (ASR) and comparing the results with experimental data collected at the Bruno Kessler Foundation (FBK) laboratory. This SOFC model can therefore be used as a starting point for the analysis of a scale-up application.