Volume 65

Transient Feature and Potential Faults Prediction for SOFC Powered All-electric Ship Propulsion System Using BP Neural Network Shengying Xiao, Jiale Wen, Yiwu Weng, Xiaojing Lv

Abstract

To reveal potential faults and dynamic performance during operation of SOFC powered all-electric ship propulsion system, this research innovatively proposes a prediction approach based on Back Propagation Neural Network (BP-NN), taking into account fuel cell overheating and thermal cracking phenomenon. A highly accurate mechanistic mathematical model is established and validated for precisely identifying transient characteristic evolution and evaluating thermal safety risk under variable operating conditions during actual ship operations.
The results show that the proposed system delivers an output power of 315.4 kW and a system efficiency of 59.8% under rated operating conditions. The model’s error is less than 5%, demonstrating its high accuracy and precision. Furthermore, thermodynamic characteristic is investigated in terms of stack temperature, output power, and stack temperature gradient under varying airflow step. Using BP-NN model, the prediction accuracies for three parameters are 98.5%, 99.2%, and 92.8%, respectively, demonstrating the high precision of the model in capturing transient behavior of SOFC system.
The results demonstrate that as airflow step decreases, dynamic response characteristics for three parameters are significantly affected, with stack temperature increasing and output power decreasing due to reduced oxygen supply, while stack temperature gradient becomes more pronounced due to weakened cooling effects. Under reduced air flow conditions, stack temperature tends to exceed safe threshold, making the stack prone to overheating failure, while under more significant reductions in airflow, the increased stack temperature gradient elevates the risk of thermal cracking failure. This work provides a technical support for the development of long-life, green, and efficient power generation technologies in the future.

Keywords SOFC, all-electric ship propulsion, transient performance prediction, BP neural network, safety evaluation